A weakly-supervised counterfactual diffusion model, IgCONDA-PET, can effectively detect anomalies in PET images by generating healthy counterparts for unhealthy images and identifying the differences.
