Automatic liver Couinaud segmentation from computed tomography scans with a gradient-enhanced hierarchical cascade deep learning network

Le résumé fourni par la source

Background Couinaud segmentation is crucial to understand the anatomical and functional structures of the liver for precise surgical planning. Couinaud segmentation is a challenging task in clinical practice due to the high similarity of intensity values between different liver segments and the high complexity of the vascular and biliary structures of the liver. Methods We develop a comprehensive method for Couinaud segmentation based on our previously developed G-UNETR++, which incorporates gradient-enhanced encoders to effectively capture 3D geometric features. Additionally, we propose a novel post-processing method, a linear polynomial-based segment transformation method to avoid the leak of a segment to its adjacent segment to improve Couinaud segmentation results. Results Experimental results on the MSD 08 dataset showed that our approach achieved better performance, with an average dice score of 0.9752 for liver segmentation and 0.8417 for Couinaud segmentation, compared to state-of-the-art methods. Conclusion The proposed method is thus effective for Couinaud segmentation for precise liver surgical planning in clinical applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.