Comparative Study of Encoder-Based U-Net Models for Brain Tumor MRI Segmentation with Tumor Analysis and Visualization

Resumen

Brain tumor segmentation from magnetic resonance imaging (MRI) plays a crucial role in clinical diagnosis, treatment planning, and disease monitoring. However, manual delineation of tumor regions is time-consuming and subject to variability among clinicians. Recent advances in deep learning, particularly convolutional neural networks (CNNs), have significantly improved the performance of medical image segmentation [5,13]. Among these architectures, U-Net is widely adopted due to its encoder–decoder structure [1,3] that effectively captures both spatial and contextual features. This study presents a comparative analysis of encoder-based U-Net architectures for auto-mated brain tumor segmentation using multi-modal MRI images from the BraTS dataset. Three models are evaluated: the standard U-Net, U-Net with a VGG16 encoder, and U-Net with a ResNet32 encoder. In addition, two preprocessing approaches, Min–Max normalization and Z-score normalization, are applied to analyze their impact on segmentation performance. The models are evaluated using loss, accuracy, and Intersection over Union (IoU). Experimental results show that incorporating encoder backbones im-proves segmentation performance compared to the baseline U-Net model. The ResNet32-based U-Net achieved the highest training IoU, while the VGG16-based U-Net with Z-score normalization achieved the best validation IoU, indicating improved performance on unseen MRI images. The frame-work additionally supports automated tumor analysis and visualization to facilitate clinical interpretation of segmentation results.