Metabolic Connectivity Alterations in Amyotrophic Lateral Sclerosis: Individual Network Analysis Based on Wasserstein Distances

Abstract

Molecular connectivity analysis with PET imaging offers a promising approach for characterizing brain network alterations in neurodegenerative disorders. In this study, we introduce Wasserstein distance (WD) as an alternative to Kullback–Leibler divergence similarity estimation (KLSE) for constructing single-subject metabolic connectivity networks. Using 18F-FDG PET data from 167 individuals with amyotrophic lateral sclerosis (ALS), 36 healthy volunteers (HV), and 25 ALS mimics, we generated WD- and KLSE-based connectivity matrices across 77 atlas-defined brain regions and evaluated corresponding graph-theory-based nodal metrics. WD- and KLSE-derived nodal measures were strongly correlated, indicating methodological consistency. Compared with HVs, age-matched subjects in the ALS group (ALSamHV) showed significant alterations in frontal, temporal, cerebellar, and occipital network nodes, with WD-based metrics revealing differences across more brain regions than the KLSE-based approach. Support vector machine classification of ALSamHV vs. HV demonstrated that both connectivity approaches matched voxel-wise PET performance with accuracy higher than 0.80 (yet significantly lower than voxel-wise data, p<0.05 ), while significantly outperforming image-based classification for ALS vs. ALS Mimics (p<0.01 ), with WD achieving the highest accuracy of 0.65 (p<0.001 ). These findings support WD-based metabolic connectivity as a sensitive, data-efficient framework for detecting disease-related network alterations and motivate its application to a broader range of PET tracers and cohorts.