Learning metrics for content-based medical image retrieval

Application of content-based image retrieval (CBIR) to medical image analysis has recently become an active research field. While many previous studies have focused on the feature design, the metric design, another key CBIR component, has not been well investigated in this application context. This paper presents a medical CBIR that adapts its similaritymetric from data by using information theoretic metric learning. Also we systematically compare our SIFT bag-of-words-based system with various plug-in similarity measures available in literature. The proposed systems are evaluated with the ImageCLEF-2011 benchmarking dataset. Our experimental results demonstrate the advantage of the proposed metric learning approach and L1 distance-based measures.