Automated Three-Dimensional Liver Reconstruction with Artificial Intelligence for Virtual Hepatectomy.

ABSTRACT

OBJECTIVE: To validate the newly developed artificial intelligence (AI)-assisted simulation by evaluating the speed of three-dimensional (3D) reconstruction and accuracy of segmental volumetry among patients with liver tumors. BACKGROUND: AI with a deep learning algorithm based on healthy liver computer tomography images has been developed to assist three-dimensional liver reconstruction in virtual hepatectomy. METHODS: 3D reconstruction using hepatic computed tomography scans of 144 patients with liver tumors was performed using two different versions of Synapse 3D (Fujifilm, Tokyo, Japan) the manual method based on the tracking algorithm and the AI-assisted method. Processing time to 3D reconstruction and volumetry of whole liver, tumor-containing and tumor-free segments were compared. RESULTS: The median total liver volume and the volume ratio of a tumor-containing and a tumor-free segment were calculated as 1035 mL, 9.4%, and 9.8% by the AI-assisted reconstruction, whereas 1120 mL, 9.9%, and 9.3% by the manual reconstruction method. The mean absolute deviations were 16.7 mL and 1.0% in the tumor-containing segment and 15.5 mL and 1.0% in the tumor-free segment. The processing time was shorter in the AI-assisted (2.1 vs. 35.0 min; p < 0.001). CONCLUSIONS: The virtual hepatectomy, including functional liver volumetric analysis, using the 3D liver models reconstructed by the AI-assisted methods, was reliable for the practical planning of liver tumor resections.