Multimodal MRI-Clinical Fusion Model Predicts 90-Day Stroke Outcome with AUC 0.885

Multimodal MRI-clinical fusion model predicted 90-day stroke functional outcome with AUC 0.885 (sensitivity 92%, specificity 76%) on internal test (n=50); external validation (n=37) showed similar trends. Retrospective, not yet externally validated in multicenter setting.

• Retrospective study with 300 internal patients (250 train/validation, 50 test) and an external cohort of 37 patients, all with acute ischemic stroke and multiparametric MRI.
• The fusion model (CNN-ViT + stacked logistic regression) outperformed both clinical-only SVM (AUC 0.878) and imaging-only model (AUC 0.782) on internal test.
• Major limitation: single-center data, small external cohort, retrospective design; prospective multicenter validation is needed before clinical use.

RadPigeon summaries are original and for information only. They are not clinical advice.