Graph-of-Differences: Anatomy-Structured Difference Alignment for Medical Image Re-Identification

The article introduces Graph-of-Differences (GoD), a novel approach for medical image re-identification (MedReID) that utilizes explicit anatomical structures to enhance identity comparisons. GoD constructs an anatomy graph for each image, facilitating soft node correspondence and computing differences based on matched anatomical regions, which leads to significant improvements in retrieval accuracy—7.1 percentage points on fundus images and 3.1 percentage points on chest X-rays over a strong baseline. This method not only enhances model performance but also provides interpretable and verifiable explanations, crucial for clinical applications where auditability is essential.