Dis-connectome Biomarkers after Acute Stroke Predicting 12-Month Functional Independence, Motor Recovery, and Spasticity

Background: Stroke is a leading cause of long-term adult disability, yet recovery is highly heterogeneous and difficult to predict on an individual basis. Conventional prognostic markers, such as initial clinical severity and lesion volume, have limited accuracy, failing to capture the complex network-level consequences of a focal brain injury. The connectome framework, which models stroke as a disruption of large-scale brain networks, offers a more mechanistic approach to understanding post-stroke deficits. However, the prognostic value of a comprehensive set of structural disconnectome metrics for specific, long-term clinical outcomes remains largely unvalidated, representing a critical gap in translating advanced neuroimaging into clinical practice. Aim: The primary aim of this thesis was to investigate whether a comprehensive set of structural disconnectome and lesion geometry metrics, derived from routinely acquired acute-phase clinical scans, could accurately and independently predict functional independence, ambulatory status, and the development of spasticity at 12 months post-stroke. Materials and Methods: In a longitudinal cohort of 28 stroke survivors (ischemic and hemorrhagic), acute-phase CT or MRI scans were used to segment lesions. These patient-specific lesion masks were virtually overlaid onto a normative healthy connectome atlas to compute structural disconnectome maps. A comprehensive set of predictive features was derived, including: (1) lesion geometry metrics (Lesion Volume, Lesion Surface, Bounding Box Volume, Ratio of Bounding Box to Lesion Volume, and Lesion Sphericity); (2) global and network-level disconnection metrics (loss in Global Efficiency, disconnection within the Somatomotor network, Inter-Hemispheric motor disconnection, and disconnection from motor to frontoparietal/dorsal attention networks); (3) nodal disconnection metrics for key sensorimotor regions (M1, S1, SMA, PMd, Thalamus); and (4) composite and laterality indices (Motor Network Disconnection Index, M1 and SMA Laterality Indices). Multivariate predictive models were developed for 12-month outcomes (modified Rankin Scale [mRS], Functional Ambulation Classification [FAC], Modified Ashworth Scale [MAS]) and for continuous motor recovery (change in NIHSS motor score). Model performance was assessed and validated using cross-validation techniques. Results: Predictive models combining baseline clinical severity (NIHSS motor score) with imaging metrics consistently outperformed imaging-only specifications. Distinct predictive signatures emerged for each outcome. Functional independence (mRS ≤ 2) was best predicted by a model including Lesion Sphericity and MNDI3 (AUC = 0.974). Ambulatory independence (FAC ≥ 4) was dominated by the SMA Laterality Index and Global Efficiency loss (AUC = 0.952). In contrast, the presence of spasticity was most strongly predicted by the initial NIHSS motor score, with a secondary contribution from disconnection within the motor network (AUC = 0.771). Continuous motor recovery was substantially explained by a combined model including lesion volume and MNDI3 (R² = 0.66), whereas an imaging-only model performed poorly (R² ≈ 0.22). Discussion: The findings suggest that while the initial severity of corticospinal injury is the primary driver of maladaptive plasticity leading to spasticity, the ultimate ceiling of functional and ambulatory recovery is more closely tied to the topological characteristics of the brain injury. The predictive power of lesion geometry and network asymmetry underscores the importance of widespread, multi-system disruption for global disability and the critical role of interhemispheric balance for complex motor tasks like gait. The independent prognostic value of disconnectome metrics demonstrates their ability to capture mechanistic aspects of neural injury not fully represented by clinical scores alone. Conclusion: Structural disconnectome metrics are powerful, independent, and outcome-specific biomarkers for long-term recovery after stroke. This work provides a robust framework for integrating network neuroscience into clinical prognostication, representing a critical step towards the development of personalized and stratified neurorehabilitation strategies