Individual-Specific Functional Connectivity-Based State Classification and Prognosis Prediction for Disorders of Consciousness.

Accurate prognosis and treatment targeting for disorders of consciousness (DOC) remain challenging due to profound neurobiological heterogeneity. Current approaches to DOC state classification and prognostic prediction, which rely on resting-state functional magnetic resonance imaging (rs-fMRI)-based functional connectivity (FC) analysis, are limited by signal blurring from group-level averaging and cross-participant spatial variability. To overcome these limitations, we propose a novel framework integrating themulti-task learning-based sparse convex alternating structure optimization (MTL-sCASO). By jointly modeling multiple subjects within a unified optimization framework, MTL-sCASO reduces the confounding effects of spatial variability across participants, while decomposing each subject's rs-fMRI signals into individual-specific and shared FC components, thereby preserving subject-specific patterns that would otherwise be obscured. Leveraging these individualized FC alongside clinical data, we develop machine learning classifiers not only for DOC state discrimination and prediction of prognostic improvement, but also identify critical FC pairs and brain network features fundamental to both tasks. Our results demonstrate superior performance over conventional methods, achieving an accuracy of 81% in state classification and 77% in prognostic prediction. This framework advances precision diagnosis and reliable prognosis in DOC by shifting rs-fMRI analysis from group-level averaging to an individualized functional-connectivity perspective. This shift establishes an FC-based analytical paradigm that overcomes neurobiological heterogeneity and enables subject-tailored clinical decision-making.