Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Predicting Stroke Recovery with Nonlinear Low-Dimensional Embeddings of Behavioral Profiles

Jax Skye¹, Dorit Kliemann¹, Aaron D. Boes, James Traer¹; ¹University of Iowa

Presenter: Jax Skye

Stroke is a leading cause of disability worldwide. Predicting functional outcomes is challenging due to heterogeneity in post-stroke deficits and recovery profiles. We assessed prediction accuracies of 101 chronic outcomes from 78 acute behavioral measures and (hypothesizing redundancy in the predictors) from low-dimensional embeddings thereof. Nonlinear 2D UMAP embeddings yielded predictions comparable to those from all predictors. We identified brain damage patterns associated with specific behavioral profiles (extrema of the patient distribution in UMAP embeddings). We show that predictions based on only four acute tests—chosen as best linear approximations to UMAP embeddings—matched prediction accuracies from all 78 tests, suggesting nonlinear dimensionality reduction offers novel and interpretable tools for understanding behavioral outcomes of brain lesions and clinical assessment.

Topic Area: Methods & Computational Tools

Extended Abstract: Full Text PDF