Enhancing Brain Stroke Diagnosis with MEDIA-BTS: A Deep Learning Approach for Efficient Medical Image Retrieval
Efficient brain stroke diagnosis relies on the accurate retrieval of relevant medical images from extensive databases to aid in clinical decision-making. Conventional methods of medical image retrieval face challenges such as variations in lesion size, differences in imaging modalities, and the difficulty of linking low-level image features to their medical significance. To overcome these obstacles, a new deep learning-based approach called MEDIA-BTS has been proposed. This method utilizes an Improved Adaptive Wiener Filter (IAWF) to preprocess input and query MRI images, enhancing image quality by suppressing noise and preserving important details. The Dual-attention based LinkNet (Duo-LinkNet) architecture integrates spatial and channel attention modules to emphasize critical regions and features, improving feature representation and retrieval accuracy. Butterfly Mating Optimization (BMO) is employed to compute similarity measures for efficient image retrieval, achieving an average precision retrieval of 99.15% and a retrieval time of 2.50s. The MEDIA-BTS model outperforms existing methods, enhancing overall accuracy by significant margins.