An Efficient Deep Learning Approach for Brain Tumor Segmentation using 3D Convolutional Neural Network

In medical application, deep learning-based biomedical semantic segmentation has provided state-of-the-art results and proven to be more efficient than manual segmentation by human interaction in various cases. One of the most popular architectures for biomedical segmentation is U-Net. In this paper, a convolutional neural architecture based on 3D U-Net but with fewer parameters and lower computational cost is used for the segmentation of brain tumors. The proposed model is able to maintain a very efficient performance and provides better results in some cases compared to conventional U-Net, while reducing memory usage, training time and inference time. The model is trained on the BraTS 2021 dataset and is able to achieve Dice scores of 0.9105, 0.884 and 0.8254 on Whole Tumor, Tumor Core and Enhancing-Tumor on the testing dataset.