Brain Tumor Segmentation Using Deep Neural Networks: A Comparative Study

Abstract

The research presents brain tumor segmentation from medical images like MRI scans using various deep neural networks. Tumors can arise anywhere in the brain and can be of different contrast, form, and magnitude. The proposed networks are designed for glioblastoma (high-grade and low-grade) tumors in MRI scans. The study explores various architectures designed for medical data image segmentation. The two-path CNN architecture implemented in the study exploits local and more global contextual features. The two-path CNN architecture was extended using three cascading architectures (Input, Local, and MF). Also, the researchers used the popular semantic segmentation architecture models, U-net, specially designed for medical image segmentation. Finally, the study compared the Cascaded and the U-net performance based on F1 score and Dice loss. It was concluded that the U-net architecture performed better than Cascading architecture and delivered a more precise boundary for the target tumor in an MRI scan. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.