Decoding Post-Stroke Cognitive Impairment After Acute Basal Ganglia Infarction: The Synergistic Role of Functional Segregation and Integration in an SVM fMRI Framework

Abstract

Objective

To investigate whether dynamic changes in resting-state functional MRI (rs-fMRI) metrics can serve as sensitive biomarkers for distinguishing acute basal ganglia cerebral infarction (BGCI) patients with post-stroke cognitive impairment (PSCI) from those without (non-PSCI).

Materials and Methods

Data on various rs-fMRI metrics dynamic functional connectivity (dFC), dynamic amplitude of low-frequency fluctuation (dALFF), and percent amplitude of fluctuation (PerAF) were acquired using a Siemens Prisma 3.0T scanner from 38 PSCI and 36 non-PSCI patients, with follow-up assessments. Functional segregation and integration were analyzed using PerAF, dALFF, and dFC. Feature extraction and selection were performed using support vector machine (SVM), followed by classifier construction and evaluation.

Results

Patients with PSCI showed decreased PerAF in the left cerebellar Crus I (lCbeCru1) and increased dALFF in the right cerebellar Crus I and left lingual gyrus compared to non-PSCI patients. Altered dFC was observed between cerebellar cognitive-related seed regions and widespread cortical areas, with increased dFC in the right cerebellar Crus II and left cuneus, and decreased dFC primarily in the inferior frontal gyrus and superior temporal gyrus. Among single-feature models, dFC achieved the best classification performance (AUC = 0.98, accuracy = 94.52%, sensitivity = 97.14%, specificity = 92.11%, precision = 91.89%). A combined feature model yielded the highest precision (94.12%).

Conclusion

SVM-based integration of PerAF, dALFF, and dFC features holds promise as a neuroimaging biomarker for PSCI in patients with BGCI. This approach may support more precise early rehabilitation strategies in clinical practice.