19
LOGO Fuzzy Application for Melanoma Cancer Risk Management Joint Research: Bilqis Amaliah (ITS) and Rahmat Widyanto (UI) 1 SocDic2011

Fuzzy Application for Melanoma Cancer Risk Management

  • Upload
    eugene

  • View
    26

  • Download
    0

Embed Size (px)

DESCRIPTION

Joint Research : Bilqis Amaliah (ITS) and Rahmat Widyanto (UI). Fuzzy Application for Melanoma Cancer Risk Management. Contents. Introduction. Problem Formulation. Goal. Problem Restriction. Method. Testing. Result. Conclusion. S uggestion and Recommendation. Background. - PowerPoint PPT Presentation

Citation preview

PowerPoint Template

Fuzzy Application for Melanoma Cancer Risk ManagementJoint Research:Bilqis Amaliah (ITS) and Rahmat Widyanto (UI)

1

SocDic2011LOGO

LOGO1ContentsTestingMethodGoalProblem FormulationIntroductionConclusionResultProblem RestrictionSuggestion and Recommendation

2SocDic2011

LOGO2Melanoma is one of skin cancer and deadly dangerousBackgroundEarly detection is necessary for the patient to get the right treatmentTakagi - Sugeno Fuzzy Inference System (TS-FIS) has a simpler computing with better accuracy than existing methods (SVM, Boosting SVM, Voted Perceptron)

3SocDic2011

LOGO3How to classify melanoma image using ABC feature and Takagi-Sugeno FIS ?Problem Formulation

Is Takagi-Sugeno FIS accuracy better than existing methods (SVM, Boosting SVM, Voted Perceptron) ?

4SocDic2011

LOGO4 Designing the image diagnosis system for determine whether melanoma or notGoal

5SocDic2011

LOGO5Image data must have a good resolution and not too small.Problem Restriction

The image is not covered by thick hair.

There is only one object in the image.

The system is built using MATLAB R2010.

6SocDic2011

LOGO6AsymmetryBorder IrregularityColor Variation3FeatureextractionMedian Filteringimage intensity values Mapping ThresholdingFlood - FillingMethod1Preprocessing2Segmentation7

6Prediction4Training5Takagi-SugenoFISSocDic2011

LOGO7Image Processing

[1] Input Image

[2] Median Filter Image[3] Grayscale Image[4] Contrasted Image[8] Result Image[7] Filled Image[6] Inverted BW Image[5] Black and White Image

8SocDic2011

LOGO8

789456Feature Extraction9123Asymmetry Asymmetry Index (AI)Lengthening Index ( )Color Variation Color homogeneity (Ch)Correlation geometry and photometry (Cpg)Border IrregularityCompactness Index (CI)Fractal Dimension (fd)Edge Abruptness (Cr)Pigmentation Transition (me, ve)

SocDic2011

LOGO9TS FIS Membership Function10

A M : [-0.2395 0.03768 0.3149] N : [0.03768 0.3149 0.592] B M : [-0.5161 0.2084 0.9329] N : [0.2084 0.9329 1.657] C M : [-58.15 0.8496 59.85] N : [0.8496 59.85 118.9] D M : [-25.59 -13.27 -0.954] N : [-13.27 -0.954 11.36] E M : [-0.1489 0.002281 0.1534] N : [0.002281 0.1534 0.3046] F M : [-108.3 -42.89 22.5] N : [-42.89 22.5 87.89] G M : [0.02139 8.275e+004 1.655e+005] N : [-8.275e+004 0.02139 8.275e+004] H M : [-253 0 253] N : [0 253 506] F M : [-0.00125 9.6 19.2] N : [-9.602 -0.00125 9.6]SocDic2011

LOGO-yang dicontohkan diatas adalah MF fitur A-Parameter batas (melanoma M dan non melanoma N) pada membership function fitur A sampai Z didapatkan melalui training data hasil ekstraksi fitur (pada proses sebelumnya)-data yang digunakan untuk training tersebut adalah data yang sudah BENAR-

1011

TS FIS RuleIf (A is (M/N) and (B is (M/N) and and (I is (M/N) then (output is (M/N) 512 rule (2^9)Because there is no special weighting on 9 features, then :

If (Melanoma) > (Non Melanoma) then output is Melanoma And otherwise -

SocDic2011

LOGO11TestingTrial Data200 DATA100 Melanoma Image(+)100 Non-Melanoma Image (-)Digit 8 - 9 : Color VariationFeature Vector DimensionDigit 1-2 : AsymmetryDigit 3 - 7 : Border Irregularity12SocDic201112Testing (cont)ExperimentPerformanceUsing 100 data of melanoma and 100 data of Non-MelanomaPerformance is measured using Accuracy13SocDic201113Testing (cont)ABC Feature ExtractionSegmentationPreprocessingOutput of PreprocessingInput Image

14Segmented Image

SocDic2011

LOGO14Testing (cont)Conclusion whether the imageis a melanoma or notTesting of Takagi-Sugeno FISTraining of Takagi-Sugeno FISABC Feature ExtractionTraining using 9 feature15

Segmented Image1 2 3 4 5 6 7 8 9

If ( ) then (output)SocDic2011

LOGO15TS-FIS performance comparison with Voted Perceptron, SVM, and SVM boosting16Classification MethodAccuracy (%)Takagi-Sugeno FIS82,5Voted Perceptron77,5SVM74,4SVMboosting75,2

SocDic2011

LOGO16Conclusion2Accuracy of TS-FIS is higher by 5% if compared to the Voted Perceptron, 8.1% higher when compared with SVM, and 7.3% higher when compared with SVMboosting.1image of melanoma can be classified based on ABC features, which is trained using Takagi-Sugeno Fuzzy Inference System17

SocDic2011

LOGO17Suggestion and RecommendationRequired the addition of trial data and feature selection on the development in order to improve performance.Improvement of segmentation by using another method18

SocDic2011

LOGO18Thank you19

KiitosSocDic2011H.NobuharaR. WidyantoSpecial Thanks :LOGO Add your company slogan

19