Novel content based image retrieval methods for medical applications

Abstract

Brain tumors are a major global cause of mortality, necessitating efficient solutions newlinefor medical image retrieval. Content-based medical image retrieval (CBMIR) systems newlineenable clinicians to efficiently retrieve relevant medical images from large databases, newlineaiding in diagnosing and treatment planning by comparing new cases, like neoplastic newlinebrain lesions or gliomas, with similar past cases, ultimately enhancing decisionmaking and bridging gaps in healthcare delivery. newlineThe primary challenge for CBIR systems is bridging the gap between low-level image newlinefeatures and high-level semantic concepts. This research addresses this gap by newlinedeveloping feature fusion strategies that combine handcrafted and deep features with newlineCNNs. The works tackle computational and storage challenges through dimensionality newlinereduction and improve retrieval accuracy with efficient indexing techniques and newlinesimilarity metrics, enhancing diagnostic interpretation and treatment planning. The newlineBraTS 2018 and 2020 datasets serve as a valuable benchmark for evaluating CBMIR newlinesystems, featuring MRI scans from multiple institutions with diverse imaging newlineprotocols and scanners. These datasets include both high-grade and low-grade newlinegliomas, with expert-provided annotations, ensuring robust training and improved newlinegeneralizability. newline