MEDIA-BTS: Medical Image Retrieval Using Butterfly Mating Optimization for Brain Stroke Diagnosis

Brain stroke diagnosis requires efficient retrieval of relevant medical images from large databases to support accurate clinical decision-making. Traditional medical image retrieval (MIR) methods often struggle with variations in lesion size, differences across imaging modalities and the challenge of relating low-level image features to their actual medical significance. To address these challenges, this paper proposes a novel deep learning-based MEDIA-BTS for retrieve relevant brain MRI images based on query image. Improved Adaptive Wiener Filter (IAWF) is used to pre-process the input and query MRI images by adaptively adjusting its filtering behavior based on local image statistics. This approach effectively suppresses noise while preserving important edges and fine anatomical details, thereby enhancing image quality for subsequent analysis. Dual-attention based LinkNet ( Duo-LinkNet) employs an encoder–decoder architecture with skip connections to preserve spatial details. It integrates spatial and channel attention modules to emphasize important regions and features. This dual-attention mechanism enhances feature representation and emphasizes clinically significant regions, improving the robustness and accuracy of retrieval. Butterfly Mating Optimization (BMO) is applied to the extracted features to compute similarity measures for accurate and efficient image retrieval. As a nature-inspired optimization technique, BMO mimics butterfly mating behavior to reduce computational complexity while retrieving images most relevant to the query. The proposed MEDIA-BTS achieves the Average precision retrieval 99.15% and retrieval time of 2.50s. The MEDIA-BTS model improves the overall accuracy by 26.95%, 1.63%, 1.55% and 3.63% better than CBAM-MA, Inceptionv3, GWO-SVM and DNN respectively.