Automatic segmentation of different lesion types

[frodo]

Identification of Multiple Sclerosis lesion subtypes and their quantitative assessments with EDSS using neuroimaging

https://link.springer.com/content/pdf/1 ... 0245-8.pdf

Abstract

It would be helpful to have an automatic segmentation method to provide an acceptable performance of Magnetic Resonance Imaging (MRI) on Multiple Sclerosis (MS) subtypes that have important roles in cure procedure. This article presents a technique to classify MS lesion subtypes in MR images and then evaluates the correlation between lesion subtypes and Expanded Disability Status Scale (EDSS).

The technique used textural features based on the Gray Level Co-occurrence Matrix (GLCM) and histogram information (mean and variance) to describe each lesion and normal tissue of FLAIR Images. Then it discriminated them with Support Vector Machine (SVM).

A comprehensive post-processing module improved the quality of segmentation. We found corresponding areas in T2-weighted (T2-w), T1-weighted (T1-w) called black holes, and T1-weighted enhancing (T1-enhancing) by mapping the extracted lesions of FLAIR slices on them. Multi-Layer Perceptron (MLP) classified the mapped lesions in three classes of T2, black holes and T1-enhancing.

Finally, the correlation between lesions volume of each class and EDSS was calculated. The performance evaluation resulted that the presented method allowed a higher value of sensitivity (86.1%) in black holes MS lesion subtypes classification. This sensitivity was higher than that obtained by others. This improvement beside the high accuracy (85.02%) and specificity (85.4%) make this pipeline an interest for clinical applications as an extra helpful way near neurologist for diagnosis.