Generalizing post-stroke prognoses from research data to clinical data.

Area of Science:

• Neuroscience
• Computational Linguistics
• Medical Imaging

Background:

• Acquired language disorders (aphasia) affect a third of stroke survivors, with unpredictable recovery.
• Current prognostic research uses brain imaging and outcome scores, aiming for broad patient generalization.
• Predicting recovery is crucial for effective patient management and treatment planning.

Purpose of the Study:

• To develop and validate prognostic models for aphasia recovery that generalize across diverse patient populations and data types.
• To investigate the impact of data acquisition timing (research vs. clinical) on model generalizability.
• To identify and mitigate confounds, such as lesion evolution, affecting cross-dataset predictions.

Main Methods:

• Utilized datasets linking structural brain imaging (MRI, CT) with language outcome scores from stroke survivors.
• Developed predictive models trained on research data and tested generalizability across countries, languages (English, Spanish), and imaging modalities.
• Addressed the confound of delayed scan acquisition relative to language assessment by projecting lesion evolution.

Main Results:

• Stroke recovery prediction models demonstrated generalization across different countries, native languages, and neuroimaging technologies.
• Models successfully generalized from research scans (long-term) to clinical scans (acute phase) after accounting for lesion changes.
• Projection of lesion evolution significantly improved the generalizability of prognostic models to clinical data.

Conclusions:

• Prognostic models for aphasia recovery can generalize effectively across diverse settings and data types.
• Accounting for lesion dynamics over time is critical for accurate prediction when applying research findings to clinical scenarios.
• Lesion growth is a significant factor in lesion-symptom mapping and should be considered in future prognostic research.