Machine Learning Predictions of Recovery in Bilingual Poststroke Aphasia: Aligning Insights With Clinical Evidence.

Area of Science:

• Neuroscience
• Computational Linguistics
• Speech-Language Pathology

Background:

• Personalized treatment planning for bilingual individuals with poststroke aphasia requires predicting therapy outcomes.
• Treated language improvement (TLI) and cross-language generalization (CLG) are key outcomes.
• Identifying predictive features is crucial for clinical decision-making.

Purpose of the Study:

• Evaluate machine learning models for predicting TLI and CLG in Spanish-English bilinguals post-stroke.
• Identify key demographic, clinical, and linguistic features that predict these outcomes.

Main Methods:

• Forty-eight bilingual individuals with poststroke aphasia received 20 speech-language therapy sessions.
• Six machine learning algorithms were used with 16 curated features (demographics, language, cognition, bilingual experience).
• Models predicted treatment responders (TLI) and CLG.

Main Results:

• Top models achieved F1 scores of 0.767 for TLI and 0.790 for CLG.
• Aphasia severity in the treated language, education, and cognitive performance predicted TLI.
• Aphasia severity in the untreated language and cognitive performance predicted CLG.

Conclusions:

• Machine learning models can predict TLI and CLG in bilingual stroke survivors.
• Aphasia severity and cognitive performance are significant predictors.
• These findings can improve clinical care for underserved populations.