Solving Inverse Problems in Medical Imaging with Score-Based Generative Models

The idea in plain language.

Combine a learned image prior with the measurement process available at reconstruction time.

How it works

A score model first learns the distribution of medical images. At reconstruction time, its sampling procedure is guided both by that learned prior and by consistency with observed measurements. Because the measurement model is introduced at this stage, the method can be applied to different measurement processes without learning a separate direct mapping for each one.

What to keep in mind

This is a research reconstruction method, not evidence of clinical readiness. A plausible reconstructed image may contain errors and requires task-specific validation.

Source: Solving Inverse Problems in Medical Imaging with Score-Based Generative Models. The original manuscript contains the methods, experiments, figures, and references. An arXiv posting date may follow an earlier conference publication. Read the linked record for version history.