Utilizing Customized 3D U-Net Framework for the Classification and Segmentation of Multi-regional Brain Tumors in Volumetric MRI Images

Precise diagnosis and treatment planning depend heavily on the segmentation and classification of brain tumors in medical pictures. The diagnostic and therapeutic challenges posed by brain tumors necessitate innovative solutions within the domain of medical imaging. This research endeavors to address this pressing need through the introduction of a pioneering approach to multi-regional tumor classification and segmentation. Utilizing volumetric MRI data, our study introduces a bespoke 3D U-Net framework meticulously tailored to the intricacies of brain tumor analysis. Our framework offers holistic solutions to two foundational objectives: classification and segmentation. In the classification facet, we strive to discriminate between various brain tumor types. Concurrently, our segmentation component aspires to precisely delineate tumor boundaries and partition distinct tumor regions. We used the BraTS 2020 dataset, and dice and Jaccard coefficient functions were employed in the model optimization process. The custom-tailored architecture of 3D U-Net, purpose-built for the nuances of brain tumor imaging, ensures resilient performance while preserving vital spatial information essential for accurate tumor delineation. Leveraging deep learning methodologies and a comprehensive dataset of volumetric MRI scans, our model showcases outstanding accuracy in classifying diverse tumor types and in the precise segmentation of tumor regions with a dice score are 88.1%, 79.4%, and 76.0% for WT dice, TC dice and ET dice. And Jaccard Scores are 79.4% for WT Jaccard, 68.2% for TC Jaccard and 64.3% for ET Jaccard. The proposed framework holds the potential to alleviate subjectivity in diagnostic processes, enhance the precision of treatment planning, and ultimately ameliorate patient outcomes.