Patch-based Segmentation with Spatial Consistency for Detection and Segmentation of Multiple Sclerosis Lesions in Brain MRI

 

Overview

Multiple Sclerosis (MS) is the most common non traumatic neurological disease in young adults. MS is typically expressed in focal white matter lesions, which are characterized by demyelination, axonal injury and axonal conduction block. These white matter lesions are visible in conventional magnetic resonance imaging (cMRI). The process of lesion segmentation is important for assessment of disease progression, evaluation of drug therapy efficiency and quantitative analysis of lesion burden. The segmentation process is still carried out in a manual or semiautomatic fashion by clinicians, because published automatic approaches have not been universal enough to be widely employed in clinical practice. Thus, MS lesion segmentation remains an open problem.
 

In our research, we presented an automatic method for segmentation of Multiple Sclerosis (MS) in brain MRI. The approach is based on similarities between multi-channel (T1, T2 and FLAIR) patches. MS lesion patch database is build using training images for which the label maps are known. Using k-nearest neighbor (k-NN) search, for each patch in the testing image, similar patches are retrieved from the database. The matching labels for these patches are then combined to result an initial segmentation map for the test case. Finally an iterative Example-Based Refinement process based on the initial segmentation map is performed. By that we ensure the spatial consistency and the neighbors relations.
 

The evaluation was preformed, in leave-one-out scheme, for each testing image in the MS lesion segmentation challenge at MICCAI 2008. The results were competitive with state-of-the-art methods achieving an average Dice index of 31%. Furthermore, the results show its general applicability to the problem of lesion segmentation.
 

Segmentation Examples

Two segmentation examples (top: UNC06, bottom: UNC02): (a) FLAIR image; (b) first iteration;

(c-e) iterations 2,3 and 5; (f) ground truth. Proposed method in Blue; Reference segmentation in Red.

Publications

Mechrez, R., Goldberger, J., & Greenspan, H. (2015, March). MS lesion segmentation using a multi-channel patch-based approach with spatial consistency.

In SPIE Medical Imaging (pp. 94130O-94130O). International Society for Optics and Photonics.

horizontal rule

For problems or questions regarding this web page contact hayit@eng.tau.ac.il.
Last updated: 12/11/15.