StrokeTimer: Robust Representation Learning for Ischemic Stroke Onset-Time Estimation from Non-contrast CT
Researchers have developed StrokeTimer, a new framework designed to estimate the onset time of ischemic strokes using non-contrast CT scans. The system employs self-supervised disentanglement and energy-guided contrastive learning to identify subtle signs of stroke and handle data imbalances and variations across different scanners. In trials on a large multi-center dataset, StrokeTimer achieved a macro AUC of 0.69 and a macro F1-score of 0.57, significantly outperforming existing methods and showing potential to aid clinical treatment decisions. AI
IMPACT This AI model could improve treatment decisions for ischemic stroke patients by providing more accurate onset-time estimations from CT scans.