3D Brain Tumor Segmentation with U-Net Network using Public Kaggle Dataset

This paper aims to implement and experiment with a deep learning model, U-Net, for effectively segmenting the 3D-brain tumor images. It helps to identify glioblastoma in MRI brain images. Manual investigation of MRI images does not provide exact information about the abnormalities in the images. Thus, various existing methods have proposed medical image processing methods, which include custom methods, traditional image segmentation methods, and classifiers. However, those methods did not provide high accuracy in prediction with lesser complexity. Hence, this paper has aimed to implement U-Net architecture for segmenting and classifying MRI brain images. It helps in obtaining efficient performance regarding brain tumor segmentation, along with detailing the stage in which the diseased person currently has after segmentation. The proposed methodology has been experimented within MATLAB software, and the efficiency has been verified. From the experiment , it is identified that the proposed model provides better performance of 98% Accuracy, 95% Precision, and 96.4% Recall. Furthermore, the size of the tumor was successfully able to be intimated by outlining it, and the stage of the Brain Tumor (Gliomas) was also intimated with the warning message to any healthcare professional. Finally, the comparative study of existing methods using Morphological operations and our proposed 3D brain tumor segmentation methodology was made and presented.