Brain Tumor Classification Using Ensemble Deep Learning Model With Content Based Medical Image Retrieval

Abstract

Brain tumors are among the world's most dangerous causes of death. Magnetic Resonance Imaging (MRI) are non-invasive imaging techniques that neuro radiologists utilize extensively to diagnose Glioma grades. Therapy planning is negatively impacted by the laborious and extremely subjective process of diagnosing brain tumors from radiological imaging, which is vulnerable to intra- and inter-observer variability. If pertinent image graphs are obtained from a large medical image collection, better medical care related to comparable previous situations can be given. Systems for retrieving images based on content are effective tools for handling large datasets. Convolutional Neural Network-based feature extraction techniques can be used in a CBIR system to efficiently automate the categorization and retrieval of comparable pathological images. This research work proposed an ensemble deep learning (DL) based classifications and medical image retrievals for MRI brain tumor images using content-based methods. In the initial stage, content-based image retrieval (CBIR) selects query or template image features and then computes measures of similarity for detecting tumors. During the image pre-processing step, an Enhanced Bilateral Filter (EBF) is used which is an improved version of the bilateral filter used in image processing to preserve edges while reducing noise. Then the image Normalization process is carried out using improved Weibull Cumulative Distribution Function (IWCDF). Following this stage, the tumor portion is identified by applying the Watershed Segmentation method to segment the images. Next, the Enhanced Whale Optimization Algorithm (EWOA) is used to extract features. The enhanced cuckoo search (ICSA) algorithm is then used to carry out the feature selection. At the indexing step, the R*-tree approach is used to improve search efficiency, and Mahalanobis Distance with relevance feedback is used to compute the similarity measure during that phase. Lastly, a model called Ensemble Deep Learning (EDL) is suggested for brain tumor diagnosis. Experiments have used brain MRI data from database of medical images. The suggested approach improves accuracy, recall, and retrieval time.