Hayley Olson, B.S.
62 Goodnight Hall @ 12:00 pm - 1:00 pm
Learn more about Prosem
Verbal fluency tasks are used to assess language and executive function after stroke, yet traditional scoring methods may lack sensitivity to subtle semantic impairments in milder stroke populations. This study examines semantic distance, a computational measure of how closely related words are in meaning, as a metric for analyzing category fluency performance. Using a human perception-based semantic distance bigram model, we evaluated fluency data from individuals with left- and right-hemisphere subacute strokes and cognitively healthy adults (CHA). ANOVA revealed significant group differences in semantic distance, with CHA producing more semantically scattered responses than both stroke groups. Support Vector Regression was used to predict semantic distance from diffusion imaging metrics, demographic information, and clinical variables. The demographics-only model yielded the best performance (RMSE = 2.64, R² = 89.7%). Advanced diffusion outperformed standard diffusion models, highlighting the potential utility of more advanced diffusion measures. These findings suggest that semantic distance may capture meaningful language changes in post-stroke populations and that future work should explore its clinical and longitudinal applications across semantic models and fluency task types.
