Tumor segmentation from brain mr images using resunet, vgg19 unet, attentionunet and vgg16 unet : a comparative study

Abstract

Medical image segmentation, particularly in the context of brain tumor analysis, is a critical task with far-reaching implications for diagnosis, treatment planning, and patient care. This comparative study delves into the efficacy of four distinct deep learning architectures—ResUNet, Vgg19 UNet, AttentionUNet, and Vgg16 UNet for the segmentation of brain tumors from magnetic resonance imaging (MRI) scans. Drawing inspiration from a wealth of prior research, this thesis paper seeks to unravel the nuances of these architectures, their performance characteristics, and their potential contributions to medical imaging. The study leverages a diverse collection of brain MRI datasets and employs a range of preprocessing techniques to ensure robustness and consistency in data representation. ResUNet, a fusion of UNet and residual connections, offers the advantage of capturing intricate tumor boundaries with fine detail. Vgg19 UNet, known for its simplicity and effectiveness, demonstrates its ability to extract features hierarchically, facilitating accurate tumor delineation. Attention U-Net introduces attention mechanisms to enhance the focus on critical regions, while Vgg16 UNet showcases the utility of earlier versions of VGG architectures. Through rigorous experimentation and quantitative analysis, the study systematically evaluates these architectures in terms of segmentation accuracy, computational efficiency, and generalization capability. Insights gleaned from this comparative exploration unveil each architecture's strengths and limitations, paving the way for nuanced decision-making in selecting the appropriate approach based on the task's requirements and available resources. The findings from this study contribute to the ever-evolving landscape of medical image analysis. By synthesizing insights from previous research, this thesis paper not only advances our understanding of tumor segmentation but also provides a comprehensive reference for researchers and practitioners engaged in the challenging realm of medical image analysis. As technology progresses and datasets expand, the lessons drawn from this comparative study are poised to inform the development of innovative solutions that enhance medical diagnostics and improve patient outcomes.