Online learning of relevance feedback from expert readers for mammogram retrieval

In content-based image retrieval (CBIR) relevance feedback schemes have been studied as a means to boost the retrieval performance in recent years. Despite the efforts in development of efficient algorithms for retrieving desired images from image databases, there often remains a gap between low-level image features and high-level semantic understanding in CBIR systems. In this paper, we investigate a technique based on online learning by relevance feedback for retrieval of mammogram images that contain perceptually similar lesions with clustered microcalcifications. Our approach applies support vector machine (SVM) regression for supervised learning and employs the concept of incremental learning to incorporate relevance feedback online. The proposed approach is demonstrated using a database of 200 mammogram images with clustered microcalcifications scored by experienced radiologists.