Preoperative CT Evaluation of Living Liver Donors Using Artificial Intelligence: Diagnostic Accuracy for Fatty Liver and Fat Fraction Quantification

With the increasing number of patients with fatty liver worldwide, living donor liver transplantation has emerged as a solution to the shortage of organs from deceased donors. Since the donor’s liver fat content significantly affects the safety and survival of both the donor and recipient, diagnosing fatty liver in the donor is crucial. Recently, the use of deep learning-based artificial intelligence (AI) in medical imaging has been on the rise, and our institution also extensively utilizes AI in imaging diagnosis. Therefore, this study aimed to analyze preoperative CT images of living donor liver transplantation using AI to measure the hepatic parenchyma's hounsfield unit (HU) values and fat fraction. We compared these measurements with traditional HU value measurements and MRI quantitative Dixon technique-based fat fraction. The study was conducted on 101 patients who underwent both CT and MRI within a month before liver transplantation. We found a strong positive linear correlation between the average HU values of the hepatic parenchyma measured by AI and those measured by radiologists, as well as between fat fraction measured by AI and the MRI quantitative Dixon technique.(HU: r=0.971, R²=0.944; Fat Fraction: r=0.815, R²=0.666) A paired t-test revealed statistically significant differences in the average HU values,(p<0.001) and fat fraction showed potential statistical significance.(p=0.05) Bland-Altman analysis showed that most differences were within clinically acceptable limits. While The AI cannot fully replace current methods, it may serve as a supplementary tool through further calibration and research.