Accurate support vector machine identification of limb-onset amyotrophic lateral sclerosis using resting-state activity of regions within executive control network

Abstract

AbstractAmyotrophic lateral sclerosis (ALS) is a motor neuron degenerative disorder facing diagnostic challenges due to its highly variable presentation and symptom overlap. In other neurodegenerative disorders, support vector machine (SVM) classifiers have utilized neuroimaging to address these challenges. Given functional alterations may be the earliest detectable in ALS, we aimed to uncover resting-state functional MRI (rs-fMRI) biomarkers for SVM classification. Resting-state networks derived from independent component analysis were compared between limb-onset ALS patients (n = 14) and controls (n = 12). A cluster within the executive control network (EXN) localizing predominantly to the anterior cingulate gyrus (ACG) was significantly decreased in limb-onset ALS. Activity of this cluster was able to develop a SVM with 86% sensitivity and 87% specificity on the validation dataset. These findings suggest the ACG and EXN may be important in classifying limb-onset ALS patients and could be incorporated into multi-modal SVM classifiers.