OpenMAP-HIE Unsupervised Diffusion Model for Neonatal Lesion Segmentation
October 22, 2026
OpenMAP-HIE uses a ControlNet-based diffusion model to perform unsupervised lesion localization on ADC images. It detects anomalies by comparing conditional and unconditioned noise predictions, using pathology-specific ADC reduction constraints to isolate neonatal hypoxic-ischemic encephalopathy without requiring voxel-wise supervision.
HOW THIS AFFECTS YOU
●
researcherYou can utilize score-based diffusion for unsupervised medical anomaly detection.
●
healthThis could improve automated lesion localization in neonatal imaging without massive labeled datasets.