Brain Tumour Detection with Threat Analysis Using U-Net Architecture

Abstract

Brain tumours are the main reason for 85% - 90% of all the primary CNS (central nervous system) tumours. Nearly 70% to 75% of Brain Tumours are undetected in early stages. The implemented model used Deep CNN and U-Net architecture to reduce this problem. The model includes detecting threat levels with lower resource requirements. The dataset is stored in NifTl-1 format (DICOM), and it uses the NiBabel Library to access the files. The U-Net Architecture using Deep CNN performs Biomedical image segmentation and the Dice Coefficient, Specificity, and Sensitivity are utilized to check the segmentation predictions. This project is an attempt with the goal of scanning a tumour, extracting its threat level, and proposing a model to deliver a more accurate result to determine the affected brain region. It is a project aimed to scan and extract the threat level of the tumour and further propose the model to provide a more accurate and reliable result for determining the affected region in the brain. The research further achieved an excellent accuracy of 96% in detecting affected areas. Also, a comparative study was performed to showcase the efficacy and dependability of a novel approach suggested in the research. Hence, the conducted research can be a success in the medical industry.