Pediatric Brain Tumor MRI Segmentation

Brain MRI is essential for pediatric brain tumor assessment, but incomplete imaging sequences are common and significantly degrade the performance of deep learning models trained on complete data. Using multi-institutional and clinical trial datasets, this study developed and evaluated strategies to handle missing MRI sequences, including dropout-trained segmentation, image synthesis, and simple substitution methods. The dropout-trained model showed robust segmentation and stable survival prediction despite missing inputs, while generative synthesis improved image quality and interpretability, supporting more reliable real-world deployment of AI tools in pediatric neuro-oncology.