Using unsupervised dimensionality reduction to identify lesion patterns predictive of post-stroke aphasia severity

Area of Science:

• Neuroscience
• Neurology
• Computational Biology

Background:

• Voxel-based methods for brain lesion analysis are limited for broad behavioral measures.
• Understanding how distributed brain damage affects language is crucial for stroke rehabilitation.

Purpose of the Study:

• To identify representative stroke patterns using non-negative matrix factorization (NMF).
• To explore associations between these stroke patterns and aphasia severity in patients with left hemisphere strokes.

Main Methods:

• Structural MRIs from 107 left hemisphere stroke patients were used to segment lesions.
• Non-negative matrix factorization (NMF) was applied to spared tissue data to identify stroke patterns ('atoms').
• Linear regression analyzed the association between NMF patterns and aphasia severity (Western Aphasia Battery - Revised Aphasia Quotient).

Main Results:

• Five prototypical stroke patterns (NMF atoms) were identified.
• Reduced spared tissue in the MCA territory correlated with more severe aphasia.
• Increased spared tissue around the insula correlated with less severe aphasia.
• Damage to superior fronto-parietal regions showed a stronger association with severity due to critical language area isolation.

Conclusions:

• NMF provides a novel approach to characterize stroke patterns and their relation to language deficits.
• Preserved tissue in specific regions, particularly the insula, may mitigate aphasia severity.
• This method enhances understanding of lesion burden and location effects on language function post-stroke.