MRI Image Segmentation using CNN and Deep Neural Network

Accurate identification and delineation of tumors in brain from MRI images which is multimodal is crucial for tracking disease progression, particularly in the case of gliomas, which are both malignant and highly diverse. To achieve precise segmentation of these tumors into their intra-tumoral subtypes, modern approaches favor deep learning algorithms over conventional methods. Among these, CNNs have emerged as a powerful tool in the field of biomedical and significantly advancing the accuracy of medical image segmentation. This study proposes a novel ensemble method that combines two distinct segmentation networks: like a 3D CNN and a U-Net. This combination, while conceptually straightforward, yields superior and more precise segmentation results. These two models individually trained using the BraTS-19 challenge dataset, resulting in segmentation maps that exhibited notable differences in terms of the identified tumor sub-regions. By skillfully assembling these diverse outputs, we achieved a final prediction that greatly improved the accuracy of segmented image.