Texture Analysis of the Brain in Amyotrophic Lateral Sclerosis

Abstract

In this paper, we applied texture analysis to evaluate cerebral degeneration in amyotrophic lateral sclerosis (ALS). Two well-known methods, the gray level co-occurrence matrix (GLCM) and the gray level aura matrix (GLAM) were employed to extract texture features from routine T1 and T2 MR images. Texture features were analyzed by statistical inference, support vector machine to determine classification rate, and receiver operator characteristic curve (ROC) analysis. Twenty control subjects (average age = 56.8± 12.4) and 19 ALS patients (average age = 56.7± 13.7) were recruited for the study. Texture features were statistically different in ALS compared to controls (p< 10−4 for T1, and p< 10−5 for T2) and provided a classification rate with more than 76%, and 82% accuracy on T1 and T2 weighted images, respectively. ROC analysis yielded area under the curves approaching 0.93, and a maximal sensitivity and specificity of 100% and 95%, respectively. Texture features moderately correlated with parenchymal brain volume suggesting that atrophy partially accounted for the texture results; however, texture features had a superior classification rate indicating that other cerebral pathology due to ALS besides atrophy was being captured by texture analysis. We conclude that texture analysis shows promise as a quantitative biomarker to study cerebral degeneration in ALS. To our best knowledge, this is the first study to apply the GLAM as a texture method to medical image analysis, showing that it provides features superior (or at least comparable) to the well known GLCM features.