Dimensionality reduction establishes specificity in lesion network mapping

BackgroundLesion network mapping (LNM) has emerged as a powerful tool for linking focal brain lesions to distributed functional networks. However, the biological specificity of these networks has been questioned. Recent mathematical derivations suggest that LNM-derived maps may trivially track the normative connectomes global degree vector rather than specific symptom-related topography, potentially rendering them biologically nonspecific. MethodsWe introduced a rigorous validation pipeline to distinguish true network specificity from low-dimensional connectome artifacts. We projected lesion connectivity maps into a low-dimensional feature space defined by the principal gradients and eigenmodes of the normative connectome. We applied this framework to a large-scale dataset of 858 lesions associated with four distinct clinical cohorts: obsessive-compulsive disorder (OCD), schizophrenia, aphasia, and epilepsy. We performed multivariate classification to determine if symptom-associated lesions occupied distinct regions of the functional manifold compared to null distributions. ResultsOur analysis revealed a sharp dissociation in network specificity across disorders. While schizophrenia-associated lesions were indistinguishable from null models (Accuracy=0.51, p=0.412), confirming the "degree artifact" hypothesis for this cohort, other disorders displayed significant network specificity. Lesions associated with OCD (Accuracy=0.58, p=0.036), aphasia (Accuracy=0.60, p=0.007), and epilepsy (Accuracy=0.61, p=0.002) occupied distinct regions of the functional manifold significantly different from the normative connectome baseline. ConclusionsThese findings demonstrate that while LNM is sensitive to connectome-level artifacts, it retains genuine biological specificity for distinct clinical phenotypes. The proposed linear projection framework offers a standardized, computationally efficient benchmark for assessing network specificity against methodological noise.