OPTIMAL Presentation Template · Image Preprocessing Step 3: Image Reorientation and Resampling The loaded images must be reoriented and resampled to the same orientation and size

IDEA

Department of Radiology and BRIC, UNC-Chapel Hill

iBEAT Tutorial

Version 1.1, Feb 10, 2012


Navigation

• About iBEAT

• Install iBEAT

• Main Window

• Image Preprocessing

• Brain Extraction

• Tissue Segmentation

• Brain Labeling

• Others

• Acknowledgement


About iBEAT

iBEAT is an Infant Brain Extraction and Analysis

Toolbox with graphical user interfaces for processing infant

brain MR images.

The iBEAT is developed by the IDEA group at the

University of North Carolina at Chapel Hill, directed by Dr.

Dinggang Shen ([email protected]).


About iBEAT

The Processing Pipeline of

Infant Brain MR Images

Infant Brain

MR Images

Brain

Images

Tissue

Images

Preprocessed

Images

Infant

Brain

Labels

Disk Disk Disk Disk Disk

Processing

Images


About iBEAT

Overview of the Processing Modules

1. Image Preprocessing: Reorient original infant T1/T2/FA images interactively to the same orientation as a selected template

image. Resampling (to the same size as the template) is applied to the reoriented images

automatically. N3 correction is then performed on the reoriented and resampled images.

2. Brain Extraction: Strip the skull and remove the cerebellum from the preprocessed image automatically. The

automatically extracted brain image can then be edited manually to obtain a more accurate brain

image (optional).

3. Tissue Segmentation: Segment gray matter, white matter and CSF from the extracted brain images.

Non-longitudinal T1/T2/FA images (single time point) are processed by coupled segmentation and

a single tissue image is generated.

Longitudinal T1/T2/FA images (multiple time points) are segmented together and a longitudinally

segmented tissue image is generated at each time point.

4. Brain Labeling: Label the segmented brain tissues with assistance of the atlas of a selected template (HAMMER

registration is used).


About iBEAT

Image modalities – T1, T2, FA

File format – analyze (.hdr with .img)

Subject image files should be arranged as follows:

The folder including images of all time points for a subject.

No blank space or ‘-’ is allowed in the subject name.

The folder including images at a time point for a subject. The folder

name consists of the subject name and month (connected with ‘-’)

in each folder

(each time point) subject name,

month and

modality

NOTE:

T2 is required for month 0 ~ 5

T1 is required for month >= 6

FA is optional for each time point

Arrangement of Subject Images

The folder including subject images

NOTE: month

must be integer.

Month 0 indicates

0~30 days


iBEAT Installation

• Linux operating system is required

• Installation steps: Download the ibeat package and unzip the ibeat.zip (e.g., the

package is unzipped in /home/programs/ibeat).

Setup environment for the ibeat as follows:

Edit the shell resource file in the home directory of the user (cd ~) :

a) Add the following two lines in the .cshrc file if csh/tcsh is used

setenv IBEAT_HOME /home/programs/ibeat

source $IBEAT_HOME/ibeatEnv.csh

b) Add the following two lines in the .bashrc file if bash is used

export IBEAT_HOME=/home/programs/ibeat

source $IBEAT_HOME/ibeatEnv.bash

Restart the shell to update the environment.

Use command ibeat to start the software.


Main Window


Main Window


Image Preprocessing

Functionality of the Image Preprocessing Module: Reorient original images interactively to the same orientation as the template image.

Resampling (to the same size as the template) is applied to the reoriented images

automatically. N3 correction is then performed on the reoriented and resampled images.

Image Preprocessing Module


Image Preprocessing

Functionality of the Image Preprocessing Module:

It consists of four parts:

Processing Flow: integrates

functions in the pipeline of

the image reorientation.

Subject Images: lists all

loaded images and the

processed images

Image View: windows for

interactive inspection of the

images.

Processing Status: show

hints for image reorientation

steps and data processing

status.



Image Preprocessing

Step 1: Image Loading

①

Image Loading Module ② ③

④ select the subject folder

subject images will be

loaded automatically.

⑤


Image Preprocessing

Functions in the Image Loading Module:

1. Load images of all subjects: select ‘Automatic’ in A –> D –> select the folder including all the subjects.

2. Add all images of a subject: select ‘Subject j’ in A and ‘Automatic’ in B –> D –> select the j-th subject folder.

3. Add all images at a time point for a subject: select ‘Subject j’ in A, ‘Month i’ in B and ‘Automatic’ in C –> D –> select the

i-th time point folder in the j-th subject folder.

4. Add an image at a time point for a subject: select ‘Subject j’ in A, ‘Month i’ in B and ‘T1,T2 or FA’ in C –> D –> select the

T1, T2 or FA image in the i-th time point folder in the j-th subject folder.

A B C D 5. Preview an image: left mouse click the image name to display three

orthogonal slices of the image.

NOTE: the displayed slices may not be consistent with the

‘Axial’, ‘Coronal’ and ‘Sagittal’ statement before reorientation.

6. Delete a selected (left mouse click) image: right mouse click in the

image list window and press the ‘Delete Images’ popup menu.

7. Delete images at a time point: left mouse click a time point (e.g.,

Month 6) –> right mouse click and press the ‘Delete Images’ popup

menu.

8. Delete all images of a subject: left mouse click a subject (e.g.,

Subject 1) –> right mouse click and press the ‘Delete Images’ popup

menu.

9. Delete selected images: use ‘Shift’ or ‘Ctrl’ key and left mouse

button to select images –> right mouse click and press the ‘Delete

Images’ popup menu.

Image List


Image Preprocessing

Step 2: Interactive Image Inspection After all longitudinal images are loaded, press ‘Finish’ to return to the Image Preprocessing Module. The loaded images

will be in the ‘Subject Images’ list.

Left mouse click an image in the ‘Subject Images’ list to display three orthogonal 2D slices and 3D slices of the image.

Right mouse click in the ‘Subject Images’ list to view the property of the selected image, or delete the image.

Review slices: mouse wheel over the 2D slices (in the ‘Axial’, ‘Coronal’ and ‘Sagittal’ windows), or use the slider.

Zoom in/out 2D slices: right mouse button down (over the 2D slices) and move up/down.

Pan 2D slices: left mouse button down (over the 2D slices) and drag.

Cross: left mouse button down (over the 2D slices) Image Preprocessing Module


Image Preprocessing

Step 2: Interactive Image Inspection (3D slices) Rotate 3D slices: press the toggled button ‘R’, then left mouse button down upon the 3D slices and move the mouse

Zoom 3D slices: press the toggled button ‘Z’, then left mouse button down upon the 3D slices and move up/down

Pan 3D slices: press the toggled button ‘P’, then left mouse button down upon the 3D slices and move the mouse

NOTE: Click the toggled button ‘R’, ‘Z’ and ‘P’ to enable or disable the related manipulation functions.


Rotate

Zoom

Pan


Image Preprocessing

Step 3: Image Reorientation and Resampling The loaded images must be reoriented and resampled to the same orientation and size as the selected template. The

template is used in the preprocessing and brain labeling steps. The standard orientation used in iBEAT is shown in the

left image and all the standard templates have the same volume size (256x256x198) and voxel size (1x1x1).

Reorient and resample the loaded images by: ③ (determine the loaded images to be reoriented and resampled) -> ④ (set

reorientation parameters, resampling is applied automatically) -> ⑤ (reorient and resample the related loaded images)

The processed images will be saved as analyze format and stored with the original image files.

① ②

③

④

⑤



Image Preprocessing

Step 3: Image Reorientation and Resampling

Preview reorientation and resampling result of a selected image: choose ‘Selected’ option in ① -> select a loaded image in

② (subject images list) -> set reorientation parameters in ③ -> ④.

Reorient and resample multiple loaded images: choose ‘Selected’ option in ① -> select multiple loaded images (use

Ctrl/Shift and mouse) in ② -> set reorientation parameters in ③ -> ④.

Reorient and resample T1/T2/FA/All loaded images: choose ‘T1/T2/FA/All’ option in ① -> set reorientation parameters in ③

-> ④

①

③

④


②


Image Preprocessing

How to set the reorientation parameters ? ( Method 1 ) Step 1: review all non-flip reorientation options by: ① (select a loaded image) -> ② (right click and

choose ‘Review Reorientation Options ’ ). The selected image will be reoriented with all non-flip

reorientation parameters, and the reoriented images with the parameters will be displayed.

Step 2: review (use ③) and determine the correct reorientation parameters by comparing the reoriented

images with the template image visually.

Step 3: select the correct reorientation parameters (see ④), the parameters at ⑤ will be updated automatically.

① ②

③


④

⑤


Image Preprocessing

How to set the reorientation parameters ? ( Method 2 )

Step 1: find the sagittal, coronal and axial correspondences. e.g. preliminarily (xT, yT, zT) = (zo, xo, yo)

Step 2: preview the reorientation result with the preliminary reorientation parameters.

Template Image A Selected Image

Template (xT, yT, zT) Original (xo, yo, zo)

z

y

x

z x

y


Image Preprocessing

How to set the reorientation parameters ? ( Method 2 ) Step 3: check if the reoriented image is consistent with the template image (may flip).

Step 4: set the correct reorientation parameters to reorient selected/T1/T2/FA/All images .

Template (xT, yT, zT) Original (xo, yo, zo)

Template Image Reoriented Image

z

y

x

z x

y

The Original Image

z x

y

(xR, yR, zR) = (zo, xo, yo)

Reoriented (xR, yR, zR)

(xT, yT, zT) = (-xR, -yR, zR)

Anterior-Posterior Flip (in the y axis), and obey the right-hand rule)

(xT, yT, zT) = (-zo, -xo, yo)


Step 4: N3 Correction After all loaded images are reoriented and resampled to the same orientation and size as the selected template,

N3 correction can be performed on the processed images by ①.

NOTE: a processed image (M001-0-T1-reoriented) is named by appending ‘-reoriented’ to the name of the

original un-processed image (M001-0-T1).

Image Preprocessing

①



Step 5: Start the Brain Extraction Module After all loaded images are preprocessed (reoriented, resampled, N3-corrected) in the Image Preprocessing

Module, press ‘Brain Extraction’ to start the Brain Extraction Module.

All the preprocessed images will be transferred in the ‘Subject Images’ list of the Brain Extraction Module

automatically, and the Image Preprocessing Module will be closed automatically.

Brain Extraction Module

Image Preprocessing


Brain Extraction

Step 1: Start the Brain Extraction Module Suppose preprocessed images are obtained by the Image Preprocessing Module, then the Brain Extraction Module can

also be started from the main window.

The preprocessed images (whose file names are ended with ‘-reoriented’, can be found where the original un-reoriented

images are) can be loaded by ② (the Image Loading Module will be started).

①

② Brain Extraction Module


Brain Extraction

Functionality of the Brain Extraction Module: To strip the skull and remove the cerebellum from the preprocessed image automatically. In addition, to edit the

automatically extracted brain image manually to obtain a more accurate brain image (optional).

The module structure is similar to the structure of the Image Preprocessing Module. Please refer to the image

preprocessing for functions such as interactive image inspection.



Brain Extraction

Step 2: Extract Brain Images Firstly choose ‘Selected/T1/T2/FA/All’ option by ①, then set parameters for brain extraction by ②, finally extract brains

from the selected loaded images (preprocessed) by ③. Generally, the default parameters can be used. The extracted brain

images will be saved as analyze format and stored with the original image files as well.

If the loaded images are skull-stripped, please check the ‘Cerebellum Removal’ only to remove the cerebellums without

skull stripping.

①

②

③



Brain Extraction

Step 3: Manually Edit the Extracted Brain Images Select the automatically extracted brain image –> right mouse click and select the ‘Manual Editing’ item on the popup

menu.

An overlay (red mask) from the automatically extracted brain image will be displayed with the un-extracted image

(preprocessed image) in the Manual Image Editing Module.

Manual Image Editing Module


Brain Extraction

Step 3: Manually Edit the Extracted Brain Images 1. Click A (the ‘Cross/Zoom/Pan/Slice’ toggled button) and then interactive inspect the image:

Review slices: mouse wheel over the 2D slices (in the ‘Axial’, ‘Coronal’ and ‘Sagittal’ windows), or use the slider.

Zoom in/out 2D slices: right mouse button down (over the 2D slices) and move up/down.

Pan 2D slices: left mouse button down (over the 2D slices) and drag.

Cross: left mouse button down (over the 2D slices).

Manual Image Editing Module

A

B C

D

E

2. Click B/C (the ‘Painter’/‘Eraser’ toggled

button) and then paint/erase regions of

interest to/from the overlay interactively.

Painter/Eraser size: use the radius slider

or just input the value (the radius size will

be indicated in the slice windows, see the

red circle under the cursor).

3D/2D Mask: select 3D/2D Mask to make

the painting or erasing operation effective

in 3D volume/2D slice space.

3. Overlay:

Click ‘Default’, ‘Clear’ and ‘Undo’ in D to

load default overlay, clear the current

overlay and undo the last painting (or

erasing) operation, respectively.

4. Preview:

Click E to preview the manually edited

brain image obtained with the current

overlay.

Click ‘Finish’ when the manually edited brain

image is satisfactory.


Brain Extraction

Step 4: Start the Tissue Segmentation Module After all brain images are extracted (cerebellum removal and manual image editing may be applied) from the loaded

images (preprocessed) in the Brain Extraction Module, press ‘Tissue Segmentation’ to start the Tissue Segmentation

Module.

All the extracted brain images will be transferred in the ‘Subject Images’ list of the Tissue Segmentation Module

automatically, and the Brain Extraction Module will be closed automatically. NOTE: an extracted brain image (M001-0-T1-

reoriented-strip) is named by appending ‘-strip’ to the name of the preprocessed image (M001-0-T1-reoriented).

Tissue Segmentation Module


Tissue Segmentation

Step 1: Start the Tissue Segmentation Module Suppose extracted brain images are obtained by the Brain Extraction Module, then the Tissue Segmentation Module can

also be started from the main window.

The extracted brain images (whose file names are ended with ‘-strip’, can be found where the original un-preprocessed and

preprocessed images are) can be loaded by ② (the Image Loading Module will be started).

②

①



Tissue Segmentation

Functionality of the Tissue Segmentation Module: To perform longitudinal and coupled tissue segmentation across multiple time points and multiple modalities

(T1/T2/FA) for each subject. Images of multiple modalities (T1/T2/FA) at each time point can be aligned automatically

before longitudinal tissue segmentation (skip the alignment if the brain images have been aligned).

The module structure is similar to the structure of the Image Reorientation Module. Please refer to the image preprocessing

for functions such as interactive image inspection.



Tissue Segmentation

Step 2: Segment Tissues and Start the Brain Labeling Module All the loaded images (extracted brain) can be processed by ①. The segmented tissue images will be saved as analyze

format and stored with the original image files as well. Note: only one coupled tissue image will be generated at

each time point. A segmented tissue image is named as, e.g., M001-0-reoriented-strip-seg, which has a ‘-seg’ postfix

but without modality information.

After all segmented tissue images are obtained in the Tissue Segmentation Module, press ② ( ‘Brain Labeling’ ) to start

the Brain Labeling Module. All the segmented tissue images will be transferred to the Brain Labeling Module.

①

②



Brain Labeling

Step 1: Start the Brain Labeling Module Suppose segmented tissue images are obtained by the Tissue Segmentation Module, then the Brain Labeling Module

can also be started from the main window.

The segmented tissue images (whose file names are ended with ‘-seg’, can be found where the original un-preprocessed,

preprocessed and extracted brain images are) can be loaded by ② (the Image Loading Module will be started).

②

①

Brain Labeling Module


Brain Labeling

Step 2: Label Brain Images A template can be selected for brain labeling. NOTE: All standard templates used in iBEAT are in the template folder of

the package. Please make the orientation, image size and file format of a custom template the same as the ones of

the standard templates, and name the custom template similarly (refer to the template files).

Firstly choose ‘Selected/All’ option by ③, then set parameters (the default ones can be used generally) for hammer

registration by ④, finally label brains for the selected loaded images (segmented tissue images) by ⑤. The brain labels will

be saved as analyze format and stored with the original image files as well (ended with ‘-aal’).

Brain Labeling Module

②

③

④

⑤

①


Brain Labeling

Brain Labels

The anatomical description of regions in ‘-aal’ image is detailed in the following table. The definition is originally from

“ N. Tzourio-Mazoyer et al, Neuroimage, 15: 273-289, 2002 ” , but now it is warped into infant spaces.


Others

Image file name regulation of the iBEAT software: subject-month-modality ->

subject-month-modality-reoriented ->

subject-month-modality-reoriented-strip ->

subject-month-reoriented-strip-seg ->

subject-month-reoriented-strip-seg-aal

The tool for the display of an image: • To select and view an image of analyze format.

The tool for interactive editing of the brain image: • To select a preprocessed image file (reoriented and resampled) and edit

the brain interactively.

• If the automatically extracted brain image is available in the same folder, it

will be detected and used as the default brain overlay.

• If the automatically extracted brain image is not available, brain extraction

can be performed firstly.

• The interactively edited brain image will be saved with the preprocessed

image and named with a postfix ‘-strip’.

Sample Data: • M001 (month 0, 6, 9, 25) in the sample folder of the iBEAT package


Others

Programming languages: • The graphical user interfaces and overall framework of the iBEAT software were implemented in

MATLAB. The image processing functions were implemented with the combination of C/C++, MATLAB,

Perl and Shell languages.

Parallel processing: • Eight separate and parallel threads are used for image processing. In the preprocessing, brain

extraction and brain labeling steps, all the images are processed by the separate threads in parallel. In

the longitudinal tissue segmentation step, all time points are processed by separate threads in parallel.

The parallel strategy accelerates the image processing largely.

Image processing efficiency: The processing speed is mainly dependent on the number of subjects, number of time points, number of

images, number of processor cores of the computer and the computing performance of each processor

core. Provide the number of subjects is numSubject, number of images is numImage, size of each image is

256x256x198, and the number of processor cores is numCore (<=8). The rough estimated processing time

is as follows:

1. Image reorientation and resampling: 1 x numImage x scale (second)

2. N3 correction: 1.7 x scale (minute)

3. Skull stripping and cerebellum removal: 17 x scale (minute)

4. Longitudinal tissue segmentation: 2.7 x scale (hour)

5. Brain labeling: 25 x scale (minute)

where scale = ceil(numImage/numCore) for 1, 2, 3, 5 and scale = ceil(numSubject/numCore) for 4. ceil(x) is

the nearest larger integer of x.


Acknowledgement

• iBEAT is developed by the IDEA group at the University of North Carolina at Chapel Hill. Dinggang Shen

initiated the project idea and direct the development of the software. Yakang Dai, Feng Shi, Li Wang

implemented functions and wrote the codes.

• The development of the software is partially supported by NIH grants EB006733, EB008374, MH088520 and

EB008760 to Dinggang Shen.

• The brain extraction function used the method proposed by Shi et al [1].

• The tissue segmentation function used the method proposed by Wang et al [2,3].

• The brain labeling used the HAMMER registration method proposed by Shen et al [4].

• Portions of functions of the iBEAT software were implemented based on the FSL library developed by the

Analysis Group, FMRIB, Oxford, UK, the MINC package developed by the McConnell Brain Imaging Centre

of the Montreal Neurological Institute, McGill University, and the ITK toolkit from the Kitware Inc.

• [1] Feng Shi, Li Wang, John H. Gilmore, Weili Lin and Dinggang Shen. Learning-based Meta-Algorithm for MRI Brain

Extraction. MICCAI 2011, Toronto, Canada, Sep. 18-22, 2011.

• [2] Li Wang, Feng Shi, Pew Thian Yap, John H Gilmore, Weili Lin and Dinggang Shen. Accurate and Consistent 4D

Segmentation of Serial Infant Brain MR Images. Multimodal Brain Image Analysis (MBIA 2011), Toronto, Canada, Sep. 18,

2011.

• [3] Li Wang, Feng Shi, Pew-Thian Yap, Weili Lin, John H. Gilmore, Dinggang Shen. Longitudinally guided level sets for

consistent tissue segmentation of neonates. Hum Brain Mapp. 2011.

• [4] Dinggang Shen and Christos Davatzikos. HAMMER: Hierarchical Attribute Matching Mechanism for Elastic Registration.

IEEE Transactions on Medical Imaging, 21(11):1421-1439, 2002.

Documents

OPTIMAL Presentation Template · Image Preprocessing Step 3: Image Reorientation and Resampling The loaded images must be reoriented and resampled to the same orientation and size