Brain Tumor Segmentation using Multimodal MRI and Convolutional Neural Network

Glioblastoma is the most common brain tumor with a high mortality rate. So, detecting the tumorous lesion and segmenting it into its subsets can be helpful to evaluate the grade of the tumor in tracking the therapeutic interventions. Moreover, image segmentation is commonly used for evaluating and visualizing the anatomy of brain tissue in MRI. On the other hand, the convolutional neural network is a network with a deep learning approach and directly learns from data without any feature extraction. In this study, we used a multimodal MRI database containing FLAR, T1 enhanced, and T2 modalities, and a convolutional neural network to segment tumors into whole tumor, core tumor, and necrotic tumor. The results showed accuracy with 85.41% for whole tumor, 90.11% for core tumor, and 79.75% for necrotic tumor. These results showed that using a convolutional neural network is reliable for brain tumor segmentation. Considering this approach used multimodal MRI, this segmentation could be separately done for each tissue.