A Deep Learning Approach for Automatic Liver Segmentation: Attention U-Net with ASPP and CBAM Integration
Mohamad O. Diab, Nadine Al Masri, Razan Al Moghrabi, Ranim El Hamwi, Zeinab Abbas, Rama Younis, Mostafa Daher, Racha Soubra · 2025
Liver segmentation is a critical step in medical image analysis for identifying anatomical structures and detecting potential abnormalities. However, this task remains challenging due to the liver's complex shape, its proximity to other abdominal organs, and the variations in intensity within medical images. To address these challenges, we propose an advanced deep learning model for automatic liver segmentation in Computed Tomography (CT) images, aiming to assist healthcare professionals with rapid diagnosis and treatment planning. Our proposed model is based on the U-Net architecture combined with three key enhancements: an Attention Gate System (AGS), a Convolutional Block Attention Module (CBAM), and an Atrous Spatial Pyramid Pooling (ASPP). This combination enhances features extraction and enables the model to capture multi-scale contextual information. The model was trained and evaluated using the Liver Tumor Segmentation (LiTS) dataset, with a total of 131 3D abdominal CT scans. The achieved Dice Score of 98.78% and Intensity over Union (IoU) of 98.53% on the testing set demonstrate strong performance of our approach and its promising potential for clinical deployment.