Identification of Tumor using Segmentation & Object Detection: A Comparison and an Analysis

Abnormal cells proliferate in the brain to form brain tumors, which frequently cause death. Localising and segmenting brain tumors using MRI is a difficult task in medical analysis, yet it is necessary for many applications. Accurate segmentation of brain tumor images is required for the identification and treatment of patients with these tumors. Even though they achieved promising segmentation results on the BRATS 2020 dataset, several of them need more testing and training time due to their complicated structure. In order to help doctors identify tumors in their early stages with high accuracy, we have proposed a two-step deep learning method for the diagnosis of three different types of tumors: necrotic, edema, enhancing, as well as brains free of tumors, using MRI images. The metrics used for evaluating the performance of these models are intersection over union, recall, and precision. The research indicates the precision of YOLO model to be 0.97 showcasing the accurate tumor detections, while recall and IoU of UNet segmentation to be 0.99 and 0.83 respectively. In this study, we addressed the issue of inaccuracy and latency in current artificial diagnosis systems by proposing a robust method for classifying brain tumors using deep learning techniques.