Brain Magnetic Resonance Imaging Tumor Segmentation

Abstract

This research project explores an advanced diagnostic approach for brain tumors using MRI images, employing a synergistic combination of classification and segmentation techniques. The methodology integrates two sophisticated neural network architectures: ResNet50 for classification and ResUNet for segmentation. The classification model utilizes the Adam optimizer and Binary Crossentropy loss function, focusing on accuracy as the primary evaluation metric. Simultaneously, the segmentation model is optimized using the Adam optimizer and the novel Focal Tversky loss function, with the Tversky coefficient as its metric. This dual-model strategy not only identifies the presence of brain tumors but also delineates their precise spatial distribution, offering a comprehensive analysis. The research's novelty lies in its integrative approach, enhancing diagnostic precision and aiding in effective treatment planning. This project is significant in the field of medical imaging analysis, particularly in neuroimaging, as it promises to refine diagnostic processes and support medical practitioners in making informed decisions, thereby potentially improving patient outcomes in neuro-oncology.