RESEARCH PERSPECTIVES OF RESEARCH PERSPECTIVES OF BIO MEDICAL BIO MEDICAL
SIGNAL PROCESSING SIGNAL PROCESSING
Dr. V. KrishnaveniAssociate Professor, Dept. of ECE
PSG College of Technologye-mail : [email protected]
1
2
Signal Processing• Ways to manipulate
signal in its original medium or an abstract representation.
• Signal can be abstracted as functions of time or spatial coordinates.
• Types of processing:– Transformation– Filtering– Detection– Estimation– Recognition and
classification– Coding (compression)– Synthesis and
reproduction– Recording, archiving– Analyzing, modeling
sin 2 500y t t ,I B x y
3
Digital Signal ProcessingDigital Signal ProcessingQ1: Q1: WHAT is DSP ?WHAT is DSP ?
Q2: Q2: WHY we need DSP?WHY we need DSP?
Q3: Q3: HOW to do DSP?HOW to do DSP?
How to understand the concept of digital signal processing? What How to understand the concept of digital signal processing? What is it’s relationship with other courses such as signal and systems, is it’s relationship with other courses such as signal and systems, communication engineering etc…,communication engineering etc…,
Solve the problem of analog signals with digital method due Solve the problem of analog signals with digital method due
to it’s advantages and limitations of ASP. However DSP is not free to it’s advantages and limitations of ASP. However DSP is not free from limitations.from limitations.
General procedure of DSP. How to implement DSP algorithm ?General procedure of DSP. How to implement DSP algorithm ?
4
WHAT is DSP ?WHAT is DSP ?• Signals generated via
physical phenomenon are analog in that – Their amplitudes are
defined over the range of real/complex numbers
– Their domains are continuous in time or space.
• Digital signal processing concerns processing signals using digital computers.– A continuous time/space
signal must be sampled to yield countable signal samples.
– The real-(complex) valued samples must be quantized to fit into internal word length.
WHY we need DSP?WHY we need DSP?Analog system performance degrades due toAnalog system performance degrades due to
•Design ComplexityDesign Complexity
•Long term drift (ageing)
•Short term drift (temperature)
•Sensitivity to voltage instability
•Batch-to-Batch component variation
•High discrete component count
•Interconnection failures 5
WHY we need DSP? WHY we need DSP? (contd…)(contd…)Advantages of Digital SystemsAdvantages of Digital Systems
•Design FlexibilityDesign Flexibility
•No short or long term driftsNo short or long term drifts
•Relative Immunity to minor power supply variations Relative Immunity to minor power supply variations
•Functional Repeatability Functional Repeatability
•High accuracy: High accuracy: Floating pointFloating point-8,16,32,64 -8,16,32,64 bitsbits
•High reliabilityHigh reliability
•Easy to interconnectEasy to interconnect
•Deal with high dimensional signalsDeal with high dimensional signals
•Low costs: Low costs: reusable, reconfigurablereusable, reconfigurable
•Adaptive capabilityAdaptive capability
Limitations of Digital SystemsLimitations of Digital Systems•Finite word length effectsFinite word length effects
6
How to do DSP ?How to do DSP ?A/DA/D DSPDSP D/AD/Axxaa(t)(t) yyaa(t)(t)FilterFilter
x(n)x(n) y(n)y(n)FilterFilter
Realization of theoretical algorithm and system (filter) on Realization of theoretical algorithm and system (filter) on software and hardware using a General purpose computer, software and hardware using a General purpose computer, General purpose DSP chip, General purpose DSP chip, Specific-design DSP chipSpecific-design DSP chip ((TITI [leading manufacture, 70%], [leading manufacture, 70%], AD, AD, MotoralaMotorala, Lucent etc.,,), Lucent etc.,,) including system architecture, chip selective, development of the including system architecture, chip selective, development of the software and hardware, etcsoftware and hardware, etc..
7
When you speak, your voice is picked up by an analog sensor in the cell phone’s microphone
An analog-to-digital converter chip converts your voice, which is an analog signal, into digital signals, represented by 1s and 0s.
The DSP compresses the digital signals and removes background noise.
In the listener’s cell phone, a digital-to-analog converter chip changes the digital signals back to an analog voice signal.
Your voice exits the phone through the speaker.
How to do DSP ? A specific example
8
9
• Given x and h, find y analyze• Given h and y, find x control• Given x and y, find h design
Three Problems in DSP
Input: x[n]Input: x[n] Output: y[n]Output: y[n]
10
Curriculum in Signal Curriculum in Signal ProcessingProcessing
• MathematicsMathematics• Signals and SystemsSignals and Systems• Communications theory and systemsCommunications theory and systems• Control theory and systemsControl theory and systems• Signal processing theory and systemsSignal processing theory and systems• Applications and researchApplications and research
Signal Processing related Signal Processing related Courses Courses
• Digital Image Processing• Advanced Signal Processing• Adaptive Signal Processing• Multirate Signal Processing• Statistical Signal Processing• Wavelets & Sub band coding• Speech Signal Processing• Bio-Medical Signal Processing • Data Compression• Multimedia Compressionetc......
11
12
Mathematics for Signal ProcessingMathematics for Signal Processing• Algebra, calculus, differential equationsAlgebra, calculus, differential equations
• Linear algebra, matrices, vector spaces, Linear algebra, matrices, vector spaces,
functional analysisfunctional analysis
• Probability, statistics, random processesProbability, statistics, random processes
• Computational mathematics, numerical Computational mathematics, numerical
analysis, algorithmsanalysis, algorithms
DSPDSP
MILITARYMILITARYSecure CommunicationsSonar ProcessingImage ProcessingRadar ProcessingNavigationMissile Guidance
VOICE/SPEECHVOICE/SPEECHSpeech RecognitionSpeech Processing/VocodingSpeech EnhancementText-to-SpeechVoice Mail
INSTRUMENTATIONINSTRUMENTATIONSpectrum AnalyzersSeismic ProcessorsDigital OscilloscopesMass Spectrometers
BIO MEDICALBIO MEDICALPatient MonitoringUltrasound EquipmentDiagnostic ToolsFetal MonitorsLife Support SystemsImage EnhancementX-ray storage/enhancement
INDUSTRIAL/CONTROLINDUSTRIAL/CONTROLRoboticsNumeric ControlPower Line MonitorsMotor/Servo Control
CONSUMERCONSUMERdigital televisiondigital televisiondigital cameradigital camerainternet music, phones and internet music, phones and videovideodigital answer machines, fax digital answer machines, fax and modemsand modemsvoice mail systemvoice mail systeminteractive entertainment interactive entertainment systemssystems
AUDIOAUDIOAV EditingDigital MixersHome TheaterPro Audio
COMMUNICATIONSCOMMUNICATIONSEcho CancellationAdaptive EqualizationDigital PBXsLine RepeatersModemsGlobal PositioningSound/Modem/Fax CardsCellular PhonesSpeaker PhonesVideo ConferencingATMs
Applications of DSP…
13
14
Modern Engineering is DesignModern Engineering is Design• Science Science studiesstudies and and describesdescribes what nature what nature
created, what already existscreated, what already exists• Engineering Engineering createscreates and and buildsbuilds what what
society wants and needs, what does not society wants and needs, what does not already existalready exist
• Engineering uses mathematics in a Engineering uses mathematics in a different perspectivedifferent perspective from science from science
15
Research – Current Research – Current ScenarioScenario
• Before 4 decades, research was done by a Before 4 decades, research was done by a small number of specialists in small number of specialists in laboratories and colleges.laboratories and colleges.
• Now, research is done by everybody in all Now, research is done by everybody in all levels of college and work.levels of college and work.
• Same true for “Design”Same true for “Design”
16
Steps in ResearchSteps in ResearchStep 1:Step 1: Identify the Topic/ProblemIdentify the Topic/Problem
Step 2: Step 2: Do an Extensive Literature SurveyDo an Extensive Literature Survey
Step3: Step3: Collect Relevant DataCollect Relevant Data
Step 4: Step 4: Propose New / Modify algorithmsPropose New / Modify algorithms
Step 5: Step 5: Interact with subject experts globallyInteract with subject experts globally
Step 6:Step 6: Identify Conferences/Journals and report your findings Identify Conferences/Journals and report your findings
Step 7: Step 7: Consolidate your work in the form of a thesisConsolidate your work in the form of a thesis
17
CASE STUDIES IN CASE STUDIES IN BIOMEDICALBIOMEDICAL
SIGNAL PROCESSING SIGNAL PROCESSING
1. EEG Signal Analysis (1D)1. EEG Signal Analysis (1D) 2. MRI Image Analysis (2D) 2. MRI Image Analysis (2D)
18
BIOMEDICAL SIGNALS
Biomedical signals carry useful information for probing, exploring, and understanding the behavior of biological system (human body) under investigation.
Different types of biomedical signals include ECG, EEG, EMG, EOG, ERG, EGG, PSG etc.,
Such recorded information cannot be readily accessed, being masked by noise or buried by other vital signals simultaneously recorded. In these cases, the raw signal has to be processed to yield useful results.
19
BIOMEDICAL SIGNAL PROCESSING Biomedical signal processing deals with the innovative applications
of signal processing methods in biomedical signals though various creative integrations of the method and biomedical knowledge.
The objectives of Biomedical Signal Processing includes
enhancement of features (waveforms) of interest, the quantitative analysis of physiological systems (from cells to
organs to the whole human organism), to extract useful information from various biological signals and
gain a better comprehension of physiological processes or to improve diagnosis, therapy, and rehabilitation in diseased patients.
In general, almost all the signal processing algorithms have the potential to be applied to various biomedical problems.
20
EEG SIGNAL ANALYSISEEG SIGNAL ANALYSIS
21
CERTAIN INVESTIGATIONS ON CERTAIN INVESTIGATIONS ON THE METHODOLOGIES FOR THE METHODOLOGIES FOR
REMOVAL OF OCULAR REMOVAL OF OCULAR ARTIFACTS FROM ARTIFACTS FROM
ELECTROENCEPHALOGRAM ELECTROENCEPHALOGRAM
22
OUTLINE OF THE PRESENTATIONOUTLINE OF THE PRESENTATION
Introduction- EEG, Artifacts in EEG, Ocular Artifacts (OA)
Literature Survey Motivation for the research Objective of the research Proposed Methodologies Conclusion and Future scope References Publications
23
ELECTROENCEPHALOGRAM (EEG)
EEG is a record of the amplified electrical activity generated by neurons in the brain.
In 1875 Richard Caton recorded the electrical activity of the brains of rabbits and monkeys directly from the brain tissue.
The first human EEG was recorded in 1924 by Hans Berger, a German psychiatrist. Since the days of Berger and the verification of his recordings by Jasper and Carmichael (1935), EEG has taken its place as a standard laboratory investigation in clinical neurophysiology and neurology.
It is used in the diagnosis of a number of clinical conditions like epilepsy, sleep disorders, brain tumors and disorders of the nervous system.
EEG recording is also used extensively in psychophysiological research and in the testing of drugs (pharmacology) (Pryse-Phillips 1997).
24
ELECTROENCEPHALOGRAM (EEG) ….. Contd
EEG signals are measured from electrodes positioned on the scalp in a 10-20 arrangement, a placement scheme devised by the International Federation of societies of EEG (Jasper 1958).
The 10-20 system was developed to standardize the collection of EEG and facilitate the comparison of studies performed at different laboratories.
F-Frontal lobeT-Temporal lobeC-Central lobeP-Parietal lobeO-Occipital lobe
25
EEG RECORDER
26A 31 Channel EEG recordingA 31 Channel EEG recording
27
ARTIFACTS IN EEG
EEG is designed to record cerebral activity.
It also records electrical activities arising from sites other than the brain.
The recorded activity that is not of cerebral origin is termed as an artifact [Selim Benbadis et al 2002] and EEG is susceptible to various artifacts such as
Physiological artifacts (Origin : Heart, muscle contraction, body, eyes etc.,)
Extra-physiological artifacts (Origin : Equipment, environment etc.,)
28
ARTIFACTS IN EEG ….. Contd
Types of physiologic artifacts
Muscle artifacts Eye blink artifacts Eye movement artifacts ECG artifacts Pulse artifacts Respiration artifacts Skin artifacts
Types of extra physiologic artifacts
Electrode popping artifacts Alternating current artifacts Artifacts due to movements in the environment
29
OCULAR ARTIFACTS (OA)
Voltage changes generated by eye movements and blinks produces large electrical potential around the eyes known as Electrooculogram (EOG).
EOG is a non-cortical activity that spreads across the scalp and contaminates the EEG [Croft et al 2002a].
Ocular Artifacts (OA) is a collective term used to describe a number of contaminating voltage potentials caused by eye movements and blinks (Jervis et al 1988).
30
A 21 Channel Ocular Artifact (EYE BLINK) contaminated EEG recording
Ocular artifacts are more prominent in all the frontal channels (FP1, FP2, F3, F4, F7, F8 and FZ) due to the placement of the corresponding frontal electrodes close to the eyes (Overton et al 1969; Terence et al 2000).
OCULAR ARTIFACTS (OA)….. ContdOCULAR ARTIFACTS (OA)….. Contd
31
OCULAR ARTIFACTS (OA)….. ContdOCULAR ARTIFACTS (OA)….. Contd
a) Uncontaminated baseline EEG b) EEG contaminated with slow blink artifactc) EEG contaminated with fast blink artifact d) EEG contaminated with vertical eye movement e) EEG contaminated with horizontal eye movement f) EEG contaminated with round eye movement
Some Ocular Artifact rich EEG rhythms….Some Ocular Artifact rich EEG rhythms….
32
OCULAR ARTIFACTS (OA)….. Contd
Ocular Artifacts are often dominant over other electrophysiological contaminating signals (ECG, EMG etc.,) as well as external interference due to power sources (Vigon et al 2000; Tatjana Zikov et al 2002)
EEG recordings are significantly distorted by OA, causing problems for analysis and interpretation by clinicians (Croft et al, 2000) and a nuisance for researchers who investigate the electrophysiology of the brain (Pivik et al 1993; Picton et al 2000)
Hence, a control procedure for filtering the OA from EEG is essential for interpreting EEG properly.
33
Technique Limitations Experimental Control (Hillyard et al 1970; Weerts et al 1973; Verleger R, 1991).
Often unrealistic or inadequate method. Subject concentrating to fulfill the requirements of this method might itself influence his/her EEG.
Rejection(Anthony BJ, 1985).
Results in a considerable loss of important useful data.
Time Domain and Frequency Domain Regression (Verleger R et al 1982; Gratton et al 1983, Woestengurg et al 1983; Gasser et al 1992)
Depend on reference EOG channel. Regression Coefficient varies for different eye movement type and frequency. Neither of these techniques take into account the propagation of the brain signals into the recorded EOG. Thus a portion of relevant EEG is always cancelled out with ocular artifact.
LITERATURE SURVEY
34
Technique Limitations
Principal Component Analysis (Berg et al 1994; Lagerlund et al 1997) & Independent Component Analysis (Scott Makeig et al 1996; Tzyy-Ping Jung et al 1998; Vigario et al 2000; Carrie A Joyce et al 2004; Shoker et al 2005)
PCA cannot completely separate eye artifacts from brain signals when both are of comparable amplitude. Automated ICA algorithms are computationally complex and requires reference EOG data.The interdependencies of the estimated independent components are not tested for their independence and uniqueness.
Wavelet based OA removal algorithm(Tatjana Zikov et al 2002)
The threshold limit was empirical and calculated from the uncontaminated baseline EEG.
LITERATURE SURVEY – Contd…..
35
Technique Limitations
Principal Component Analysis (Berg et al 1994; Lagerlund et al 1997) & Independent Component Analysis (Scott Makeig et al 1996; Tzyy-Ping Jung et al 1998; Vigario et al 2000; Carrie A Joyce et al 2004; Shoker et al 2005)
PCA cannot completely separate eye artifacts from brain signals when both are of comparable amplitude. Automated ICA algorithms are computationally complex and requires reference EOG data.The interdependencies of the estimated independent components are not tested for their independence and uniqueness.
Wavelet based OA removal algorithm(Tatjana Zikov et al 2002)
The threshold limit was empirical and calculated from the uncontaminated baseline EEG.
LITERATURE SURVEY – Contd…..
36
MOTIVATION FOR THE RESEARCHMOTIVATION FOR THE RESEARCH
OA significantly distort EEG recordings, and consequently the presence of OA needs to be accounted for in any EEG study, and researchers must take into account the effect of OA on the EEG.
Hence, devising methods for successful removal of OA from EEG recordings have been still a major challenge, and this work is confined to the development of certain methodologies for removal of OA from EEG.
37
MOTIVATION FOR THE RESEARCH - Contd…..
ICA algorithms obtain components that are approximately independent. However, there is no guarantee that any particular ICA algorithm can capture the individual source signal in its components (Carrie A Joyce et al 2004).
Hence, the performance of the ICA algorithm employed for artifact removal need to be investigated for their actual independence to identify the best separating algorithm
available in the literature.
38
MOTIVATION FOR THE RESEARCH - Contd…..
An ICA based automatic method for removal of OA proposed by (Carrie A Joyce et al 2004) requires six measured EOG channels which are generally not available and pose problems if previously recorded EEG data are to be processed.
The method proposed in (Shoker et al 2005) is computationally complex because of the high dimensionality of the feature space.
The results of these studies (Carrie A Joyce et all 2004, Shoker et al 2005) does not imply that SOBI (Second Order Blind Identification) algorithm is the overall best approach for decomposing EEG sensor data into meaningful components, and has not been completely validated by the authors.
Therefore, efficient algorithms to classify EEG and ocular artifact components obtained from the best separating algorithm is essential.
39
MOTIVATION FOR THE RESEARCH - Contd…..
In wavelet based denoising method for removal of ocular artifacts from EEG, the threshold limit was empirical and calculated from the uncontaminated baseline EEG (Tatjana Zikov et al 2002).
Using wavelets for artifact removal in EEG, selection of an appropriate threshold limit and thresholding function is context sensitive.
Hence, the wavelet based methodologies for OA removal to achieve better performance is required.
40
OBJECTIVE OF THE RESEARCH
To propose efficient algorithms for removing ocular artifacts from EEG which satisfies the following criteria:
Minimization of the magnitude of the Ocular Artifacts.
Retainment of the underlying brain signal besides the exact preservation of the high frequency components of the original signal.
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
41
PROPOSED METHODOLOGIES
Two different approaches are proposed for removal of OA from EEG
1. Component based approach
i. A Quantitative comparison of various ICA algorithms ii. A hybrid ICA- Kalman Predictor algorithm iii. A hybrid ICA-Neural Network algorithm
2. Wavelet based approach
i. A method based on Successive Thresholding of Wavelet Coefficients. ii. A method based on Adaptive Thresholding of Wavelet Coefficients. iii. OA Zone Identification and Removal of OA using Wavelet Transform.
42
EEG DATA
EEG DATA and EEGLAB Toolbox is obtained from Swartz Center for Computational Neuroscience,Institute for Neural Computation, University of California San Diego
EEG DATA:http://sccn.ucsd.edu/~arno/fam2data/publicly_available_EEG_data.html
EEGLAB Toolbox:
http://sccn.ucsd.edu/eeglab/
43
Component based approach
1. Quantitative comparison of the ICA algorithms for identifying the best
separating algorithm
44
Blind source separation (BSS) – cocktail party problem : speech signals from different speakers recorded using many sensors. Problem is to separate the voices of individual speakers.
ICA – a tool for BSS. Some assumptions about the sources:
Sources are statistically independent. measured signals are linear mixtures of
source signals. propagation delays of the mixing medium
are negligible. No. of sensors No. of independent sources.
INDEPENDENT COMPONENT ANALYSIS (ICA)
45
Basic ICA Model
As(t) x(t) T
1 ns(t) [s (t), ..., s (t)]T
1 nx(t) [x (t), ..., x (t)]
A is the n x n mixing matrix
W is the n x n demixing matrix
- SOURCE SIGNALS
- MEASURED MIXED SIGNALS
- ESTIMATED SOURCE SIGNALST
1 nˆ ˆ ˆs(t) [s (t), ..., s (t)]
46
ICA Algorithms applied to EEG in this study for artifact removal
Infomax (Bell et al, 1995)
Extended Infomax (Lee et al, 1996)
Fast ICA (Hyvarinen et al 1997)
SOBI (Second Order Blind Identification) (Beloucharani et al, 1997)
TDSEP (Temporal Decorrelation source SEParation ) (Ziehe et al, 1998)
JADE (Joint Approximate Diagonalization of Eigen matrices) (Jean-François Cardoso, 1999)
MS-ICA (Molgedey Schuster) (Molgedey et al, 1994)
SHIBBS (Shifted Blocks Blind Separation) (Cardoso et al, 1996)
OGWE (Optimized Generalized Weighted Estimator) (Juan et al, 1999)
Kernel-ICA (Bach et al, 2002)
47
…………...S1 S2 Sm
Mixing Matrix A
ICA Algorithm
……………Ŝ1 Ŝ2 Ŝm
Identify the artifact (Visually / Automatically)Remove the artifact
x1 x2 xm
……………X1 X2 Xm
…….
METHODOLOGY FOR ARTIFACT REMOVAL USING ICAMETHODOLOGY FOR ARTIFACT REMOVAL USING ICA
48
FP1
FP2
F3
F4
F7
F8
FZ
Ocular Artifact contaminated EEG recording Independent Components using JADE
1
2
3
4
5
6
7
491 2 3 4 5 6 7 8 9 10
0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
ICA ALGORITHMS
AV
ER
AG
E O
F M
I ES
TIM
ATE
1. ICA-MS2. SOBI3. INFOMAX ICA4. TDSEP5. EXTENDED INFOMAX6. KERNAL ICA7. FAST ICA8. SHIBBS9. OGWE10. JADE
1 2 NI (X ,X ,....X )1 2 NI (S ,S ,....S ) for various ICA algorithms
MI OF RAW EEG ICA-MS SOBI INFOMAX TDSEP EXTENDED
INFOMAXKERNAL
ICAFAST ICA SHIBBS OGWE JADE
5.1758 1.4298 1.3120 1.2095 0.8520 0.7516 0.7264 0.7230 0.7160 0.7030 0.6488
AVERAGE MI OF THE RECORDED EEG AND THE INDEPENDENT COMPONENTS OBTAINED BY VARIOUS ICA ALGORITHMS
50
Pair wise MI estimates of the independent components
Square root of variances for the pair wise MI estimates
Dependency Matrix(JADE)
1 2 3 4 5 6 7
1
2
3
4
5
6
7
0.05
0.1
0.15
0.2
0.25
Inde
pend
ent c
ompo
nent
s
Independent components M
utua
l Inf
orm
atio
n
Variability Matrix (JADE)
1 2 3 4 5 6 7
1
2
3
4
5
6
70
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Inde
pend
ent c
ompo
nent
s
Squ
are
root
s of v
aria
nces
Independent components
51
Component based approach
2. OA Removal using a hybrid ICA- Kalman Predictor algorithm
52
KALMAN FILTER
Kalman filtering is a process of implementing a set of mathematical equations which implements a predictor-corrector type estimator that is optimal in the sense that it minimizes the estimated error covariance when some presumed conditions are met (Greg Welch and Gary Bishop 2004).
It provides an efficient computational (recursive) means to estimate the state of a process. The filter is very powerful in several aspects: it supports estimations of past, present, and even future states, and it can do so even when the precise nature of the modeled system is unknown.
53
The time update (predictor equations) projects the current state estimate ahead in time. The measurement update (corrector equations) adjusts the projected estimate by an actual measurement at that time.
Ongoing discrete Kalman filter cycle
TIME UPDATE
PREDICT future state
PREDICT error covariance
TIME UPDATE
PREDICT future state
PREDICT error covariance
MEASUREMENT UPDATE
COMPUTE Kalman Gain
CORRECT future state
CORRECT error covariance
MEASUREMENT UPDATE
COMPUTE Kalman Gain
CORRECT future state
CORRECT error covariance
54
Block diagram of the proposed automatic ocular artifact removal system using JADE algorithm and Kalman predictor
Independent components obtained using JADE are tracked using a Kalman Predictor which is embedded with Adaptive Autoregressive (AAR) model.
55
The order of the predictor and the update coefficient determines the performance of the predictor in estimating the signal, and these parameters are selected based on the minimum mean square error obtained.
MSE plot for various model orders against update coefficient
Selection of Model order and update coefficient
56
Update Coefficient
Mean square errorp=2 p=3 p=4 p=5 p=6 p=7 p=8 p=9 p=10
0.1 1.17640 0.95543 0.83825 0.73945 0.67546 0.61000 0.56366 0.48368 0.45059
0.15 1.15500 0.74568 0.57421 0.51488 0.46110 0.38947 0.34468 0.29514 0.26385
0.2 0.94886 0.59869 0.46313 0.39476 0.33109 0.27648 0.25280 0.19649 0.17996
0.25 0.94411 0.52050 0.39238 0.32297 0.27791 0.22840 0.20033 0.16279 0.14662
0.3 1.00210 0.47635 0.35844 0.29117 0.25167 0.19603 0.17776 0.14141 0.13000
0.35 0.99947 0.44845 0.35664 0.29098 0.27460 0.18408 0.16164 0.12521 0.11765
0.4 1.05410 0.44686 0.46145 0.35242 0.31948 0.17102 0.15398 0.11637 0.10732
0.45 1.13780 0.58078 0.4709 0.39835 0.37475 0.16824 0.14110 0.11505 0.10527
0.5 1.35950 0.74955 0.50036 0.37423 0.35124 0.16415 0.14080 0.11372 0.10379
0.55 3.54460 1.16370 0.46467 0.36505 0.35896 0.17190 0.14684 0.11384 0.10327
0.6 7.97470 1.89560 1.50400 1.24260 1.18940 0.20922 0.16536 0.13501 0.12000
Mean square error obtained for various model orders and update coefficients
57
METRICS FOR EVALUATING THE PERFORMANCE OF THE CLASSIFIER
The performance of the classifier was evaluated in terms of the sensitivity, specificity and average detection rate (Tarassenko, 1998)
Sensitivity is a measure of the ability of the classifier to detect EOG components.
Specificity is a measure of the ability of the classifier to specify EEG components.
Average detection rate is the average of sensitivity and specificity.
58
Time Domain plots
The contaminated EEG signal and the corrected EEG are compared by inspecting their visual appearance of the time domain plots for each channel data, to ensure the retainment of underlying brain signal and to visualize the minimization of the amplitude of the ocular artifact.
Similarity measure (ShokerL et al 2005)
Calculates the difference between a segment of ocular artifact free EEG data before and after correction in which the difference is measured by the similarity of the two waveforms, and is defined by
Both these metrics ensure that the observations are faithfully reconstructed in time domain, both in terms of subjective visual inspection and objective performance metrics.
METRICS FOR EVALUATING THE PERFORMANCE OF THE OA REMOVAL METHODOLOGIES
dB
N1=10log 1- EN i =1
f [n] - f [n]i i
59
Power Spectral Density plot Spectral power in the lower frequency bands (0-13 Hz) [Gasser et al 1985] should be reduced and the higher frequency bands should not be affected [Somsen et al 1998].
PSD plots helps us to check whether the power of the spectral components of ocular artifacts has been reduced and whether the high frequency components are exactly preserved.
Frequency Correlation plot (Andreas Jung 2003) Correlation between the contaminated EEG and the corrected EEG is computed.
The low frequency spectrum should be less correlated and the high frequency spectrum should be highly correlated.
METRICS FOR EVALUATING THE PERFORMANCE OF THE OA REMOVAL METHODOLOGIES….. Contd
x,y
w 25
w1C =w 2 w 2
w1 w1
* *. * x(l) y(l) + x(l)y(l)
* *x(l)x(l) * y(l)y(l)
60
RESULTS FOR JADE-KALMAN ALGORITHM
To evaluate the performance of the proposed JADE-KALMAN algorithm, 372 EOG components 2161 EEG components are analyzed.
Similarity Measure and Standard Deviation
Threshold Sensitivity Specificity Average detection rate
0.1 44.5 % 97.75 % 71.13 %
0.15 65.75 % 96.6 % 81.16 %
0.2 72 % 92.8 % 82.4 %
0.3 79.3 % 83.8% 81.5 %
Parameter Threshold 0.1 Threshold 0.15 Threshold 0.2 Threshold 0.3
0.1435 0.4575 0.5168 2.0811
0.4455 1.0661 1.0953 2.8100db
db
61
Results for JADE-KALMAN Algorithm (10 second epoch)
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
Time domain plot
Frequency correlationPower Spectral Density
62
Results for JADE-KALMAN Algorithm (1 second epoch)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
Time domain plot
Power Spectral Density Frequency correlation
63
Results for JADE-KALMAN Algorithm (Eye movement)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
Time domain plot
Power Spectral Density Frequency correlation
64
Inferences: JADE-KALMAN algorithm exhibits good degree of
specificity (92.8%) ensuring minimum loss of the underlying brain signal for a threshold value of 0.2.
Since the value of sensitivity is less (72%), the algorithm missed approximately 30% of the EOG components.
The magnitude of the high frequency contents are not preserved as evident from PSD and frequency correlation plots.
Further to enhance the performance of the classifier, both in terms of its sensitivity and specificity, a new hybrid ICA-NN algorithm is proposed.
65
Component based approach
3. OA Removal using a hybrid ICA-NN algorithm
66
Block diagram of the proposed automatic ocular artifact removal system using JADE algorithm
and Neural Network
Raw EEG
67
ICA-NN Algorithm
PNN (Polynomial Neural Network) trained by GMDH (Group Method of Data Handling) algorithm.
FNN (Feed Forward Neural network) trained by Back Propagation algorithm.
Auto Regressive (AR) coefficients are used as input features to neural network.
68
JADE-PNN Algorithm PNN is a multilayered network based on evolutionary principle.
GMDH (Group Method of Data Handling) Training Algorithm is used to find the following parameters of the network * Weights for each neuron.
* Total number of neurons in each layer * Total number of layers
GMDH requires two non-intersecting subsets as training dataset and examining dataset along with testing dataset.
Training dataset is used for finding weights.
Examining dataset is used for finding the best neurons.
Testing dataset is used for testing the classifier’s performance in detecting the EOG and EEG components.
69
Structure of the trained Polynomial Neural Network to classify the independent components obtained from JADE algorithm
70
z1(1) = 0.75827*1 + 0.50852 * u97 + 0.85907 * u98 - 0.10723 * u97* u98;z2(1) = 0.70742*1 +1.3931 * u2 +0.4726* u85 + 0.38296 * u2* u85;z3(1) = 0.75827*1 + 0.50852 * u97 + 0.85907 * u98 - 0.10723 * u97* u98;
z4(1) = 0.57238 *1 + 1.377 * u2 + 0.26579 * u61 + 0.4594 * u2* u61;z5(1) = 0.70742*1 +1.3931 * u2 +0.4726* u85 + 0.38296 * u2* u85;z6(1) = 0.65842*1 + 0.35681* u67 +0.86528 * u92-0.033001 * u67* u92;z7(1) = 0.73729 *1+ 0.48345 * u97 +0.96379 * u104- 0.1949* u97* u104;z8(1) = 0.74838*1 + 0.52138 * u25 +1.0585* u56- 0.11204* u25* u56;z1(2) = -0.17655 *1 + 0.58637 * z1(1) + 0.30952 * z2(1) + 0.70452 * z1(1) *
z2(1);z2(2) = -0.21672 *1 + 0.64903 * z3(1) + 0.29864 * z4(1) + 0.74661 * z3(1) *
z4(1);z3(2) = -0.20438 *1 + 0.62977 * z5(1) + 0.31258 * z6(1) + 0.71985 * z5(1)*
z6(1);z4(2) = -0.14459 *1 + 0.55598 * z7(1) + 0.31728 * z8(1) + 0.63401 * z7(1)*
z8(1);z1(3) = -0.019766 *1 +0.48808 * z1(2) + 0.43042 * z2(2) + 0.1829(3) * z21*
z2(2);z2(3) = -0.052048 *1 +0.47732 * z3(2) + 0.50716 * z4(2) + 0.17377 * z3(2)*
z4(2);z1(4) = -0.16397 *1 + 0.83886 * z1(3) + 0.75138 * z2(3) - 0.42587 * z1(3)*
z2(3);
Polynomial EquationPolynomial Equation
71
RESULTS FOR JADE-PNN ALGORITHM
Training samples:164 EOG and 215 EEG samples Testing samples: 208 EOG and 1946 EEG samples
0.56474
1.0578
Similarity Measure and Standard Deviation
db
db
Data set Sensitivity Specificity Average detection rate
Training data 76 % 79 % 77.5 %
Testing data 55 % 78 % 66.5%
72
Results for JADE-PNN Algorithm (10 second epoch)
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-20
0
20
40
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
73
Results for JADE-PNN Algorithm (1 second epoch)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
74
Results for JADE-PNN Algorithm (Eye movement)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
75
JADE-FNN Algorithm
Structure of the trained Feed Forward Neural Network to classify the independent components obtained from JADE algorithm
76
0.28977
0.4662
Similarity Measure and Standard Deviation
db
db
Data set Sensitivity Specificity Average detection rate
Training data 96 % 98 % 97 %
Testing data 89 % 94 % 91.5 %
RESULTS FOR JADE-FNN ALGORITHM
Training samples:164 EOG and 215 EEG samples Testing samples: 208 EOG and 1946 EEG samples
77
Results for JADE-FNN Algorithm (10 second epoch)
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
78
Results for JADE-FNN Algorithm (1 second epoch)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
79
Results for JADE-FNN Algorithm (Eye movement)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
80
COMPARISON OF JADE-KALMAN, JADE-PNN AND JADE-FNN ALGORITHMS
Algorithm Data set Sensitivity Specificity Average Detection Rate
JADE-KALMAN
(0.2 Threshold)
Training data 72.5 % 93.4 % 83 %
Testing data 71.51 % 92.2 % 81.8 %
JADE-PNN Training data 76 % 79 % 77.5 %
Testing data 55 % 78 % 66.5 %
JADE-FNN Training data 96 % 98 % 97 %
Testing data 89 % 94 % 91.5 %
Similarity Measure and Standard DeviationParameter JADE-KALMAN JADE-PNN JADE-FNN
0.5168 0.56474 0.28977
1.0953 1.0578 0.4662
dbdb
Training samples:164 EOG and 215 EEG samples Testing samples: 208 EOG and 1946 EEG samples
81
Inferences:
Minimization of ocular artifacts and retainment of the underlying brain signal is appreciable using JADE-FNN compared to JADE-PNN and JADE-KALMAN algorithms.
The results for ICA-Kalman Predictor and ICA-NN algorithms purely depend on the accurate detection of EOG component from the independent components obtained using the JADE algorithm.
The success of ICA-NN classifier solely depend on the training samples.
It is worth noting that the component based approaches do not exactly preserve the high frequency contents of the original signal, since the identification and removal of artifacts is being carried out by observing the contaminated signal in time domain only.
Hence, an investigative study of the contaminated EEG in time domain as well as in frequency domain might throw more light into the process of removing OA from EEG.
82
Wavelet based approach
4. Removal of OA using Successive Thresholding of wavelet
coefficients
83
Why Wavelets for EEG ?
EEG signal is non-stationary in both time and space.
Specific components in EEG may be localized in time, space and scale. Wavelet analysis provides flexible control over the resolution with which EEG components and events can be localized in time, space and scale.
Wavelets possess the ability to optimize the window size of its analyzing functions over the entire range of scales in EEG.
Hence both the large and small scale structures of EEG can be resolved.
This information helps in choosing the accurate bands necessary for the analysis of EEG.
84
STEPS IN DENOISING EEG
Apply Wavelet
Transform
Threshold the Noisy
Wavelet coefficients
Apply InverseWavelet
TransformNoisy EEG
Wavelet coefficients
Signal coefficients
Denoised EEG
85
86
Contaminated EEG signal is dividedinto frames of 2 second epochs
Wavelet Transform is applied and the
coefficients for all scales are obtained
Threshold value is calculated
for each scale
Using the Thresholding function
the noisy coefficients are shrinked
Apply IWT
Concatenate all the Denoised epochs
Have more epochs
yes
no
87
Successive Thresholding of Wavelet Coefficients
In (Tatjana Zikov et al 2002), the threshold limit was calculated from the baseline EEG, which is presumably artifact-free. The recording procedure to obtain such an artifact-free EEG, calls for a co-operative patient and is not only tedious but also very rarely free from contaminations.
The proposed successive thresholding algorithm eliminates the need for calculating the threshold limit from the artifact-free EEG data.
88
STEP 1: Wavelet transform is used to decompose the recorded EEG signal x[n] which is contaminated by ocular artifacts and the wavelet coefficients at each scale where, are obtained.
STEP 2: Wavelet coefficients at each scale are thresholded based on the selected threshold limit by applying an appropriate thresholding function. Thresholded wavelet coefficients are the estimate of the coefficient values of
STEP 3: Denoised (reconstructed) signal is obtained by applying the inverse wavelet transform on the thresholded wavelet coefficients.
STEP 4: Previous steps are repeated successively for appropriate number of times (stages) with the reconstructed signal as input till the desired performance goal is achieved.
j 1,....j J1 2n n nu u .......unU j J
nU j j
1 2n n n
ˆ v v .......vnV j J x [n]tr
ˆnV j
SUCCESSIVE THRESHOLDING ALGORITHM
89
Choice of the wavelet transform and the decomposition level
Stationary Wavelet Transform Time Invariant Transform. Also has better sampling rates in the lower frequency bands when compared to
DWT.
Mother Wavelet ‘Coif’ wavelet is chosen as the basis function since the shape of its mother
wavelet resembles the shape of the eye blink artifact.
Decomposition Level To have reasonable computational complexity, the decomposition level is taken
to be 5.
Threshold Limit Determines how thresholds are computed. i) Donoho’s universal threshold (Donoho et al 1995)
ii) Threshold based on statistics of the signal Tatjana Zikov et al (2002) Tj= mean (Hj) + 2.std (Hj)
Thresholding function Determines how thresholds are applied to the data.
22 ln njt
90
Results for Successive Thresholding Algorithm
db
db
Parameter
Threshold limit
Donoho mean + 2σFirststage
Secondstage
First stage
Second stage
Thirdstage
0.7894 0.8561 0.1883 0.3465 0.5357
0.5167 0.5232 0.0802 0.2665 0.3395
91
Results for Successive Thresholding Algorithm using Donoho’s Threshold limit – 10 sec epoch
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
Second stage
First stage
92
Results for Successive Thresholding Algorithm using Donoho’s Threshold limit – 1 sec epoch
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
First stage
Second stage
93
Results for Successive Thresholding Algorithm (second stage output) using Donoho’s Threshold limit for
eye movement artifact
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
94
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
First stage Second stage
Third stage
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)C
orre
latio
n co
effic
ient
Results for Successive Thresholding Algorithm usingResults for Successive Thresholding Algorithm using mean + 2σ Threshold limit – 1 sec epochThreshold limit – 1 sec epoch
95
Results for Successive Thresholding Algorithm (third stage output) using mean + 2σ Threshold limit for
eye movement artifact
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
96
Inferences:
1. Comparing the proposed wavelet based successive thresholding algorithm with component based approaches, the preservation of the high frequency contents of the original signal is good in the wavelet based approach, which is evident from the PSD and frequency correlation plots
2. Comparing the results obtained for the threshold limits, Donoho and , mean + 2σ it is noted that, thresholding using Donoho’s threshold limit, minimization of OA (both eye blink and eye movement artifacts) is better compared to the other. However preservation of underlying brain signal besides retaining the high frequency contents is good in the method which uses the mean + 2σ threshold limit.
3. Threshold values are calculated from the uncontaminated EEG itself.
4. The proposed algorithm results in intensive computations as it requires two/three successive thresholding stages to achieve the desired performance.
5. The threshold limits used are selected empirically and it is worth noting that the selection of an appropriate threshold limit and threshold function is context sensitive and needs further investigation.
97
Wavelet based approach
5. Removal of OA using Adaptive Thresholding of wavelet
coefficients
98
A nonlinear time-scale adaptive denoising system based on wavelet shrinkage scheme is proposed for removing OA from EEG.
SWTSoft-like
thresholding ISWT
CalculateRISK
Recorded EEG
Gradient based adaptation to find optimal threshold
Artifact Free EEG
Block diagram of the proposed adaptive system for removal of ocular artifact from EEG
99
Adaptive Thresholding Algorithm
STEP 1: Stationary Wavelet Transform with Coif as the basis function is used to decompose the recorded EEG contaminated by ocular artifacts.
STEP 2: Soft like thresholding function is used to find the time-scale adaptive threshold values from the initial threshold value based on MSE risk value estimated by using Stein’s Unbiased Risk Estimate (Xiao-Ping Zhang, 1998).
STEP 3: Inverse stationary wavelet transform is applied to the thresholded wavelet coefficients to obtain the artifact free EEG signal.
100
db
db
Results for Adaptive Thresholding Algorithm
Parameter
Threshold limit
Donoho
0.9513 0.74170.5054 0.4995
mean + 2σ
101
Results for Adaptive Thresholding Algorithm using Donoho’s Threshold limit
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
102
Results for Adaptive Thresholding Algorithm using Donoho’s Threshold limit for eye movement artifact
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
103
Results for Adaptive Thresholding Algorithm using mean + 2σ Threshold limit
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
104
Results for Adaptive Thresholding Algorithm using mean + 2σ Threshold limit for eye movement artifact
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
105
db
db
Parameter Successive Thresholding Adaptive ThresholdingDonoho
(second stage)Mean + 2 std(third stage)
Donoho
Mean + 2 std
0.8561 0.5357 0.9513 0.74170.5167 0.3395 0.5054 0.4995
Comparison of wavelet based successive thresholding and adaptive thresholding algorithms
106
Inferences:1. Comparing the results obtained for the adaptive thresholding method
for the threshold limits, Donoho and Mean+2std, it is noted that, thresholding using Donoho’s threshold limit, minimization of OA (both eye blink and eye movement) is good compared to the Mean+2std threshold limit. However preservation of underlying brain signal is better in the method which uses the mean + 2std threshold limit.
2. Since the OA removal algorithms (successive thresholding and adaptive thresholding) are applied to the entire length of the contaminated EEG, in addition to the minimization of OA, signal gets affected even in the non-OA zones, hence the retainment of the underlying brain signal is not appreciable, which is important for clinical diagnosis.
• Hence, there is a need for automatic identification of the slow varying OAs and application of the OA removal algorithm only to the OA affected zones, is essential.
107
Wavelet based approach
6. OA Zone Identification and Removal of OA using Wavelet Transform
108
Step 1: Contaminated EEG is decomposed by using Haar wavelet upto four levels and the coefficients are reconstructed.
Step 2: First order difference is calculated for this array.
Step 3: Mean of all the positive values and negative values are calculated.
Step 4: Positive threshold, negative threshold and difference threshold values are empirically computed using the mean values.
Step 5: Based on these three threshold values, the difference array is checked to identify the start and end time index values of the OA affected time zones.
Step 6: SWT is applied with Coif as the basis function to the contaminated EEG with OA zones identified.
Step 7: For each identified OA zone, successive thresholding algorithm / adaptive thresholding algorithm is applied to threshold the wavelet coefficients.
Step 8: ISWT is applied to the thresholded wavelet coefficients to obtain the artifact free EEG signal.
OA Zone Identification AlgorithmOA Zone Identification Algorithm
109
Decomposition of the recorded EEG with Haar wavelet and identification of OA zones
It is worth noting that reconstructed coefficients at level four results in a step function with a rising edge for a change in the state of the eyes from open to closed and a step function with a falling edge for a change in state of the eyes from closed to open.
110
A sample of EEG with both eye blink and eye movement artifact with OA affected time zones identified
0 1 2 3 4 5 6 7 8 9 10-400
-300
-200
-100
0
100
Time (seconds)
Ampl
itude
(uV)
111
COMPARISON OF SIMILARITY MEASURE
Algorithm
THRESHOLD LIMIT: DONOHO THRESHOLD LIMIT: MEAN + 2STD
SUCCESSIVE THRSEHOLDING
ADAPTIVE THRESHOLDING
SUCCESSIVE THRESHOLDING
ADAPTIVE THRESHOLDING
First stage
Second stage
First stage
Second stage
Third stage
Without zone identification εdB
0.7894 0.8561 0.9513 0.1883 0.3465 0.5357 0.7417
With zone identification
εdB
0.2240 0.2268 0.2235 0.2233 0.2248 0.2279 0.2217
112
Results for Successive Thresholding Algorithm applied to the identified OA zone using Donoho’s Threshold limit
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)C
orre
latio
n co
effic
ient
First stage
Second stage
113
Results for Successive Thresholding Algorithm applied to the identified OA zone using mean + 2σ
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-20
0
20
40
60
80
100
120
140
160
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
First stage Second stage Third stage
114
0 10 20 30 40 50 60
-20
-10
0
10
20
30
40
50
60
70
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
Results for Results for Adaptive Thresholding AlgorithmAdaptive Thresholding Algorithm applied to the applied to the identified OA zone using identified OA zone using mean + 2mean + 2σσ
0 10 20 30 40 50 60-40
-20
0
20
40
60
Frequency (Hz)A
mpl
itude
(dB
)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
115
Results for Adaptive Thresholding Algorithm applied to the identified OA zone using Donoho’s threshold limit
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
0 1 2 3 4 5 6 7 8 9 10-100
-50
0
50
100
150
200
Time (seconds)
Am
plitu
de (u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-30
-20
-10
0
10
20
30
40
50
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-40
-20
0
20
40
60
80
100
120
140
160
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)C
orre
latio
n co
effic
ient
116
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-150
-100
-50
0
50
100
150
Time(seconds)
Am
plitu
de(u
V)
EEG with artifactCorrected EEG
0 10 20 30 40 50 60-40
-30
-20
-10
0
10
20
30
40
50
60
Frequency (Hz)
Am
plitu
de (d
B)
EEG with artifactCorrected EEG
0 10 20 30 40 50 600
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Frequency (Hz)
Cor
rela
tion
coef
ficie
nt
(b) Power spectral density (c) Frequency correlation
Results for Results for Adaptive Thresholding AlgorithmAdaptive Thresholding Algorithm applied to the applied to the identified OA zone using identified OA zone using Donoho’s threshold limit Donoho’s threshold limit
117
Inferences:
1. The results obtained validate that the de-noising algorithms when applied to the identified OA zones alone, achieve superior performance to that of the existing methodologies.
2. Among the techniques compared, the Adaptive Thresholding technique using the Donoho’s threshold limit gives the best result in minimizing the magnitude of OA besides preserving the underlying brain activity and retaining the high frequency components of the original signal.
118
CONCLUSION
The component based approaches do not exactly preserve the high frequency components of the original signal, since the identification of the artifact components and subsequent removal of the artifacts is being carried out by observing the contaminated signal in time domain only.
Wavelet based algorithms retains the necessary underlying brain activity both in time and frequency domain.
Among the proposed wavelet based algorithms, adaptive thresholding algorithm applied to the identified OA zone using Donoho’s threshold limit gives the best result in minimizing the magnitude of OA and also preserves the underlying brain activity besides retaining the high frequency components of the original signal.
It eliminates the need to estimate the threshold limit from the uncontaminated EEG.
The need for reference EOG signals is not necessary.
119
FUTURE SCOPE Suitability of MILCA algorithm for removal of OA from EEG. To investigate EMD for removal of OA from EEG. Further studies can be carried out by implementing a more
powerful de-noising procedure using fast wavelet estimation to get better performance in terms of retainment of the underlying brain activity and the preservation of high frequency components.
New algorithms combining ICA and wavelet based denoising can be proposed and investigated for removal of OA from EEG recordings containing epileptic seizures without distorting the recorded ictal activity.
Further, advanced classification methods to identify components containing artifacts from the independent components is also promising.
To design and develop a Wireless EEG Recorder
120
Existing EEG Recorder Existing EEG Recorder
121
LIMITATIONS OF THE EXISTING SYSTEMLIMITATIONS OF THE EXISTING SYSTEM
Difficulty in mobility of the patient due to the wires.Difficulty in mobility of the patient due to the wires. Requires skin preparation.Requires skin preparation. Long term monitoring as in the case of epilepsy requires Long term monitoring as in the case of epilepsy requires refilling of gel.refilling of gel. Skin re-growth degrades signal quality.Skin re-growth degrades signal quality. Skin may develop allergy.Skin may develop allergy. Short circuit between two adjacent observation points may Short circuit between two adjacent observation points may occur due to excessive application of the gel.occur due to excessive application of the gel. The patient may have the fear of electrical shock. The patient may have the fear of electrical shock. High probability of misplacement of the electrodes by the High probability of misplacement of the electrodes by the technician.technician. Peadiatric measurements are a challenging issue.Peadiatric measurements are a challenging issue.
122
123
REFERENCES1. Hughes JR, “EEG in clinical Practice”, Boston MA, Butterworth’s, 1982.2. Croft RJ, Barry RJ, “Removal of ocular artifact from the EEG: A review,” Clinical Neurophysiology,
Vol. 30, No.1, pp.5-19, 2000.3. Kandaswamy A, Krishnaveni V, Jayaraman S, Malmurugan N, Ramadoss K, “Removal of Ocular
Artifacts from EEG - A Survey,” IETE Journal of Research, Vol.52, No.2, pp.21-130, 2005.4. Gratton. G, Coles MG, Donchin E, “A new method for off-line removal of ocular artifact,”
Electroencephalography and Clinical Neurophysiology, Vol.55, No.4, pp.468-484, 1983.5. Woestengurg JC, Verbaten MN, Slangen JL, “The removal of the eye movement artifact from the
EEG by regression analysis in the frequency domain,” Biological Physiology, Vol.16, pp.127-147, 1982.
6. Lagerlund TD, Sharbrough FW, Busacker NE, “Spatial filtering of multichannel electroencephalographic recordings through principal component analysis by singular value decomposition,” Clinical Neurophysiology, Vol.14, No.1, pp.73 – 82, 1997.
7. Joliffe I T, “Principal Component Analysis,” Springer Verlag New York, 1986.8. Xiao-Ping Zhang, Member, IEEE and M.Desai, Published in IEEE signal Processing letters, Vol. 5,
No. 10, 1998.9. Tzyy-Ping Jung, Scott Makeig, Colin Humphries, Te-won Lee, Martin J Mckeown, Vincent Iragui,
Terrence J Sejnowski, “Extended ICA removes Artifacts from Electroencephalographic recordings,” Advances in Neural Information Processing Systems,MIT Press Cambridge MA, Vol.10, pp.894-900, 1998.
10. Vigario R, Jaakko Sarela, Veikko Jousmaki, Matti Hamalainen, Erkki Oja, “Independent Component Approach to the Analysis of EEG and MEG Recordings,” IEEE Transactions on Biomedical Engineering, Vol.47, No.5, pp.589-593, 2000.
124
12. Delorme.A, Makeig.. S, Sejnowski T, “Automatic artifact rejection for EEG data using high-order statistics and independent component analysis,” Proceedings of the Third International ICA Conference, pp.9-12, 2000.
13. Carrie A.Joyce, Irina F Gorodnitsky, Marta Kutas, “Automatic removal of eye movement and blink artifacts from EEG data using blind component separation,” Psychophysiology, Vol.41, No.2, pp.313-325, 2004.
14. Nicolaou N, Nasuto SJ, “Temporal Independent Component Analysis for automatic artefact removal from EEG,” 2nd International Conference on Medical Signal and Information Processing, Malta, pp.5-8, 2004.
15. Tatjana Zikov, Stephane Bibian, Guy A. Dumont, Mihai Huzmezan, “A wavelet based de-noising technique for ocular artifact correction of the Electroencepahalogram,” 24th International conference of the IEEE Engineering in Medicine and Biology Society, Huston, Texas, 2002.
16. Krishnaveni V Jayaraman S, Malmurugan N, Kandaswamy A, Ramadoss K, “Non adaptive Thresholding methods for correcting ocular artifacts in EEG,” ACAD No.13, 2004.
17. Krishnaveni V, Jayaraman S, Aravind S, Hariharasudhan V, Ramadoss K, “Automatic Identification and Removal of Ocular Artifacts from EEG using Wavelet Transform,” Measurement Science Review, Vol.6, No. 4, 2006.
18. L Cohen, ‘Time-Frequency Distributions- A Review’, Proc. of IEEE, 77(7), 941-981, 1989. 19. Vincent J Samar, Ajit Bopardikar, Raghuveer Rao, Kenneth Swartz, “Wavelet Analysis of
Neuroelectric waveforms: A Conceptual Tutorial,” Brain and Language, Vol.66, pp.7-60, 1999.20. Stephane Mallat, “A wavelet tour of signal processing, Academic press,” Elsevier (USA), 1999.21. I Daubechies, ‘Orthonormal Bases of Compactly Supported Wavelets’, Comm. Pure and Applied
Math., 41, 909-996, 1988 22. S G Mallat, ‘A Theory for Multiresolution Signal Decomposition: The Wavelet Representation’, IEEE
Trans. Pattern Reco. and Machine Int., 11(7), 674-693, 1989. 23. M Vetterli, and J Kovacevic, ‘Wavelets and Subband Coding’, Prentice Hall, NJ, 1995.24. I Daubechies, ‘Ten Lectures on Wavelets’, 61, SIAM Publications, 1992.25. Quiroga, R.Q., 2000: Obtaining Single Stimulus Evoked Potentials with Wavelet Denoising, Physica,
145: 278-292.
REFERENCES…. ContdREFERENCES…. Contd
125
26. Donoho, D.L. (1995). De-noising by soft thresholding. IEEE Transactions on Information Theory. Volume 41: 613-627.
27. Donoho, D.L. et al. (1995). Wavelet shrinkage: Asymptopia?. Journal of Royal Statistics Society B(2):301-369.
28. Beylkin, G. (1992). On the representation of operators in bases of compactly supported wavelets. SIAM Journal of Numerical Analysis. 6-6:1716-1740.
29. Elvir Causevic, Robert E. Morley, M. Victor Wickerhauser, Arnaud E. Jacquin, Fast Wavelet Estimation of Weak Biosignals.
30. Burke, M.J., and Gleeson, D.T. (2000). A micropower dry-electrode ECG Preamplifier. IEEE Transactions on Biomedical Engineering. Volume 47:2.
31. Causevic, E. (2001). Fast Wavelet Estiomation of Weak Biosignals. Ph.D. Thesis, Washington University.32. Coifman, R. R. and M. V. Wickerhauser (1995). Adapted waveform ‘de-noising’ for medical signals and
images. IEEE Engineering in Medicine and Biology 14((5) September/October): 578-586.33. Coifman, R.R., Donoho, D.L. (1995). Translation invariant denoising. Technical Report 475. Department of
Statistics, Stanford University.34. Donoho, D.L. (1995). De-noising by soft thresholding. IEEE Transactions on Information Theory. Volume
41: 613-627.35. Donoho, D.L. et al. (1995). Wavelet shrinkage: Asymptopia?. Journal of Royal Statistics Society B.
57(2):301-369.36. R R Coifman and D L Donoho, “Translation Invariant denoising,” Lecture notes in Statistics, No.103,
pp.125-159, 1995.37. Donald B Percival, Andrew Walden 2000, “Wavelet methods for Time series analysis,” Cambridge
University Press.38. Xiao-Ping Zhang and Mita D. Desai, “Adaptive Denoising Based on SURE Risk ,” IEEE Signal processing,
No.10, pp.265-267, 1998.39. http://www.sccn.ucsd.edu/~arno/famzdata/publicly_available_EEG_data.html40. Shoker L, Sanei S and Chambers J, “Artifact removal from electroencephalograms using a hybrid BSS–SVM
algorithm,” IEEE Signal Process., No.12, pp.721–724, 2005.
REFERENCES…. ContdREFERENCES…. Contd
126
LIST OF PUBLICATIONSInternational Journals:1. “Non Adaptive Thresholding Methods for correcting ocular artifacts in EEG” Academic Open Internet Journal, Technical College, Bourgas 8010, Bulgaria Volume 13, 2004
2. “Comparison of Independent Component Analysis Algorithms for Removal of Ocular Artifacts from Electroencephalogram” Measurement Science Review Journal – Measurement in Biomedicine, Volume 5, Section 2, 2005 pp 68-78
3. “Application of Mutual Information based Least dependent Component Analysis (MILCA) for Removal of Ocular Artifacts from Electroencephalogram” International Journal of Biomedical Sciences, Volume 1 Number 1, 2006 ISSN 1306-1216, 63-74
4. “Automatic Identification and Removal of Ocular Artifacts from EEG using Wavelet Transform” Measurement Science Review Journal – Measurement in Biomedicine, Volume 6, Section 2, No.4, 2006 pp 45-572.
127
LIST OF PUBLICATIONS - contd
5. “Removal of Ocular Artifacts from EEG using Adaptive Thresholding of Wavelet Coefficients” Institute of Physics Publishing, UK, Journal of Neural Engineering Vol. 3 (2006) pp 338–346
6. “Automatic removal of ocular artifacts using JADE algorithm and neural network” International Journal of Biomedical Sciences, Volume 2 Number 1, 2007 ISSN 1306-1216, pp 10-21
7. “Removal of Ocular Artifacts from EEG – A Survey” IETEJournal of Research, Vol 51, No.2, Mar-Apr 2005, pp 121-130
National Journal:8. V Krishnaveni, S Jayaraman, N Malmurugan, K Ramadoss and Chaitanya Mathi, “Quantitative Evaluation of Signal Separation Algorithms for removal of ocular artifacts from EEG” National Journal of Technology, No.2 Vol.1, June 2005, pp 47-53
128
LIST OF PUBLICATIONS - contdInternational Conferences:9. “Automatic Removal of Ocular Artifacts from EEG using TDSEP (Temporal Decorrelation Source Separation) Algorithm” International conference on Robotics, Vision, Signal Processing and Information, Universiti Sains, Penang, Malaysia, July 20-2210. “Automatic Identification of Ocular Artifacts from Independent Components of EEG signals using Kalman Predictor” Thirteenth International Conference on Advanced Computing & Communications December 14-17, 200511. “Automatic Removal of Ocular Artifacts from Electroencephalogram using Least Dependent Component Analysis and Wavelet Transform” International Conference on Resource Utilisation and Intelligent Systems, Kongu Engineering College, Perundurai, Erode, January 4-6, 2006
National Conferences:12. “Ocular Artifact correction in EEG using Wavelet Thresholding”, First National Conference on Advances in Electronic Communications Conducted by National Engineering College Kovilpatti during June 3rd & 4th 2004.13. “Performance study of denoising algorithms based on wavelet thresholding and Independent Component Analysis for correction of Ocular Artifacts in EEG” First National Conference on Advances in Electronic Communications Conducted by National Engineering College, Kovilpatti during June 3rd & 4th 2004
CITATIONS
129
130
MRI IMAGE ANALYSISMRI IMAGE ANALYSIS
CERTAIN INVESTIGATIONS ON MAGNETIC RESONANCE
IMAGE DENOISING METHODS BASED ON NEUTROSOPHIC SET
THEORY
131
OUTLINE OF THE PRESENTATION
Introduction - MR Imaging, Noise in MR Images, Denoising Literature Survey Motivation for the Thesis Objective of the Thesis Proposed Methodologies MRI Dataset and Validation Strategies Conclusion and Future Scope References List of Publications
132
Magnetic Resonance Imaging (MRI) is a powerful medical imaging technique used in diagnosis and clinical applications.
It is used to visualize the detailed internal structure of the human body in order to demonstrate pathological or other physiological alterations of living tissues.
MR images are acquired complete non-invasively without causing any radiation risk to the patients.
133
4
The process to obtain the MR Image is
The magnitude of the MRI signal is the square root of the sum of the squares of two independent Gaussian variables, it follows Rician distributionRician distribution.
The Rician noise is signal dependent and is therefore difficult to separate from the signal.
IDFT
(Gudbjartsson and Partz 1995)
135
In parallel MRI, data collected by each receiver element in k- space is subsampled data.
Statistical point of view, the reconstruction will affect the distribution of the noise in the reconstructed data.
Variance of noise may vary for different image location.
It is highly desirable to develop filtering method that remove this noise.
(Thunberg and Zetterberg 2007, Dietrich et al 2008)
136
Effect of Noise in MR Images
The noise in MR images not only affect the diagnostic and visual quality of MR Images.
This is also problematic for further image processing tasks:
o Segmentation of important features o Classification of images for computer aided
diagnosticso Three dimensional image reconstruction &
image registration etc.
137
Denoising – important preprocessing step used to improve image quality by reducing the noise component while preserving all the image features.
Preservation of the meaningful image structures is a necessary condition for developing MRI denoising methods.
Denoising138
The existing MR image denoising methods can be classified as:
a) Filtering approach,
b) Transform based approach &
c) Statistical approach.
Literature Survey139
Spatial
Temporal
Bilateral
Trilateral
Fast NLM
Blockwise optimized NLM
Unbiased NLM
Dynamic NLM
Enhanced NLM
Adaptive Rician NLM (ARNLM)
Adaptive 4th order PDE
4th order complex PDE
Adaptive ADF
Noise Driven ADF
Noise Adaptive ADF
Anisotropic Diffusion Filter (ADF)
4th order Partial Differential Equations (PDE)
Nonlocal Mean (NLM) filter
Combination of Domain and Range Filters
Linear Non-Linear
Filtering Approach Transform based Approach Statistical Approach
Wavelet Transform
Wavelet Thresholding
Wavelet domain filter
Wavelet packet analysis
Adaptive Multiscale Product Thresholding
Multiwavelet
Undecimated Wavelet
Total Variation (TV) minimization scheme
Maximum likelihood (ML) Estimation
Phase Error Estimation
Nonparametric estimation
Singularity function analysis
Linear Minimum Mean Square Error Estimation
Nonlocal ML (NLML)
Restricted local ML (RLML)
140
Spatially uniform noise distribution
MRI Denoising methods Limitations / Disadvantages
Spatial filtering (McVeigh et al 1985)
Blurring edges by averaging pixels with non similar patterns (suitable only for Gaussian noise).
Temporal filtering (McVeigh et al 1985)
Proper selection of frequency response for the filter is important in order to avoid aliasing.
Anisotropic diffusion filtering (Gerig et al 1992, Sijbers et al 1999, Murase et al 2001, Lysaker et al 2003, Samsonov & Johnson 2004, Tang et al 2007, Krissian & Aja-Fernandez 2009, Lu et al 2009, Rajan et al 2009, Zhang & Ma 2010)
Usually erases small features and transforms image statistics due to its edge enhancement causes blocky (staircase) effect in the image.
141
142
MRI Denoising methods
Limitations / Disadvantages
Nonlocal mean (Coupe et al 2006, Manjon et al 2007, Wiest-Daessle et al 2007, Coupe et al 2008, Manjon et al 2008, Wiest-Daessle et al 2008, Gal et al 2009, Liu et al 2010, Manjon et al 2010)
Computational burden due to its complexity of calculating the weight of the pixel/voxel.
Bilateral and Trilateral filtering (Wong & Chung 2004, Walker et al 2006, Hamarneh & Hradsky 2007)
large scale structures are preserved, while small structures are considered as noise and are removed.
Spatially uniform noise distribution
143
MRI Denoising methods Limitations / Disadvantages
Wavelet based denoising (Healy & Weaver 1992, Nowak 1999, Wood & Johnson 1999, Zaroubi & Goelman 2000, Alexander et al 2000, Pizurica et al 2003, Bao & Zhang 2003, Placidi et al 2003, Wu et al 2003, Mrazek et al 2005, Delakis et al 2007, Ashamol et al 2008, Yu & Zhao 2008, Anand & Sahambi 2008, Tan & Shi 2009, Anand & Sahambi 2010)
May introduce characteristic artifacts that can be quiet problematic and also difficult to confirm the scale and threshold of the wavelet.
Nonparametric neighborhood statistics method (Awate & Whitaker 2005, Awate & Whitaker 2007)
Optimization of information theoretic metric using expectation-maximization algorithm to avoid ill-fitting prior models is a difficult task.
Spatially uniform noise distribution
Spatially non-uniform noise distribution
MRI Denoising methods
Limitations / Disadvantages
Noise Adaptive Nonlinear Diffusion filter (Samsonov & Johnson 2004)
Requires the calculation of the noise map from the receiver coil matrix. This may be implemented by MRI scanner’s reconstruction software, (available with manufacturers).
Wavelet based denoising (Delakis et al 2007)
Local noise variance is estimated from the wavelet decomposition high frequency subband after discarding edge pixels. Therefore, the discarded edge pixels and details in the periphery may result in loss of information.
Adaptive Nonlocal means (Manjon et al 2010)
Computational burden due to algorithmic complexity. So, neglecting the voxels/ blocks with small weights (i.e., most dissimilar patches to the current one) for speeding up the filter.
144
From these literature, it is found that the post processing noise reduction techniques is considered to be the only means of achieving a desired MR Image quality which is of extreme importance in more accurate diagnosis.
A trade-off between noise reduction and preservation of the useful image features still persists and is a challenging problem in MR Image denoising.
Motivation for the Thesis145
This thesis work is confined to development of certain methodologies for denoising MR images acquired using both single-coil and multi-coil acquisition based on Neutrosophic Set (NS) theory to achieve a balance between noise reduction and structure preservation.
146
Motivation for the Thesis (Contd..)
Objective of the Thesis147
Propose and analyze new efficient algorithms for denoising MR Images based on Neutrosophic set (NS) logic which satisfies the following criteria:
o Removal of noise from the noisy MR Image.o Preservation of the anatomical structures present in the image.
148
1. For denoising, single-coil acquired MR Image which has uniformally distributed Rician noise (Gudbjartsson and Partz 1995, Dietrich et al 2008):
i. Neutrosophic set approach of median filtering (NS median) ii. Neutrosophic set approach of Wiener filtering (NS Wiener) iii.Nonlocal neutrosophic set approach of Wiener filtering (NLNS Wiener).
Proposed Methodologies
149
2. For denoising, multi-coil acquired MR Image which has spatially varying noise (Thunberg and Zetterberg 2007):
i. Nonlocal neutrosophic set approach of spatially adaptive median filtering (NLNSAM).
3. Application: NLNS Wiener method is used as the pre-processing step for automated brain tumor segmentation.
The MRI datasets used to analyze the proposed methodologies are:
o Simulated MR Images obtained from the Brainweb database (Kwan et al 1999), (http://brainweb.bic.mni.mcgill.ca/brainweb/).
o Clinical MR Images collected from PSG Institute of Medical Sciences and Research (PSG IMS & R), Coimbatore, Tamilnadu, India.
MRI Dataset150
The evaluation of the denoising method depends on the balance between noise removal and the structure preservation.
For Noise Removal:o Peak signal to noise ratio (PSNR),o Structural similarity index (SSIM), o Bhattacharyya coefficient (BC) & o Mean absolute difference (MAD)
Structure preservation, Visual inspection of the denoised images and the residual images.
Validation Strategies151
Single-coil acquired MR Image Denoising
Neutrosophic Set Approach of Median Filter (NS MEDIAN)
152
Neutrosophic Set approach of MRI denoising153
Algorithm for NS Median The new neutrosophic approach to MRI denoising is described as below:
Step 1: Transform the image into NS domain;
Step 2: Use - median filtering operation on the true subset T to obtain ;
T
Step 3: Compute the entropy of the indeterminate subset )(,ˆˆ iEnI I
Step 4: if
)(
)()1(
ˆ
ˆˆ
iEn
iEniEn
I
II, go to Step 5;Else TT ˆ , go to Step 2;
Step 5: Transform subset T from the neutrosophic domain into the gray level
domain.
154
FLOW CHART155
The performance of the proposed NS median method is compared with standard algorithms such as:
Anisotropic diffusion filter (ADF) (Perona and Malik 1990),
Total variation (TV) minimization scheme (Rudin et al 1992)
Nonlocal maximum likelihood (NLML) method (He andGreenshields 2009).
Results and Discussion156
Quantitative Analysis of NS Median - PSNR157
T1 weighted MR images
23.4 dB
158
T1 weighted MR images
Quantitative Analysis of NS Median - SSIM
0.9604
159
T1 weighted MR images
Quantitative Analysis of NS Median - BC
0.854
160
T1 weighted MR images
Quantitative Analysis of NS Median - MAD
8.27
161
MR ImageDenoising Methods
Performance Metrics
PSNR (dB) SSIM BC MAD
T1- Weighted brain MRI with MS lesion corrupted by 7% Rician noise
TV 23.61 0.9521 0.84 7.19
ADF 22.58 0.9637 0.832 7.42
NLML 24.33 0.9657 0.867 6.44
NS Median 25.52 0.9691 0.89 6.28
PD- Weighted brain MRI corrupted by 9% Rician noise
TV 23.06 0.923 0.822 9.69
ADF 23.24 0.9386 0.834 9.96
NLML 23.29 0.9594 0.849 8.27
NS Median 23.4 0.9604 0.854 8.21
T2- Weighted brain MRI with MS lesion corrupted by 15% Rician noise
TV 18.21 0.8261 0.795 13.62
ADF 18.35 0.841 0.805 13.75
NLML 18.68 0.914 0.827 12.51
NS Median 19.71 0.922 0.831 11.92
Performance Comparison
162
Denoising results of simulated T1 weighted axial MR image: (a) Original image (b) Original image corrupted with 9% of Rician noise (c) denoised with TV method (d) denoised with ADF method (e) denoised with NLML method (f) denoised with the proposed NS Median method (g) , (h), (i) and (j) are the corresponding residual images
163
Denoising results of simulated T2 weighted axial MR image: (a) Original image (b) Original image corrupted with 15% of Rician noise (c) denoised with TV method (d) denoised with ADF method (e) denoised with NLML method (f) denoised with the proposed NS Median method (g) , (h), (i) and (j) are the corresponding residual images
164
Denoising results of small part of the T1 weighted axial MR image with MS lesions: (a) Original image (b) Original image corrupted with 9% of Rician noise (c) denoised with TV method (d) denoised with ADF method (e) denoised with NLML method (f) denoised with the proposed NS Median method (g) , (h), (i) and (j) are the corresponding residual images
165Analysis on Clinical Dataset
Denoising results of clinical T2 weighted Coronal MRI. (a) Original image (b) denoised with TV method (c) denoised with ADF method (d) denoised with NLML method (e) denoised with the proposed NS Median method (f), (g) , (h) and (i) are the corresponding residual images
166
Denoising results of clinical T1 weighted axial MRI with granulomatous lesion pathology. (a) Original image (b) denoised with TV method (c) denoised with ADF method (d) denoised with NLML method (e) denoised with the proposed NS Median method (f), (g) , (h) and (i) are the corresponding residual images
These results demonstrate that the proposed NS median filter performs better than the TV minimization scheme and AD filter.
The NLML method performs well compared to the proposed NS median filter at low noise levels (1% - 3% of Rician noise) and as the noise level increases (4% - 15% of Rician noise), the proposed NS median filter performs better than the NLML method.
It is found from the residual images, that the traces of anatomical structures are less seen in the proposed NS median filter compared to TV, ADF and NLML methods.
Even though the proposed NS median method produces the good denoised images which are having strong correlation with the original noise-free images but the residual images of the NS median method have fewer traces of the anatomical structures.
Hence, there is a need for improvement in the proposed method.
Inferences 167
168Performance ComparisonMR Image Denoising Methods
Performance Metrics
PSNR (dB) SSIM BC MAD
PD- Weighted brain MRI with MS lesions corrupted by 7% Rician noise
Wiener 24.42 0.9427 0.835 7.2TV 24.54 0.9545 0.84 7.5ADF 25.37 0.9642 0.832 7.12NLM 25.36 0.9647 0.85 6.52NLML 26.52 0.9694 0.876 6.45RLML 26.62 0.9734 0.883 6.4NS Median 27.19 0.9787 0.89 6.18NS Wiener 29.5 0.9889 0.9 4.52NLNS Wiener 31.37 0.9895 0.902 4.18
T2-Weighted brain MRI with MS lesions corrupted by 9% Rician noise
Wiener 22.12 0.8337 0.82 10.06TV 22.59 0.9345 0.809 9.92ADF 22.67 0.9466 0.816 10.09NLM 23.39 0.9495 0.823 8.86NLML 23.75 0.9512 0.835 8.73RLML 23.96 0.9594 0.843 8.36NS Median 24.4 0.9696 0.85 8.28NS Wiener 26.19 0.9887 0.86 6.39NLNS Wiener 27.24 0.9816 0.88 5.65
T1- Weighted brain MRI with MS lesions corrupted by 15% Rician noise
Wiener 16.78 0.7870 0.808 13.71TV 16.82 0.7901 0.809 13.69ADF 16.84 0.8127 0.806 13.74NLM 16.88 0.8370 0.83 13.241NLML 18.96 0.8426 0.836 12.87RLML 19.22 0.8842 0.841 12.21NS Median 20.51 0.9062 0.846 11.657NS Wiener 23.53 0.9263 0.855 9.17NLNS Wiener 25.23 0.9284 0.867 8.65
Application
Automated Brain Tumor Segmentation on MR Images
169
To enhance the performance of the automated brain tumor segmentation technique, the NLNS Wiener filtering is used as the preprocessing step to improve the MR Image quality.
The fuzzy image enhancement technique is used in the NLNS Wiener filtered image.
The k- means clustering method is used for brain tumor segmentation.
The metrics used for evaluation of the segmentations are
1. Sensitivity 2. Specificity
3. False position rate (FPR) 4. False negative Rate (FNR)
5. Jaccard similarity Rate 6. Dice Coefficient
Procedure170
171
Results of Automatic Brain tumor segmentation method with pre-processed using NLNS Wiener filter. (a) Original image, (b) Filtered image using nonlocal NS Wiener filter, (c) Enhanced image using Fuzzy intensification (d) Tumor traced by experts, (e) Tumor segmented by the proposed method
Results and Discussion
172Results of automatic brain tumor segmentation method with pre-processed using NLNS Wiener filter
Image Accuracy Sensitivity Specificity FPR FNR Jaccard Dice
Case 1 0.9961 0.8419 0.9992 0.0008 0.1581 0.8103 0.8952Case 2 0.9681 0.8659 0.9982 0.0018 0.1341 0.7367 0.8484Case 3 0.9816 0.9054 0.9882 0.0118 0.0946 0.7960 0.8864Case 4 0.9874 0.8744 0.9935 0.0065 0.1256 0.8023 0.8703Case 5 0.9993 0.9757 0.9996 0.0004 0.0233 0.9431 0.9707Case 6 0.9861 0.8539 0.9892 0.0108 0.1461 0.8215 0.8852Case 7 0.9542 0.8734 0.9854 0.0146 0.1266 0.8365 0.8885Case 8 0.9819 0.9154 0.9982 0.0018 0.0846 0.8960 0.8964Case 9 0.9878 0.8774 0.9945 0.0045 0.1226 0.8123 0.8803
Case 10 0.9953 0.9757 0.9986 0.0014 0.0233 0.9331 0.9207Case 11 0.9931 0.8615 0.9965 0.0035 0.1385 0.8726 0.8957Case 12 0.9784 0.9659 0.9982 0.0018 0.0341 0.8367 0.8788Case 13 0.9719 0.9154 0.9982 0.0018 0.0846 0.8965 0.8990Case 14 0.9943 0.9757 0.9996 0.0004 0.0233 0.9523 0.9776Case 15 0.9681 0.8672 0.9962 0.0038 0.1328 0.7367 0.8484Case 16 0.9961 0.8419 0.9992 0.0008 0.1581 0.8103 0.8952Case 17 0.9642 0.8734 0.9864 0.0136 0.1266 0.8567 0.8975Case 18 0.9876 0.9154 0.9870 0.0130 0.0846 0.7960 0.8864Case 19 0.9874 0.8744 0.9935 0.0065 0.1256 0.8023 0.8703Case 20 0.9956 0.9754 0.9943 0.0057 0.0236 0.9435 0.9710Average 0.9837 0.9012 0.9947 0.0053 0.0988 0.8446 0.8981
173Performance comparison of automatic brain tumor segmentation by using various denoising methods
Denoising Method
Accur-acy
Sensit-ivity
Specifi-city
FPR FNR Jaccard Dice
Wiener 0.9554 0.8523 0.9625 0.0375 0.1477 0.8056 0.8562
TV 0.9678 0.8589 0.9687 0.0313 0.1411 0.8094 0.8596
ADF 0.9548 0.8687 0.9716 0.0284 0.1313 0.8148 0.8654
NLM 0.9715 0.8756 0.9795 0.0205 0.1244 0.8231 0.8714
NLML 0.9792 0.8897 0.9846 0.0154 0.1103 0.8358 0.8860
RLML 0.9805 0.8974 0.9901 0.0099 0.1026 0.8368 0.8904
NS Median 0.9789 0.8895 0.9845 0.0155 0.1105 0.8352 0.8852
NS Wiener 0.9812 0.8983 0.9910 0.0090 0.1017 0.8406 0.8913
NLNS Wiener 0.9837 0.9012 0.9947 0.0053 0.0985 0.8446 0.8981
174
Conclusions and Future Scope
175
Denoising is necessary to be performed to improve the image quality for more accurate diagnosis.
Three methods are proposed for denoising MR images with uniformly distributed Rician noise based on neutrosophic set theory namely NS median, NS Wiener and NLNS Wiener methods. All the methods provide better performance in terms of quantitative and qualitative measures compared to other methods available in the literature.
For denoising MR images with spatially varying noise distribution, NLNSAM filtering method is proposed and it works well compared with ARNLM (Manjon et al 2010) method.
For automated brain tumor segmentation, NLNS Wiener filtering is used as the pre-processing step for in order to enhance the performance of segmentation and an average of high accuracy of 98.37%, high specificity of 99.47% and lower missing rate of 0.53%
Conclusions
The suitability of LMMSE (Linear Minimum Mean Square Error) and ML (Maximum likelihood) estimation methods combined with NS approach for denoising MR images can be investigated.
New algorithms combining NS and wavelet denoising can be proposed and investigated for denoising MR images.
For multiple coil images, the noise may also follow noncentral chi distribution. In future, a method can be developed to denoise MR images with noncentral chi distribution.
Another extension to this work is to develop the FPGA implementation of the NS median, NS Wiener, and NLNS Wiener methods.
Future Scope 176
References1.Aja-Fernandez, S, Alberola-López, C & Westin, CF 2008a, ‘Noise and signal estimation in magnitude MRI and Rician distributed images:
A LMMSE approach’, IEEE Transactions on Image Processing, vol. 17, no. 8, pp. 1383–1398. 2. Aja-Fernandez, S, Niethammer, M, Kubicki, M, Shenton, ME & Westin, CF 2008b, ‘Restoration of DWI data using a Rician LMMSE
estimator’, IEEE Transactions on Medical Imaging, vol. 27, no. 10, pp. 1389–1403.3. Aja-Fernandez, S, Tristan-Vega, A & Alberola-Lopez, C 2009, ‘Noise estimation in single and multiple coil MR data based on statistical
models’, Magnetic Resonance Imaging, vol. 27, no. 10, pp. 1397-1409.4. Alexander, ME, Baumgartner, R, Summers, AR, Windischberger, C, Klarhoefer, M, Moser, E & Somorjai, RL 2000, ‘A wavelet-based
method for improving signal-to-noise ratio and contrast in MR images’ Magnetic Resonance Imaging, vol. 18, no. 2, pp. 169–180.5. Anand, CS & Sahambi, JS 2008, ‘MRI Denoising Using Bilateral Filter in Redundant Wavelet Domain’, Proceedings of the IEEE Region
tenth Conference, pp.1-6.6. Anand, CS & Sahambi, JS 2010, ‘Wavelet domain non-linear filtering for MRI denoising’, Magnetic Resonance Imaging, vol. 28, no.6,
pp. 842-861. 7. Ashbacher, C 2002, Introduction to Neutrosophic Logic, American Research Press, Rehoboth.
8. Awate, SP & Whitaker, RT 2005, ‘Nonparametric neighborhood statistics for MRI denoising’, Proceedings of the nineteenth international conference on information processing in medical imaging (IPMI), ed. G.E. Christensen & M. Sonka, IPMI, Colorado, vol. 3565, pp.
677-688.9. Awate, SP & Whitaker, RT 2007, ‘Feature-preserving MRI denoising: A nonparametric empirical Bayes approach’, IEEE Transactions
on Medical Imaging, vol. 26, no. 9, pp. 1242–1255.10.Bao, P & Zhang, L 2003, ‘Noise reduction for magnetic resonance images via adaptive multiscale products thresholding’, IEEE
Transactions on Medical Imaging, vol. 22, no.9, pp. 1089–1099. 11.Bao, L, Li, Y, Zhu, Y, Pu, ZB & Magnin, IE 2008, ‘Sparse representation based MRI denoising with total variation’, Proceedings of the
ninth international conference on signal processing, pp. 2154-2157. 12.Bhattacharyya, A 1943, ‘On a measure of divergence between two statistical populations defined by their probability distributions’,
Bulletin of the Calcutta Mathematical Society, vol. 35, pp.99-109. 13.Bloch, F 1946, ‘Nuclear induction’, Physical Review, vol. 70, no. 7-8, pp. 460-474.
Buades, A, Coll, B & Morel, JM 2005, ‘A review of image denoising algorithms, with a new one’, Multiscale Modeling Simulation, vol. 4, no. 2, pp. 490-530.
14.Clark, MC, Hall, LO, Goldgof, DB, Velthuzen, R, Murtagh, FR & Silbiger MS 1998, ‘Automatic tumor segmentation using knowledge based techniques’, IEEE Transactions on Medical Imaging, vol.17, no. 2, pp. 187-201.
177
15.Clarke, LP, Velthuizen, RP, Heine, JJ, Vaidyanathan, M, Hall, LO, Thatchar, RW & Silbiger, ML 1995, ‘MRI Segmentation: Methods and Applications’, Magnetic Resonance Imaging, vol. 13 no. 3, pp. 343-368.
16.Constantinides, CD, Atalar, E & McVeigh, ER 1997, ‘Signal-to-noise measurements in magnitude images from NMR phased arrays’, Magnetic Resonance in Medicine, vol. 38, no. 5, pp. 852–857.
17.Coupe, P, Yger, P & Barillot, C 2006, ‘Fast nonlocal means denoising for MR images’. Proceedings of the ninth international conference on medical image computing and computer-assisted intervention (MICCAI), ed. R. Larsen, M. Nielsen & J. Sporring,
MICCAI, Denmark, pp. 33-40.18.Coupe, P, Yger, P, Prima, S, Hellier, P, Kervrann, C & Barillot, C 2008, ‘An optimized blockwise nonlocal means denoising filter for 3-
D magnetic resonance images’, IEEE Transactions on Medical Imaging, vol. 27, no. 4, pp. 425–441.19.Coupe, P, Manjon, JV, Gedamu, E, Arnold, D, Robles, M, & Collins, DL 2009, ‘An Object-based Method for Rician noise estimation in
MR images’, Proceedings of the twelfth international conference on medical image computing and computer-assisted intervention (MICCAI), ed. G. Yang, D. Hawkes, D. Rueckert, A. Noble & C. Taylor, MICCAI, London, pp. 601-608.
20.Coupe, P, Manjon, JV, Gedamu, E, Arnold, D, Robles, M & Collins, DL 2010, ‘Robust Rician noise estimation for MR images’, Medical Image Analysis, vol. 14, no. 4, pp. 483-493.
21.Coupe, P, Manjon, JV, Robles, M & Collins, DL 2012, ‘Adaptive multiresolution nonlocal means filter for 3D MR image denoising’, IET Image Processing, vol. 6, no. 5, pp. 558-568.
22.Damadian, R 1971, ‘Tumor detection by nuclear magnetic resonance’, Science, vol. 171, no. 3976, pp. 1151–1153. Delakis, I, Hammad, O & Kitney, RI 2007, ‘Wavelet-based de-noising algorithm for images acquired with parallel magnetic resonance
imaging (MRI)’, Physics in Medicine and Biology, vol. 52, no. 13, pp. 3741–3751.23.Dietrich, O, Raya, JG, Reeder, SB, Reiser, MF & Schoenberg, SO 2007, ‘Measurement of signal-to-noise ratios in MR images: influence
of multichannel coils, parallel imaging, and reconstruction filters’, Magnetic Resonance Imaging, vol. 26, no. 2, pp. 375–385.24.Dietrich, O, Raya, JG, Reeder, SB, Ingrisch, M, Reiser, MF & Schoenberg, SO 2008, ‘Influence of multichannel combination, parallel
imaging and other reconstruction techniques on MRI noise characteristics’, Magnetic Resonance Imaging, vol. 26, no. 6, pp. 754–762.25.Carr, H 1952, Free Precession Techniques in Nuclear Magnetic Resonance. PhD thesis, Harvard University, Cambridge, USA.
26.Cheng, HD & Guo, Y 2008, ‘A new neutrosophic approach to image thresholding’, New Mathemetics and Natural Computation, vol. 4, no. 3, pp. 291-308.
27.Cheng, HD, Guo, Y & Zhang, Y 2011, ‘A novel image segmentation approach based on neutrosophic set and improved fuzzy c-means algorithm’, New Mathemetics and Natural Computation, vol. 7, no. 1, pp. 155-171.
28.Donoho, DL 1995, ‘De-Noising by Soft-Thresholding’, IEEE Transactions on information theory, vol. 41, no. 3, pp. 613-627. Dou, W, Ruan, S, Chen, Y, Bloyet, D & Constants, JM 2007, ‘A framework of fuzzy information fusion for segmentation of brain
tumor tissues on MR images’, Image and Vision Computing, vol. 25, no. 2, pp. 164-171.
178
179
29.Edelstein, WA, Bottomley, PA & Pfeifer, LM 1984, ‘A signal-to-noise calibration procedure for NMR imaging systems, Medical Physics, vol.11, no. 2, pp. 180-185.
30.Gal, Y, Mehnert, AJH, Bradley, AP, McMahon, K, Kennedy, D & Crozier, S 2009, ‘Denoising of Dynamic Contrast- Enhanced MR Images Using Dynamic Nonlocal Means’, IEEE Transactions on Medical Imaging, vol. 29, no. 2, pp. 302-310.
31.Gerig, G, Kubler, O, Kikinis, R & Jolesz, FA 1992, ‘Nonlinear anisotropic filtering of MRI data’ IEEE Transactions on Medical Imaging, vol. 11, no. 2, pp. 221–232.
32.Golshan, HM, Hasanzedeh, RPR & Yousefzadeh, SC 2013, ‘An MRI denoising method using data redundancy and local SNR estimation’, Magnetic Resonance Imaging, vol. 31, no. 7, pp. 1206-1217.
33.Gonzalez, RC & Woods, RE 2002, Digital Image Processing, second edition, Prentice-Hall, New Delhi, India. 34.Gordillo, N, Montseny, E & Sobrevilla, P 2013, ‘State of the art survey on MRI brain tumor segmentation’, Magnetic Resonance
Imaging, vol. 31, no. 8, pp. 1426-1438. 35.Griswold, MA, Jakob, PM, Heidemann, RM, Nittka, M, Jellus, V & Wang, J 2002, ‘Generalized autocalibrating partially parallel
acquisitions (GRAPPA)’, Magnetic Resonance in Medicine, vol. 47, no. 6, pp. 1202–1210.
36.Gudbjartsson, H & Patz, S 1995, ‘The Rician distribution of noisy MRI data’, Magnetic Resonance in Medicine, vol. 34, no. 6, pp. 910–914.
37.Guo, W & Huang, F 2009, ‘Adaptive total variation based filtering for MRI images with spatially inhomogeneous noise and artifacts’, Proceedings of the IEEE international symposium on biomedical imaging: nano to macro, pp. 101-104.
38.Guo, Y, Cheng, HD & Zhang, Y 2009, ‘A new neutrosophic approach to image denoising’, New Mathemetics and Natural Computation, vol. 5, no. 3, pp. 653-662.
39.Guo, Y & Cheng, HD 2009, ‘New neutrosophic approach to image segmentation’, Pattern Recognition, vol. 42, no. 5, pp. 587-595.40.Guo, Y & Sengur, A 2013, ‘A novel color image segmentation approach based on neutrosophic set and modified fuzzy c-means’, Circits,
Systems and Signal Processing, vol. 32, no. 4, pp. 1699-1723. 41.Hamarneh, G & Hradsky, J 2007, ‘Bilateral Filtering of diffusion tensor magnetic resonance images’, IEEE Transactions on Image
Processing, vol. 16, no. 10, pp. 2463-2475.42.Harati, V, Khayati, R & Farzan A 2011, ‘Fully automated tumor segmentation based on improved fuzzy connectedness algorithm in brain MR images’, Computers in Medicine and Biology, vol. 41, no. 7, pp. 483-492.
43.He L & Greenshields IR 2009, ‘A Nonlocal Maximum Likelihood Estimation Method for Rician noise reduction in MR images’, IEEE
Transactions on Medical Imaging, vol. 28, no. 2, pp. 165-172.44.Healy, DM & Weaver, JB 1992, ‘Two applications of wavelet transforms in magnetic resonance imaging’, IEEE Transactions on
Information Theory, vol. 38, no. 2, pp. 840-860.
18045.Henkelman, RM 1985, ‘Measurement of signal intensities in the presence of noise in MR images’, Medical Physics, vol. 12, no. 2, pp. 232–233.
46.James, ML & Gambhir SS 2012, ‘A Molecular imaging primer: modalities, imaging agents and applications’, Physiological Reviews, vol. 92, no. 2, pp. 897-965.
47.Jiang, L & Yang, W 2003, ‘Adaptive magnetic resonance image denoising using mixture model and wavelet shrinkage’, Proceedings in the seventh international conference on digital image computing: techniques and applications (DICTA), ed. C. Sun, H. Talbot, S.
Ourselin & T. Adriaasen, DICTA, Sydney, pp. 831-838. 48.Karabatak, E, Guo, Y & Sengur, A 2013, ‘Modified neutrosophic approach to color image segmentation’, Electronic Imaging, vol. 22,
no. 1, 013005.49.Kaus, MR, Warfield, SK, Nabavi, A, Black PM, Jolesz, FA & Kikinis, R 2001, ‘Automated segmentation of MRI of brain tumors’,
Radiology, vol. 218, no. 2, pp. 586-591. 50.Krissian, K & Aja-Fernández, S 2009, ‘Noise driven Anisotropic Diffusion filtering of MRI’, IEEE Transactions on Image Processing,
vol. 18, no. 10, pp. 2265-2274.51.Kuperman, V 2000, Magnetic Resonance Imaging Physical Principles and Applications, Academic Press, San Diego, USA.
52.Kwan, RK, Evans, AC & Pike, GB 1999, ‘MRI simulation based evaluation of image processing and classification methods’, IEEE Transactions on Medical Imaging, vol. 18, no. 11, pp. 1085-1097.
53.Larkman, DJ & Nunes, RG 2007, ‘Parallel magnetic resonance imaging’, Physics in Medicine and Biology, vol. 52, no. 7, pp. 15–55. 54.Lauterbur, PC 1973, ‘Image formation by induced local interactions: examples of employing nuclear magnetic resonance’, Nature, vol.
242, pp. 190–191.55.Lefohn, A, Cates, J & Whitaker R 2003, ‘Interactive, GPU-based level sets for 3D brain tumor segmentation’ Proceedings of the sixth
international conference on medical image computing and computer-assisted intervention (MICCAI), ed. R. E. Ellis & T. M. Peters, MICCAI, Montreal, pp. 564-572.
56.Liu, J, Udupa, JK, Odhner, D, Hackney, D & Moonis, G 2005, ‘A system for brain tumor volume estimation via MR imaging and fuzzy connectedness’, Computerized Medical Imaging and Graphics, vol. 29, no. 1, pp. 21-34.
57.Liu, H, Yang, C, Pan, N., Song, E & Green, R 2010, ‘Denoising 3D MR images by the enhanced non-local means filter for Rician noise’, Magnetic Resonance Imaging, vol. 28, no. 10, pp. 1485-1496.
58.Lu, B, Deng, C, Liu, Q & Li, J 2009, ‘Four Order Adaptive PDE Method for MRI Denoising’, Proceedings of the third international conference on Bioinformatics and Biomedical Engineering, pp. 1-4.
59.Luo, J, Zhu, Y & Magnin, IE 2009, ‘Denoising by Averaging Reconstructed Images: Application to Magnetic Resonance Images’, IEEE Transactions on Biomedical Engineering, vol. 56, no. 3, pp. 666-674.
60.Luo, J, Zhu, Y & Hiba, B 2010, ‘Medical image denoising using one-dimensional singularity function model’, Computerized Medical Imaging and Graphics, vol. 34, no. 2, pp. 167-176.
181
61.Lysaker, M, Lundervold, A & Tai, XC 2003, ‘Noise removal using fourth-order partial differential equation with applications to medical magnetic resonance images in space and time’, IEEE Transactions on Image Processing, vol. 12, no. 12, pp. 1579–1590.
62.MacQueen, J 1967, ‘Some methods for classification and analysis of multivariate observations’, Proceedings of the fifth Berkeley symposium on Mathematics, Statistics and Probability, pp. 281-297.
63.Macovski, A 1996, ‘Noise in MRI’ Magnetic Resonance in Medicine, vol. 36, no. 3, pp. 494–497.64.Manjon, JV, Robles, M & Thacker, NA 2007, ‘Multispectral MRI de-noising using non-local means’, Proceedings of the Medical Image
Understanding and Analysis, pp. 41–46.65.Manjon, JV, Carbonell-Caballer, J, Lull, JJ, García-Marti, G, Martí-Bonmati, L & Robles, M 2008, ‘MRI denoising using non-local
means’, Medical Image Analysis, vol. 12, no. , pp. 514–523.66.Manjon, JV, Coupe, P, Marti-Bonmati, L, Collins, DL & Robles, M 2010, ‘Adaptive Non-Local Means Denoising of MR images with
spatially varying noise levels’, Magnetic Resonance Imaging, vol. 31, no. 1, pp. 192-203.67.Manjon, JV, Coupe, P, Buades, A, Collins, DL & Robles, M 2012, ‘New methods for MRI denoising based on sparseness and self-
similarity’, Medical Image Analysis, vol. 16, no. 1, pp. 18-27.68.Mathew, JM & Simon, P 2014, ‘Color texture image segmentation based on neutrosophic set and nonsubsmpled contourlet
transformation’, Proceedings of the first international conference on applied algorithms (ICAA), ed. P. Gupta 7 C. Zaroliagis, ICAA, Kolkata, pp. 164-173.
69.McGibney, G & Smith MR 1993, ‘Unbiased signal-to-noise ratio measure for magnetic resonance images’, Medical Physics, vol. 20, no. 4, pp. 1077–1078.
70.McVeigh, ER, Henkelman, RM & Bronskill, MJ 1985, ‘Noise and filtration in magnetic resonance imaging’, Medical Physics, vol. 12, no. 5, pp. 586–591.
71.Moon, N, Bullitt, E, Van Leemput, K & Gerig, G 2002, ‘Model based brain and tumor segmentation’, Proceedings of the sixteenth international conference on pattern recognition, Quebec, pp. 528-531.
72.Moonis, G, Liu, J, Udupa, JK & Hackney, DB 2002, ‘Estimation of tumor volume with fuzzy connectedness segmentation of MR images’, American Journal of Neuroradiology, vol. 23, no. 3, pp. 352-363.
73.Nowak, RD, 1999, ‘Wavelet-based Rician noise removal for magnetic resonance imaging’, IEEE Transaction on Image Processing, vol. 8, no. 10, pp. 1408–1419.
74.Pal, SK & King, RA 1980, ‘Image enhancement using fuzzy set’, Electronics Letter, vol. 16, no. 10, pp. 376-378. 75.Perona, P & Malik, J 1990, ‘Scale-space and edge detection using anisotropic diffusion’, IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 12, no. 7, pp. 629–639.78.Phillips, WE, Velthuizen, RP, Phuphanich, S, Hall, LO, Clark, MC & Silbiger, MS 1995, ‘Application of fuzzy c-means segmentation
technique for tissue differentiation in MR images of hemorrhagic giloblastoma multiframe’, Magnetic Resonance Imaging, vol. 13, no. 2, pp. 277-290.
182
79.Pizurica, A, Philips, W, Lemahieu, I & Acheroy, M 2003, ‘Versatile wavelet domain noise filtration technique for medical imaging’, IEEE Transactions on Medical Imaging, vol. 22, no. 3, pp. 323–31.
80.Placidi, G, Alecci, M & Sotgiu, A 2003, ‘Post-processing noise removal algorithm for magnetic resonance imaging based on edge detection and wavelet analysis’, Physics in Medicine and Biology, vol. 48, no. 13, pp. 1987-1995.
81.Prastawa, M, Bullitt, E, Ho, S & Gerig, G 2004, ‘A brain tumor segmentation framework based on outlier detection’, Medical Image Analysis, vol. 8, no. 3, pp. 275-283.
82.Prinosil, J, Smekal, Z & Bartusek, K 2010, ‘Wavelet Thresholding Techniques in MRI Domain’, Proceedings of the International Conference on Biosciences, pp. 58-63.
83.Pruessmann, KP, Weiger, M, Scheidegger, MB & Boesiger, P 1999, ‘SENSE: sensitivity encoding for fast MRI’, Magnetic Resonance in Medicine, vol. 42, no. 5, pp. 952–962.
84.Purcell, E, Torrey, H & Pound, R 1946, ‘Resonance Absorption by Nuclear Magnetic Moments in a Solid’, Physical Review, vol. 69, no. 1-2, pp. 37-38.
85.Rabi, II, Zacharias, JR, Millman, S & Kusch, P 1938, ‘A new method of measuring nuclear magnetic moment’, Physical Review , vol. 53, no. 4, pp. 318–327.
86.Rajan, J, Jeurissen, B, Sijbers, J & Kannan, K 2009, ‘Denoising Magnetic Resonance Images using Fourth Order Complex Diffusion’, Proceedings of the thirteenth international conference on Machine Vision and Image Processing, pp. 123-127.
87.Rajan, J, Poot, D, Juntu, J & Sijbers, J 2010, ‘Noise measurement from magnitude MRI using local estimates of variance and skewness’, Physics in Medicine and Biology, vol. 55, no. 16, pp. 441-449.
88.Rajan, J, Jeurissen, B, Verhoye, M, Audekerke, JV & Sijbers, J 2011, ‘Maximum likelihood estimation-based denoising of magnetic resonance images using restriced local neighborhoods’, Physics in Medicine and Biology, vol. 56, no. 16, pp. 5221-5234.
89.Rajan, J, Veraat, J, Audekerke, JV, Verhoye, M & Sijbers, J 2012, ‘Nonlocal maximum likelihood estimation method for denoising multiple-coil magnetic resonance images’, Magnetic Resonance Imaging, vol. 30, no. 10, pp. 1512-1518.
90.Redpath, TW 1998, ‘Signal-to-noise ratio in MRI’, The British Journal of Radiology, vol. 71, no. 847, pp. 704-707.91.Rodriguez, AO 2004, ‘Principles of magnetic resonance imaging’, Revista Mexicana De Fisica, vol. 50, no. 3, pp. 272-286.
92.Roemer, PB, Edelstein, WA., Hayes, CE, Souza, SP & Mueller, OM 1990, ‘The NMR phased array’, Magnetic Resonance in Medicine, vol. 16, no. 2, pp. 192–225.
93.Rudin, LI, Osher, S & Fatemi, E 1992, ‘Nonlinear total variation based noise removal algorithms’, Physica D: Nonlinear Phenomena, vol. 60, no. 1-4, pp. 259-268.
94.Rummeny, EJ, Reimer, P & Heindel, W (eds.) 2009, MR Imaging of the Body, Second Edition, Georg Thieme Verlag, Stuttgart, Germany. Avaliable from: Google books.
18395. Samarandache, F 2003, A unifying field in logics Neutrosophic logic, in Neutrosophy. Neutrosophic Set, Neutrosophic Probability, third ed., American Research Press, Rehoboth.
96. Samsonov, AA & Johnson CR 2004, ‘Noise-adaptive nonlinear diffusion filtering of MR images with spatially varying noise levels’, Magnetic Resonance in Medicine, vol. 52, no. 4, pp. 798–806.
97. Schoenberg, SO, Dietrich, O & Reiser, MF (eds.) 2007, Parallel Imaging in Clinical MR Applications, Springer-Verlag, Berling, Germany.
98. Sengur, A & Guo Y 2011, ‘Color texture image segmentation based on neutrosophic set and wavelet transformation’, Computer Vision and Image Understanding, vol. 115, no. 8, pp. 1134-1144.
99. Shan, J, Cheng, HD & Wang, Y 2012, ‘A novel segmentation method for breast ultrasound images based on neutrosophic I-means clustering’, Medical Physics, vol. 39, no. 9, pp. 5669-5682.
100.Shattuck, DW, Prasad, G, Mirza, M, Narr, KL & Toga, AW 2009, ‘Online resource for validation of brain segmentation methods’, Neuroimage, vol. 45, no. 2, pp. 431-439.
101.Yang, GZ, Burger, P, Firmin, DN & Underwood, SR 1995, ‘Structure Adaptive Anisotropic Filtering for Magnetic Resonance Image Enhancement’, Proceedings of the sixth international conference on Computer Analysis of Images and Patterns (CAIP), ed. V. Hlavac
& R. Sara, CAIP, Prague, vol. 970, pp. 384 –391.102.Yang, X & Fei, B 2011, ‘A wavelet Multiscale denoising algorithm for magnetic resonance images’, Measurement Science and
Technology, vol. 22, no. 2, pp. 1-12. 103.You, YL & Kaveh, M 2000, ‘Fourth-order partial differential equation for noise removal’, IEEE Transactions on Image Processing, vol.
9, no. 10, pp. 1723–1730. 104.Yul H & Zhao, L 2008, ‘An Efficient Denoising Procedure for Magnetic Resonance Imaging’, Proceedings of the second international
conference on Bioinformatics and Biomedical Engineering, pp. 2628-2630.105.Zadeh, LA 1975, ‘Fuzzy logic and approximate reasoning’, Synthese, vol. 30, no. 3-4, pp. 407-428.
106.Zaroubi, S & Goelman, G 2000, ‘Complex denoising of MR data via wavelet analysis: application for functional MRI’, Magnetic Resonance Imaging, vol. 18, no. 1, pp. 59–68.
107.Zhang, F & Ma, L 2010, ‘MRI Denoising Using the Anisotropic Coupled Diffusion Equations’, Proceedings of the third international conference on Biomedical Engineering and Informatics, pp. 397-401.
108.Zhang, M, Zhang, L & Cheng, HD 2010a, ‘A neutrosophic approach to image segmentation based on watershed method’, Signal Processing, vol. 90, no. 5, 1510-1517.
109.Zhang, M, Zhang, L & Cheng, HD 2010b, ‘Segmentation of ultrasound breast images based on a neutrosophic method’, Optical Engineering, vol. 49, no. 11, pp. 117001-12.
110.Zhu, H, Li, Y, Ibrahim, JG, Shi, X, An, H, Chen, Y, Gao, W, Lin, W, Rowe, DB & Peterson, BS 2009, ‘Regression models for identifying noise sources in magnetic resonance imaging’, Journal of the American Statistical Association, vol. 104, no. 486, pp. 623-
637.
List of Publications184
1. Mohan, J, Krishnaveni, V & Yanhui Guo 2014, ‘A survey on the magnetic resonance image denoising methods’, Elsevier-Biomedical Signal Processing and Control, vol. 9, pp. 59-69. Impact Factor: 1.532
2. Mohan, J, Krishnaveni, V & Yanhui Guo 2013, ‘MRI denoising using non local neutrosophic set approach of wiener filtering’, Elsevier-Biomedical Signal Processing and Control, vol. 8, no.6, pp. 779-791. Impact Factor: 1.532
3. Mohan, J, Krishnaveni, V & Yanhui Guo 2013, ‘ A new neutrosophic approach of Wiener filtering for MRI denoising’, Measurement Science Review, vol. 13, no.4, pp. 177-186. Impact Factor: 1.233
4. Mohan, J, Krishnaveni, V & Yanhui Guo 2012, ‘Performance comparison of MRI denoising techniques based on neutrosophic set approach’, European Journal of Scientific Research, vol. 86, no.3, pp. 307-318.
5. Mohan, J, Thilaga Shri Chandra, AP, Krishnaveni, V & Yanhui Guo 2012, ‘Evaluation of Neutrosophic set approach filtering technique for image denoising’, International Journal of Multimedia & its applications, vol. 4, no. 4, pp. 73-81.
6. Mohan, J, Krishnaveni, V & Yanhui Guo 2012, ‘Performance Analysis of Neutrosophic set approach of median filtering for MRI denoising’, International Journal of Elec. and Commn. Engg & Tech., vol. 3, no. 2, pp. 148-163.
International Journals
185
1. Mohan, J, Krishnaveni, V, Yanhui Guo & Kanchana Jeganathan 2012, ‘MRI Denoising based on Neutrosophic Wiener filtering’, in Proceedings of the IEEE International Conference on Imaging Systems and Techniques University of Manchester, Manchester, UK, pp. 327-331.
2. Mohan, J, Krishnaveni, V & Yanhui Guo 2012, ‘Validating the Neutrosophic Approach of MRI Denoising based on Structural Similarity’, in Proceedings of the IET Image Processing Conference, University of Westminster, London, UK, pp. 1-6.
3. Mohan, J, Thilaga Shri Chandra, AP, Krishnaveni, V & Yanhui Guo 2012, ‘Image Denoising based on Neutrosophic Wiener Filtering’ in Proceedings of the Advances in Computing& Inform. Technology, AISC, ed. N. Meghanathan et al, Springer-Verlag Berlin Heidelberg, vol. 177, pp. 861-869.
4. Mohan, J, Krishnaveni, V & Yanhui Guo 2011, ‘A Neutrosophic Approach of MRI Denoising’, in Proceedings of the IEEE International Conference on Image Information Processing, Simla, India, pp. 1-6.
International Conferences
CHALLENGES IN BIOMEDICAL IMAGE PROCESSING
• In todays health care, imaging plays an important role throughout the entire clinical process from diagnostics and treatment planning to surgical procedures.
• Since most imaging modalities have gone directly digital, with continually increasing resolution, medical image processing has to face the challenges arising from large data volumes.
186
Analysis of Images to Detect Abnormalities in Endoscopy
* The increase incidence and burden of gastroenterological diseases, challenges both the surveillance strategies and resources for monitoring at risk patients and the possibility of early detection and accurate staging of lesions and tissue alterations.
* The need is to relieve the clinicians and health system of part of the bottlenecks commonly found in endoscopic surveillance, and to reduce the number of unnecessary biopsy taken from the patients, has asked for an ever increasing effort from the imaging community to provide to the endoscopists tools to the able to identify suspicious areas, suggest bioptic sites, perform a virtual histology, or estimate the tissue diagnosis.
187
Lung Cancer Lung cancer is the leading cause of cancer-related death worldwide.
Screening high risk individuals for lung cancer with low-dose CT scans is now being implemented in the United States and other countries are expected to follow soon.
In CT lung cancer screening, many millions of CT scans will have to be analyzed, which is an enormous burden for radiologists. Therefore there is a lot of interest to develop computer algorithms to optimize screening.
188
Digital Mammography
The Digital Mammography DREAM Challenge will attempt to improve the predictive accuracy of digital mammography for the early detection of breast cancer.
Primary benefit of this Challenge will be to establish new quantitative tools - machine learning, deep learning or other - that can help decrease the recall rate of screening mammography, with a potential impact on shifting the balance of routine breast cancer screening towards more benefit and less harm.
189
PET segmentation challenge using a data management and processing infrastructure
Automated segmentation of PET images for the delineation of tumor volumes has been the focus of intense research efforts for the last few years .
There has also been a few limited efforts to compare several methods on common datasets
But in the majority of cases, each method has been evaluated on different image sets according to different evaluation criteria, and a comparison of currently available methods based on literature analysis only is thus challenging.
190
Spine Imaging Researchers are encouraged to evaluate novel and existing methods on important, spine related image analysis tasks.
Left atrial wall thicknessFocuses on algorithms for segmentation of left atrial wall thickness from MRI and CT
191
Identify nerve structures in ultrasound images of the neck
Building a model so that we can identify nerve structures in a dataset of ultrasound images of the neck. Doing so would improve catheter placement and contribute to a more pain free future.
192
193
Some More Interesting Some More Interesting Research Topics in Digital Research Topics in Digital
Signal ProcessingSignal Processing
194
Source Coding and Compression
• Source coding and compression deal with reducing the bit rate required to store and communicate complex signals and images with minimal distortion.
• Lossy compression and lossless compression can be investigated for networked systems of signal, image and video sources.
195
Computational Imaging and Inverse Problems
• The latest medical imaging, radar imaging and microscopy systems rely heavily on the availability of fast computational signal and image processing algorithms.
• These algorithms include image reconstruction, image restoration, multimodality image registration, tracking, detection and classification.
196
Compressive Sampling and Sensing• CS is an emerging theory which permits radically
new sensing devices that simultaneously acquire and compress certain signals using very efficient randomized sensing protocols.
• The implications of the CS theory are very far-reaching and will likely impact analog-to-digital conversion, data compression, medical imaging, sensor networks, digital communication, statistical model selection, and more.
Tracking Rapid Nonlinear Dynamic Changes in both Time and Frequency
A completely new class of wavelet based algorithms can be developed to track very rapid changes in nonlinear dynamic models and to map these into nonlinear frequency response behaviours with applications to EEG and neuro imaging.
197
Image Signal Processing
• Content based image retrieval• Object detection and tracking• Face detection• Traditional vehicle plate detection and
recognition• Text detection and recognition• Optical character recognizer• Steganography
198
Speech Signal ProcessingSpeech Signal Processing
* Speech and Speaker Recognition
* Speech Enhancement and Transmission
* Speech Analysis
* Natural Language Processing
Natural language processing (NLP) is the ability of a
computer program to understand human speech as it is
spoken. 199
Software Defined RadioSoftware Defined Radio
200
• Analysis of signal processing algorithms used for software defined radios.
• Computer architectures for software defined radios.
201
Detailed analysis of different tracking and navigation application including: aircraft positioning, target tracking for radar and sonar applications, car collision detection, and positioning and tracking in homeland security applications.
Define the requirements for each application such as sampling rate, accuracy, latency, range.
Tracking ApplicationsTracking Applications
202
Definition:“A radar adapts intelligently to its environments on the basis of a plurality of potential information sources”Benefits:Adaptive Radar Scheduler/Data Product GenerationAdaptive Transmit/Receive ChainEnhanced Real-Time AdaptivityOpen Topics:Adaptive Power AllocationDigitization and Processing of RF and Front-EndDiversity Technology (Time/Frequency/Spatial/Embedded Domain )High-Performance Computing Platform and Programming ModelKnowledge-Aided Processing and LearningEnvironmental Dynamic Database and Data Mining
SP Applications in Reconfigurable/Cognitive Radar
203
Signal Processing for Sensing and Sensor Networks
• Millions of sensors collect environmental data from the ground, oceans, and space but only a small minority are networked and integrated into a globally accessible distributed database.
• There are many challenges that must be overcome to in order to build such networks.
• Analysis of signal processing algorithms used for wireless sensor networks: positioning, tracking, data fusion, sensor processing. Analysis of DSP architectures used in sensor networks.
204
Number of Connected Objects Expected to Reach 50bn by 2020
Internet of Things
205
Definition:All devices and places are universally IP enabled and an integral part of the Internet .These devices are called smart objects whose examples are mobile phones, personal health devices and home automation, to industrial automation, smart metering and environmental monitoring systems.Benefits:Better Environmental Monitoring, Energy Savings, Smart Grids, MoreEfficient Factories, Better Logistics, Better Healthcare and Smart Homes.SP Related Open Topics:Ubiquitous Information/Signal/Data Capturing/SensorWireless Embedded Technology (Proprietary networking solutions)RFID: Circuits and IntegrationInformation/Signal/Data Coding and CompressionSecurity Authentication, Key Management and Routing Algorithms
SP in Internet of Things
206