A Content-Based Medical Image Retrieval System for Lung Diseases Using Mask AttnRCNNpro Segmentation and Hybrid Distance Approach

At present, Content-Based Medical Image Retrieval Systems (CBMIRS) are a novel and potentially useful technology though they lack clinical validation. The study aims to assess how CBMIRS helps in interpretation of chest X-ray (CXR) images in patients who have lung disease. This paper proposes Lung-CBMIR, a new hybrid model that aims to enhance retrieval precision and computational complexity for lung disease X-ray images. The new system combines Mask AttnR-CNNpro, an improved segmentation model that uses attention mechanisms to precisely segment lung areas. Feature extraction is done through Local Binary Patterns (LBP) for texture features, shape descriptors for geometric pattern, and DenseNet+, which utilizes three dense blocks and strategic pooling methods to achieve deep feature extraction. The Bobcat-Fish Hybrid Optimizer (BFHO) method proposed in this paper integrates Bobcat Optimization exploration ability with the exploitation capability of Catch Fish Optimization for optimal selection of features. There is also a novel hybrid distance metric, combining Mahalanobis and Cosine distances, that improves image similarity measurement. Furthermore, rank the images based on their relevance to the query and compile them into a feature vector. Lastly, the DeepCL-Net classifier, which is a combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, facilitates effective classification of lung illnesses like pneumonia, infiltrates, and lung nodules. The proposed Lung-CBMIR system is found to attain an accuracy of 98.75%, F1-score of 98.13%, and MCC of 0.9801, better than state-of-the-art models like CNN-AE of 95.58% and VGG-19 of 96.81%. The results confirm that the proposed system greatly improves retrieval accuracy, lowers computational complexity, and yields a strong tool for lung disease diagnosis in CBMIR tasks. The abbreviation and their concern description are manifested in Table 1.