Post-stroke aphasia analysis using topological alterations in brain functional networks.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Aphasia affects nearly one-third of stroke survivors, impacting language function.
• Mechanisms of network reorganization and objective biomarkers for aphasia subtypes remain unclear.
• Rapid and accurate classification of aphasia subtypes presents a significant clinical challenge.

Purpose of the Study:

• To develop a diagnostic framework for classifying post-stroke aphasia using resting-state fMRI-derived functional brain networks.
• To identify topological changes in brain networks associated with aphasia.
• To establish a robust and reproducible method for objective aphasia classification.

Main Methods:

• A diagnostic framework was developed analyzing topological changes in resting-state functional brain networks.
• A feature selection pipeline combined topological features, ReliefF algorithm, elbow method, and cross-validation.
• A cubic Support Vector Machine (SVM) classifier was employed for patient classification.

Main Results:

• The model achieved 88.70% accuracy and 92.92% F1 score on a public dataset, validated by 88.1% accuracy and 92.76% F1 score on an in-house dataset.
• Post-stroke aphasia patients (PWA) exhibited higher global/local network efficiency, increased clustering, and shorter path lengths compared to non-aphasic stroke patients.
• Distinct neural patterns were identified for Anomic, Broca, Conduction, and Global aphasia subtypes, indicating divergent pathophysiology.

Conclusions:

• The proposed framework significantly improves classification accuracy, interpretability, and reproducibility for aphasia.
• This approach offers a potential basis for a new objective diagnostic tool for aphasia.
• Identifying distinct neural patterns for aphasia subtypes supports targeted therapeutic strategies.