Koushyar Kowkabzadeh of the Chalmers Signals and Systems Department, Division of Biomedical Engineering will present his masters thesis "Accurate brain tissue segmentation of MR images".
Date: Monday 16th of August 2010
Time: 10:00-11:00
Place: Lecture hall Kammaren, Sahlgrenska Universitetssjukhuset (between sections J and K of main hospital building entrance floor)
Abstract:
Accurate brain tissue segmentation of MR images has been one of the most important research areas for several years. It is important to have an accurate segmentation of different brain tissue types for various applications such as radiotherapy planning, image-guided interventions, surgical planning by using EEG or fMRI information and brain disease studies such as Alzheimer and MS. However, all the mentioned applications are crucially dependent on the level of accuracy of brain tissue segmentation. Thus, this is important to choose the best algorithm to reach the most accurate tissue segmentation.
In this project different software packages for brain tissue segmentation where studied to pick the most sophisticated ones for the evaluation and comparison. SPM8, FSL and FreeSurfer elected which are three most widely used brain tissue segmentation software packages. The evaluation for each software package carried out by performing tissue segmentation over 18 brain phantoms provided by Brainweb with 6 different noise levels and 3 different RF inhomogeneity values. Then the resulted images for each tissue type from each software compared voxel by voxel with the ground truth images for each tissue type. To provide a volumetric voxel by voxel comparison between ground truth images and output images from software packages the 3dOverlap command line from AFNI software package used together with Jimmy Shen’s library in MATLAB.
Finally to compare the performance of these three software packages the misclassification rate were calculated for each tissue type.
