Automated Liver Segmentation Using 2D U-Net: A Deep Learning Approach for MRI Analysis and Surgical Planning

Abstract

The liver's complexity in structure and function makes it susceptible to various forms of damage, highlighting the need for prompt and accurate diagnosis of liver tumors to ensure timely medical intervention. Image processing has emerged as a valuable tool for diagnostic support, especially in complex liver surgeries. This study presents a deep learning-based approach for automated liver segmentation on magnetic resonance imaging (MRI) using a U-Net model. Our approach is designed to enhance preoperative planning by improving image analysis and aiding surgical teams in reducing potential complications. We utilized a public MRI dataset of 310 liver image series, representing diverse clinical cases and contrast types. Data augmentation techniques were applied to increase the model's robustness. After selecting the most suitable contrast type, we trained the U-Net model and evaluated its performance using metrics focused on segmentation accuracy. The model achieved a Dice Similarity Coefficient (DSC) of 94.37% and a 95th Percentile Hausdorff Distance (HD95) of 3.99 mm, demonstrating high precision in liver segmentations. These results emphasize the potential of this model as a reliable, MRI-based tool for clinical support in liver surgery planning and risk management.