A CONTENT BASED MEDICAL IMAGE RETRIEVAL WITH HYBRID FEATURE EXTRACTION AND DEEP LEARNING MODEL FOR BRAIN TUMOR CLASSIFICATION

Abstract

Abstract: When detected early, brain tumors (BT) are also frequently treatable conditions. The accuracy with which the abnormality in the query image is recognized and determines the diagnosis of the condition. Utilizing a combination of Feature Extraction (FE) and Similarity Matching (SM) techniques, computer-aided automated systems like Content-Based Medical Image Retrieval (CBIR) approach are used to retrieve Query-Based (QB) images from a huge database. For the BT to be effectively treated, an accurate diagnosis is essential. In order to help the radiologist diagnose the BT, this work suggests an intelligent CBIR system that recovers similar pathology carrying Magnetic Resonance Images (MRI) of the brain from a medical database. Since images within the same disease class differ in terms of severity, density, and other features, a single Feature Vector (FV) will not be very effective in identifying similar images in the medical domain. A two-level hierarchical CBMIR system was suggested in this study for addressing this issue. It finds the most similar images in the specified class after first classifying the Query Image (QI) of a BT as benign or malignant. During the image preprocessing step, image Normalization, and Noise Reduction (NR) is carried out. After this step, the image segmentation using WatersShed (WS) method is employed for the identification of the tumor part in MRI image and FE using Hybrid Fuzzy (HF) Dove Swarm Optimization Algorithm (DSOA). Then the Feature Selection (FS) is performed by making use of the modified BAT (mBAT) algorithm. At last, an effective method for BT diagnosis is suggested a Convolutional Neural Network (NN) (CNN). Brain MRI data from a Medical imaging (MI) database have been used in experiments. By increasing precision, recall, and retrieval time, the suggested method produces effective retrieval outcomes.