JMIR Publications

ABSTRACT

In this study, we develop a content-based medical image retrieval (CBMIR) system meticulously designed to cater to seven distinct types of brain tumors as seen in magnetic resonance brain images. Our system is tailored to assist radiologists and healthcare professionals in efficiently retrieving pertinent historical medical images, thereby substantially augmenting the quality of clinical diagnoses and the progress of endeavors in radiology workflow research. The core innovation of our study is the introduction of a state-of-the-art deep learning-based feature extraction algorithm specifically engineered for the CBMIR system. We employ GoogLeNet as the primary architecture for the deep learning network. To further enhance the system's capacity for capturing nuanced and generalized local features, we incorporate generalized-mean pooling. Additionally, we implement an embedding layer to effectively reduce image dimensions. The empirical findings of our research demonstrate the performance and robustness of the proposed CBMIR system. Our system achieves a remarkable mean average precision score of 89.16% and an equally impressive Precision@10 score of 94.08%. These metrics affirm the system's efficacy in retrieving relevant medical images. Furthermore, we seamlessly integrate the CBMIR service into a picture archiving and communication system (PACS) by successfully harmonizing two open-source projects. This milestone marks significant progress toward establishing our CBMIR system as an indispensable tool for both clinical practice and medical research, with the potential to significantly advance brain tumor diagnosis and research.