Content-based medical image retrieval using deep learning-based features and hybrid meta-heuristic optimization

Medical imaging is essential to the medical profession because it gives physicians access to crucial data on interior body structures for clinical analysis and treatment decisions that help them identify and cure a wide range of illnesses. A significant collection of medical photos has been created as a result of the rapid increase in medical diagnoses, yet it can be difficult to locate similar medical images within such a large database. This article describes a technique for deep learning-based convolutional neural network (CNN) -based Content-Based Medical Image Retrieval (CBMIR) to deal with this problem as well as Modified Cosine Similarity (MCS)-based matching. The aim of this approach is to enhance the accuracy and efficiency of CBMIR by utilizing the power of deep learning and advanced optimization techniques. The proposed model includes two major phases: (a) the training stage, and (b) the testing stage. In the training stage, the pre-processing, feature extraction, and optimal feature selection process take place. The database images are pre-processed using the Gaussian filter, Contrast Limited Adaptive Histogram Equalization (CLAHE), and Gaussian smoothing. Then, the deep features of database images are extracted using the Inception V3 CNN model and VGG19, respectively. The extracted features are combined, and the optimal features are selected from them. This selection is done through the new Coyote-Moth Optimization Algorithm (CMOA). This CMOA model is the conceptual amalgamation of the standard Moth-flame optimization (MFO) and coyote optimization Algorithm (COA).