MRI Segmentation MIDAS, 2007, 2010

Similar documents
MIDAS Processing of Segmentation Data for Brain Lesions

MIDAS Signal Calibration

ASAP_2.0 (Automatic Software for ASL Processing) USER S MANUAL

EMSegmenter Tutorial (Advanced Mode)

Learning to Identify Fuzzy Regions in Magnetic Resonance Images

PRANA. Project Review and Analysis. March (Pre-release and still under development!)

An Introduction To Automatic Tissue Classification Of Brain MRI. Colm Elliott Mar 2014

ZipStudy. Utility for Packaging Study Files. A. Maudsley, 6/2015

Norbert Schuff VA Medical Center and UCSF

MR IMAGE SEGMENTATION

EMSegment Tutorial. How to Define and Fine-Tune Automatic Brain Compartment Segmentation and the Detection of White Matter Hyperintensities

Supplementary methods

mritc: A Package for MRI Tissue Classification

Detecting White Matter Lesions in Lupus

Knowledge-Based Segmentation of Brain MRI Scans Using the Insight Toolkit

Methods for data preprocessing

GLM for fmri data analysis Lab Exercise 1

Math in image processing

Detecting White Matter Lesions in Lupus

Automated MR Image Analysis Pipelines

Processing math: 100% Intensity Normalization

Kernel Based Fuzzy Ant Clustering with Partition validity

Histograms. h(r k ) = n k. p(r k )= n k /NM. Histogram: number of times intensity level rk appears in the image

Knowledge-Based Segmentation of Brain MRI Scans Using the Insight Toolkit

FROM IMAGE RECONSTRUCTION TO CONNECTIVITY ANALYSIS: A JOURNEY THROUGH THE BRAIN'S WIRING. Francesca Pizzorni Ferrarese

Automatic Registration-Based Segmentation for Neonatal Brains Using ANTs and Atropos

Preprocessing II: Between Subjects John Ashburner

QUANTITATION OF THE PREMATURE INFANT BRAIN VOLUME FROM MR IMAGES USING WATERSHED TRANSFORM AND BAYESIAN SEGMENTATION

Computational Neuroanatomy

White Matter Lesion Segmentation (WMLS) Manual

MIDAS Project Definition

Brain Portion Peeling from T2 Axial MRI Head Scans using Clustering and Morphological Operation

Structural MRI of Amygdala Tutorial: Observation, Segmentation, Quantification

The BATCH Program. Andrew A. Maudsley, Ph.D. University of Miami 2005 A. A. Maudsley

CHAPTER 2. Morphometry on rodent brains. A.E.H. Scheenstra J. Dijkstra L. van der Weerd

AN AUTOMATED SEGMENTATION FRAMEWORK FOR BRAIN MRI VOLUMES BASED ON ADAPTIVE MEAN-SHIFT CLUSTERING

Single Voxel Spectroscopy Data Processing and Analysis Tools

Comparison Study of Clinical 3D MRI Brain Segmentation Evaluation

Normalization for clinical data

An ITK Filter for Bayesian Segmentation: itkbayesianclassifierimagefilter

Ventricle slice detection in MRI images using Hough Transform and Object Matching techniques

Single Subject Demo Data Instructions 1) click "New" and answer "No" to the "spatially preprocess" question.

Playing with data from lab

Automatic Generation of Training Data for Brain Tissue Classification from MRI

Subvoxel Segmentation and Representation of Brain Cortex Using Fuzzy Clustering and Gradient Vector Diffusion

Multiple Sclerosis Brain MRI Segmentation Workflow deployment on the EGEE grid

FSL Workshop Session 3 David Smith & John Clithero

Slicer3 Training Compendium. Slicer3 Training Tutorial ARCTIC (v1.2)

ANALYSIS OF FUNCTIONAL MAGNETIC RESONANCE IMAGING DATA USING SPM99: VOXEL-BASED MORPHOMETRY DONNA ROSE ADDIS

LST: A lesion segmentation tool for SPM

GLIRT: Groupwise and Longitudinal Image Registration Toolbox

Automated Brain-Tissue Segmentation by Multi-Feature SVM Classification

Neuroimaging and mathematical modelling Lesson 2: Voxel Based Morphometry

Brain Explorer for Connectomic Analysis (BECA) Software Manual

Basic principles of MR image analysis. Basic principles of MR image analysis. Basic principles of MR image analysis

Structural MRI of Amygdala Tutorial: Observation, Segmentation, Quantification

Structural Segmentation

Norbert Schuff Professor of Radiology VA Medical Center and UCSF

MIDAS FAQ. 1) Supported MR Systems and SI Acquisition Protocols. 2) Recommended Computer Systems

Magnetic resonance image tissue classification using an automatic method

Structural Segmentation

Automatic Generation of Training Data for Brain Tissue Classification from MRI

Stroke Quantification Tool (Sonia) Ver User Manual

Brain Mask User s Guide

Slicer3 Training Tutorial Using EM Segmenter with Non- Human Primate Images

SPM Introduction. SPM : Overview. SPM: Preprocessing SPM! SPM: Preprocessing. Scott Peltier. FMRI Laboratory University of Michigan

SPM Introduction SPM! Scott Peltier. FMRI Laboratory University of Michigan. Software to perform computation, manipulation and display of imaging data

Norbert Schuff Professor of Radiology VA Medical Center and UCSF

Fast Fuzzy Clustering of Infrared Images. 2. brfcm

The organization of the human cerebral cortex estimated by intrinsic functional connectivity

Webpage: Volume 3, Issue V, May 2015 eissn:

BrainMask. Quick Start

Neuroimaging Group Pipeline Quick Start Manual

Modified Expectation Maximization Method for Automatic Segmentation of MR Brain Images

Automatic MS Lesion Segmentation by Outlier Detection and Information Theoretic Region Partitioning Release 0.00

Application of fuzzy set theory in image analysis. Nataša Sladoje Centre for Image Analysis

Ischemic Stroke Lesion Segmentation Proceedings 5th October 2015 Munich, Germany

MARS: Multiple Atlases Robust Segmentation

QUANTITATION OF THE PREMATURE INFANT BRAIN VOLUME FROM MR IMAGES USING WATERSHED TRANSFORM AND BAYESIAN SEGMENTATION

Performance Evaluation of the TINA Medical Image Segmentation Algorithm on Brainweb Simulated Images

International Journal of Engineering Science Invention Research & Development; Vol. I Issue III September e-issn:

VBM Tutorial. 1 Getting Started. John Ashburner. March 12, 2015

Fuzzy Multi-channel Clustering with Individualized Spatial Priors for Segmenting Brain Lesions and Infarcts

Analysis of fmri data within Brainvisa Example with the Saccades database

IS MRI Based Structure a Mediator for Lead s Effect on Cognitive Function?

ADAPTIVE GRAPH CUTS WITH TISSUE PRIORS FOR BRAIN MRI SEGMENTATION

Ensemble registration: Combining groupwise registration and segmentation

Surface-based Analysis: Inter-subject Registration and Smoothing

Measuring baseline whole-brain perfusion on GE 3.0T using arterial spin labeling (ASL) MRI

BDP: BrainSuite Diffusion Pipeline. Chitresh Bhushan

An integrated software solution for improving neuroimaging data archival, management, and processing - The experience from the Queen Square MS Centre

BrainMask. Quick Start

UGviewer: a medical image viewer

Segmenting Glioma in Multi-Modal Images using a Generative Model for Brain Lesion Segmentation

syngo.mr Neuro 3D: Your All-In-One Post Processing, Visualization and Reporting Engine for BOLD Functional and Diffusion Tensor MR Imaging Datasets

Whole Body MRI Intensity Standardization

Atlas of Classifiers for Brain MRI Segmentation

Tissue Tracking: Applications for Brain MRI Classification

Image Registration + Other Stuff

Fuzzy Ant Clustering by Centroid Positioning

Transcription:

MRI Segmentation MIDAS, 2007, 2010 Lawrence O. Hall, Dmitry Goldgof, Yuhua Gu, Prodip Hore Dept. of Computer Science & Engineering University of South Florida CONTENTS: 1. Introduction... 1 2. Installing the program... 2 3. Starting the Program... 3 3.1. Widget Interface... 3 3.2. Segmentation with an Other Tissue Mask... 3 3.3. Batch Operation... 4 4. Data Formats... 5 5. Importing non-midas Segmentations... 5 6. Acknowledgments... 7 1. Introduction This program provides automatic segmentation of cerebrospinal fluid (CSF), white matter, and gray matter from MR imaging of the human brain. Up to three MRI data types can be used: T1- weighted, Proton-Density weighted and T2-weighted images. Since T1 and T2/PD commonly have different spatial resolutions, data alignment is performed by registration software if two or three image types are used by segmentation methods. Four methods are implemented: 1. Single pass algorithm (volume based). The single pass algorithm partitions the entire image volume by scanning the data from the disk only once. We provide a final computationally inexpensive segmentation refinement step after the Fuzzy clustering algorithm to model the clusters as Gaussians and compute the covariance matrix associated with features. Then the final segmentation is produced computing posterior probabilities assuming uniform priors for the 3 classes. 2. Bipartite-Merger. The Brfcm algorithm over-classifies each slice of a volume in to 9 clusters (excluding air). Out of these, some are purely extra cranial (skull etc.) and the rest contain intracranial tissues. The Bipartite merger then optimally matches the clusters between any two adjacent slices of a volume using a minimally weighted bipartite matching algorithm. Thus, in our case we obtain 9 chains (a chain is a series of similar matched clusters in the Z direction) of clusters. Heuristics are then used to label which chain is what i.e. which chain is skull (BET mask used for it), which one is CSF, gray matter or white matter. After labeling, split gray matter 1

or white matter chains are then merged to form complete gray matter or white matter segmentation. 3. FSL/FAST (old version). See the FSL website [http://www.fmrib.ox.ac.uk/fsl/] for a full description. 4. FSL/FAST 4.1. 2. Installing the program The program is automatically installed as part of the MIDAS system. If a separate installation is required the following procedure can be used: 1. Unzip "MriSeg" package, then move all files and folders to the desired location: (e.g.) c:\midas\bin\ 2. Go to c:\midas\bin\ Use edit software to open: runmrisegui.bat (UI mode) Or runmrisegbm.bat (batch mode) Change: Set Midasbindir= c:\midas\bin\ Then save it. 3. Copy c:\midas\bin\native-properties\windows\nativecodes.properties to $home\$usr\midas\modules\nativecodes.properties. NOTE: For the Linux version installing, the MriSeg programs and registration programs permission needs to be modified to allow execute permission before running. Ex: Go to INSTALL_PATH/Midas/Bin/mriseg/bin/Linux Type: chmod 755 * Go to INSTALL_PATH/Midas/Bin/native-codes/Linux Type: chmod 755 * Similar changes will be made runmrisegui.linux or runmrisegbm.linux export RegLib= INSTALL_PATH/Midas/Bin/javadir Before copying INSTALL_PATH/Midas/Bin/native-properties/Linux/nativecodes.properties to $home/$usr/midas/modules/nativecodes.properties. Use edit software to open this file too. Change three locations: Mriseg.folder= INSTALL_PATH/Midas/Bin/mriseg Mriseg.bin.folder= INSTALL_PATH/ Midas/mriseg/bin/Linux Binaries.folder= INSTALL_PATH/ Midas/native-codes/Linux 2

3. Starting the Program 3.1. Widget Interface The program can be run from the MIDASTools button bar icon:. The program can also be run from the command line. Go to INSTALL_PATH\Midas\Bin\ and run: runmrisegui.bat (windows) or runmrisegui.linux (linux). The program starts with the widget interface shown in Fig 1: Fig. 1 Fig. 2 The first operation must be to click the Browse button, which opens up the MIDAS browser (Fig. 2). Selection must be done on the study level, then click Done. Next, select the segmentation method and options. Currently (Nov 2010), the FSL with bias field correction and partial volume output is the method used for 3T data acquired for the MIDAS project. The default value for fractional intensity threshold for FSL4.1 is 0.5, whereas a value of 0.3 was used for previous version of FSL. Then click Process to start processing. 3.2. Segmentation with an Other Tissue Mask This option can be used when there are lesions in the brain, for the purpose of: a) having a separate tissue map covering the lesion volume, and b) stopping the voxels in the lesion from affecting the quality of the segmentation of the normal tissue regions. This option is only available with FSL/FAST 4.1 segmentation method. 3

The procedure requires the following steps: 1. The user must manually create a mask describing the lesion volume(s) (VOIs) at the T1 image resolution. This can be done using an external segmentation package (e.g. Slicer or MIPAV) 2. Save the VOI mask as either Analyze or NIFTI hdr/img pair files with the file names other.hdr and other.img and located under the same directory as the T1 image. Data format should be byte or integer. 3. In the segmentation program, select the FSL/FAST 4.1 method, and click the Change fractional intensity threshold and additional flag for FSL box under the main widget. Then under the pop-up window, put --class=4 option in the additional flag text box, then click OK button. The rest of the segmentation processing will be exactly the same way as the regular segmentation. The result of the segmentation will contain 4 tissue types: CSF, Grey-Matter, White- Matter, and Other, the pair of image files for Other tissue is simply a copy of original other.hdr and other.img files. Therefore, if the Save partial volume option was selected for the segmentation results, the result for Other will remain the 0/1 value of the original mask. 3.3. Batch Operation Run: setpath.bat type: Java edu.miami.midas.mriseg.runmriseg and you will get usage information of this program: -f <program name: fsl or clp or bip or spass> choose different segmentation tool -s <1, 2a, 2b, 3 > number of features to be used, default is 1, 2a (T1 and T2) 2b(T1 and PD), 3 (all features) -p <path for subject.xml> ex: c:\data\mn012\subject.xml -d <Study ID> -fit: <The fractional intensity threshold value for brain extraction tool: 0-1, default is 0.2> -b : do bias field correction, but do not save -ob: do bias field correction and save corrected data -ov: save partial volume option (not all packages have this option) -a: use apriori probability maps (FSL package only)" -3d: <followed by two numbers, which are the parameters for 3d bias correction (s ingle pass package only), e.g. -b -3d 2 3 > ex (1): java edu.miami.midas.mriseg.runmriseg -f fsl s 1 -p D:\data\VOL016\subject.xml -d 1.3.12.2.1107.5.2.13.20560.30000005072118294371800000001 -ob (use FSL package, T1 feature is used, do bias field correction and save corrected data) ex (2): java edu.miami.midas.mriseg.runmriseg -f spass s 1 -p D:\data\VOL016\subject.xml -d 1.3.12.2.1107.5.2.13.20560.30000005072118294371800000001 -b -3d 2 3 -ov (use single pass package, T1 feature is used, do bias field correction but don t save corrected data, output partial volume result) Segmentation with FSL4.1 in batch mode. 4

If this segmentation procedure is called from the IDL BATCH program, it runs the batch file named SEG_FSL4_T1.BAT, which is located under MIDAS bin directory. IDL batch passes the subjectxml path, studyid, and additional options such as ob. These are then passed to the Java program. The actually call to the JAVA class looks like this: JAVA %fullpath%runmriseg f fsl4 p C:\project_dir\subject_dir\subject.xml d studyid Additional options include: -s # number of image channel, default is 1 -fit 0.## user defined fractional intensity threshold, between 0~1, default is 0.2 -B -b save bias corrected image, default -nopve do not do partial volume classification -ov save partial volume classification, default -a use prior probability maps for initialization, must give path of probability map, currently not supported. --class=4 using 4 tissue types, this option works with additional mask image other.hdr/other.img under mri directory. 4. Data Formats The format used for the tissue segmentation results is based on the Analyze format: - Orientation is same as MRI - Data type is byte, with value 0 or 255 equivalent where 255 indicates it is a particular tissue type and 0 means it is not In the MIDAS XML description, the segmentation data is stored as a FRAME, and the descriptor of the corresponding tissue type is given using the FRAME_TYPE definition tag multiple images representing each tissue type are concatenated into a single file, a corresponding header file is created too. 5. Importing non-midas Segmentations Results from other segmentation programs, such as SPM, can be imported in ANALYZE format (*.hdr/*.img pair). Start the MIDAS Importer, then click on the Data Management button and click on the Import Segmentation button (Figure 3). 5

Figure 3 Click on the GM button (Figure 4) for importing segmented gray matter images file, browse to a directory that has segmented images, select the *.img file, and repeat this procedure for importing *.img files of white matter and CSF images. Figure 4 6

Click on the Input 1 button (Figure 5) under the heading MR files used to create Segmentation to identify the type of MRI data used (T1-weighted and T2-weighted and/or PD-weighted) for segmentation. For example, the segmented image files in the PracticeSVSData directory were created using a T1-weighted MRI data only, so select the MRI_T1 Volume node, click the Save button and wait until the Analyze Format to import?!.. widget vanishes. Figure 5 6. Acknowledgments This program was developed by Yuhua Gu, and Prodip Hore, and used the method proposed by Dr. Lawrence O. Hall and Dr. Dmitry Goldgof. Implementation of FSL 4.1 was done by Jiping Zhan in 2010. This work was supported by NIH grant EB000822 under the MIDAS project. 7