Improved U-Net architecture with VGG-16 for brain tumor segmentation

Automated assessment and segmentation of Brain MRI images facilitate towards detection of neurological diseases and disorders. In this paper, we propose an improved U-Net with VGG-16 to segment Brain MRI images and identify region-of-interest (tumor cells). We compare results of improved U-Net with a custom-designed U-Net architecture by analyzing the TCGA-LGG dataset (3929 images) from the TCI archive, and achieve pixel accuracies of 0.994 and 0.9975 from basic U-Net and improved U-Net architectures, respectively. Our results outperformed common CNN-based state-of-the-art works.

Automated summary

Automated assessment and segmentation of Brain MRI images using an improved U-Net with VGG-16 can effectively detect neurological diseases and disorders, outperforming common CNN-based state-of-the-art works.