An Ensemble Deep Learning and Modified Wild Horse Optimization based Pixel Selection for Secured Content-based Medical Image Retrieval

Due to the restricted capacity for focus on the Human Visual System, clinical specialists may ignore minor lesions in different Medical Images (MI), making MI-based diagnosis a difficult process that could negatively impact clinical therapy. On the other hand, this issue could be handled through an effective Content-Based MI Retrieval (CBMIR) approach to examine comparable cases from the prior medical database. The retrieval engine utilizes Feature Vectors (FV), which are High-Level (HL) image representations maintained by a CBIR structure, to match and rank Query Images (QI) based on similarity. In this research work, an Ensemble Learning (EL) and Modified Wild Horse Optimization (MWHO) based pixel selection with Henon chaotic map (MWHO-HCE) is developed. Next, the image retrieval process includes a series of processes, namely Densenet-121-based feature extraction, Enhanced Sparrow Search Optimization Algorithm (ESSO) based Hyper Parameter (HP) optimizer, and Ensemble deep learning classifier (EDLC) such as Resnet50 inspectionV2 and Lenet are used to classify the lung diseases for the given dataset. Utilizing the elcap test image database, the suggested technique's effectiveness is evaluated. The study outcomes indicate the effectiveness of the suggested strategy for MIR (MI Retrieval).