Exploring CNN-Based Deep Learning Architectures for Automatic Brain Tumor Segmentation from MRI Scans

Brain tumor localization and segmentation from magnetic resonance imaging (MRI) are challenging yet crucial tasks for a range of medical analysis applications. Early and accurate segmentation can significantly improve treatment options and increase patients' chances of survival. Convolutional Neural Networks (CNNs) have recently shown remarkable performance in image segmentation tasks within computer vision. This research focuses on exploring these segmentation models to detect and localize brain tumors. It presents a comparative study based on the exploration some advanced CNN-based deep learning models, in particular UNet architecture and its variants, such as ResUNet, UNet++, UNet3+, and Attention-UNet, using a LGG dataset comprising MRI scans. Experimental results have shown that among these models, the Attention-UNet outperformed the others across all performance metrics, achieving a mean Intersection over Union (mIoU) of 70.01%, an IoU of 82.55%, and a Dice coefficient of 90.43%. They also underscore the potential of UNet-based approach in real-world diagnostic scenarios, where rapid testing for brain tumors could be highly beneficial.