Temporal Glioma Prediction

This study presents a self-supervised temporal deep-learning model designed to improve recurrence prediction in pediatric gliomas using longitudinal MRI data. The model, trained on nearly 4,000 scans from 715 patients, significantly outperformed traditional methods, with up to a 58.5% improvement in F1 score and strong predictive accuracy across multiple datasets. These findings suggest temporal deep learning can enhance personalized surveillance and may be applicable to other cancers and chronic diseases.