Revolutionizing Medical Diagnosis: A Deep Learning Approach for Content-Based Image Retrieval

Medical imaging is essential for disease diagnosis and techniques such as magnetic resonance imaging (MRI), X-ray, electroencephalogram (EEG) and Optical coherence tomography (OCT) provide vast volumes of data. Medical professionals have challenges in identifying pertinent diagnostic images due to the growing complexity and size of medical image repositories. In order to solve this, this study suggests a novel Content-Based Medical Image Retrieval (CBMIR) system based on Vision Transformers (ViT) that can reliably and precisely extract pertinent medical images. To facilitate supervised learning, the system utilizes a dataset of 50,000 images from various modalities. The model initializes with preprocessing the images using Contrast Limited Adaptive Histogram Equalization (CLAHE) for contrast enhancement and the Non-Local Means (NLM) algorithm for noise removal. Retrieval accuracy, precision, recall and mean Average Precision (mAP) measures were used to assess the system performance. With a retrieval accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$98.5\%$</tex> for 40 retrieved images and a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{mAP}$</tex> improvement to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$98.7{\text{\% }}$</tex> for 100 retrieved images, the proposed CBMIR approach achieved remarkable results. The system showed resilience and scalability across all modalities including OCT, EEG, MRI and X-ray. The predicted results especially for MRI data confirm that the model can manage intricate imaging characteristics and provide precise results. The suggested ViTbased CBMIR system offers a revolutionary tool for clinical diagnoses and medical research by greatly improving the effectiveness and dependability of medical image retrieval.