Lesion-DDPM: Lesion-Enhanced 3D Diffusion for MS MRI Synthesis
Researchers have developed Lesion-DDPM, a novel 3D conditional diffusion framework designed to synthesize medical images for multiple sclerosis (MS) research. This method specifically enhances the generation of images that accurately represent small, sparse lesions, a common challenge in MS datasets. By incorporating multi-level anatomical mask injection and a lesion-weighted reconstruction loss, Lesion-DDPM aims to improve the robustness of neuroimaging machine learning models. Experiments show that models trained on Lesion-DDPM-generated scans achieve better performance on downstream lesion segmentation tasks compared to those trained on existing synthetic datasets or real scans alone. AI
IMPACT Improves synthetic data quality for MS research, potentially accelerating development of diagnostic AI models.