Optimized Liver Tumor Segmentation: Integrating U-Net with Graph Cut Algorithms

This paper presents an innovative approach to liver tumor segmentation, combining the strengths of U-Net architecture with Graph Cut refinement. This paper focuses on enhancing the precision of segmentation in computed tomography (CT) scans, crucial for accurate diagnosis and treatment planning in liver cancer cases. Our methodology utilized the U-Net model for initial segmentation, owing to its efficiency in handling complex biomedical images. To further refine these results, we integrated the Graph Cut algorithm, which significantly improved the accuracy of tumor boundary delineation. We validated our approach using the LiTS17 dataset, comprising diverse CT scans with varying tumor characteristics. Our results demonstrated a Dice Coefficient of 96.4 and an Accuracy of 99.4, outperforming several existing segmentation methods. This fusion of deep learning and optimization algorithms not only provided superior segmentation results but also highlighted the potential of combining different computational techniques for enhanced performance in medical image analysis.