Predicting language treatment response in bilingual aphasia using neural network-based patient models

Area of Science:

• Computational linguistics
• Neuroscience
• Speech-language pathology

Background:

• Predicting language therapy outcomes in bilinguals with aphasia (BWA) is challenging due to numerous pre- and poststroke factors.
• Computational models can simulate language impairment and treatment responses in BWA to guide therapy.

Purpose of the Study:

• To utilize the BiLex computational model to simulate language deficits and treatment responses in a cohort of Spanish-English BWA.
• To assess BiLex's accuracy in predicting naming ability, poststroke impairment, and treatment response in treated and untreated languages.

Main Methods:

• The BiLex computational model was employed to simulate prestroke naming ability and poststroke naming impairment in Spanish and English for 13 BWA.
• Simulations covered treatment response in the treated language and cross-language generalization to the untreated language.
• A cross-validation approach was used to test the model's generalizability on unseen patient data.

Main Results:

• BiLex accurately and robustly predicted treatment effects in the treated language for BWA.
• The model captured varying degrees of cross-language generalization in the untreated language.
• Cross-validation confirmed BiLex's ability to generalize predictions to patients not included in model training.

Conclusions:

• The BiLex model demonstrates significant potential for predicting language therapy outcomes in bilinguals with aphasia.
• Computational modeling, specifically using BiLex, can assist in developing tailored rehabilitation strategies for BWA.
• These findings highlight the utility of computational approaches in optimizing language recovery for bilingual individuals post-stroke.