Integrating Vision Transformers and Autoencoders for Liver Image Segmentation in Medical Imaging

Medical image segmentation is recognized as a critical task in medical imaging, crucial for enabling diagnosis, disease monitoring, and treatment planning. Liver image segmentation is particularly challenging due to the sophisticated structure of the liver and the present of surrounding organs and tissues that obscure clear boundaries. Traditional statistical methods often struggle with scalability, robustness, and adaptability to complex datasets. And convolutional neural network-based models face limitations in capturing global dependencies and contextual information due to their local receptive fields. To overcome these challenges, a novel segmentation framework that integrates a vision transformer with an autoencoder architecture, leveraging self-attention mechanisms was proposed. It was shown that the proposed method can capture global context and long-range dependencies. By focusing on both global and local features, this methodology offers a powerful solution to the challenges in the field of medical image segmentation, showcasing the potential of transformerbased models in advancing the field.