Predicting post-hepatectomy liver failure in patients with hepatocellular carcinoma: nomograms based on deep learning
Boryeong Jeong1,2, Subin Heo1,3, Seung Soo Lee4
1Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea.
Objectives:
This study aimed to develop nomograms for predicting post-hepatectomy liver failure (PHLF) in patients with hepatocellular carcinoma (HCC), using deep learning analysis of Gadoxetic acid-enhanced hepatobiliary (HBP) MRI.
Methods:
This retrospective study analyzed patients who underwent gadoxetic acid-enhanced MRI and hepatectomy for HCC between 2016 and 2020 at two referral centers. Using a deep learning algorithm, volumes and signal intensities of whole non-tumor liver, expected remnant liver, and spleen were measured on HBP images. Two multivariable logistic regression models were formulated to predict PHLF, defined and graded by the International Study Group of Liver Surgery: one based on whole non-tumor liver measurements (whole liver model) and the other on expected remnant liver measurements (remnant liver model). The models were presented as nomograms and a web-based calculator. Discrimination performance was evaluated using the area under the receiver operating curve (AUC), with internal validation through 1000-fold bootstrapping.
Results:
The study included 1760 patients (1395 male; mean age ± standard deviation, 60 ± 10 years), with 137 (7.8%) developing PHLF. Nomogram predictors included sex, gamma-glutamyl transpeptidase, prothrombin time international normalized ratio, platelets, extent of liver resection, and MRI variables derived from the liver volume, liver-to-spleen signal intensity ratio, and spleen volume. The whole liver and the remnant liver nomograms demonstrated strong predictive performance for PHLF (optimism-corrected AUC of 0.78 and 0.81, respectively) and symptomatic (grades B and C) PHLF (optimism-corrected AUC of 0.81 and 0.84, respectively).
Conclusion:
Nomograms based on deep learning analysis of gadoxetic acid-enhanced HBP images accurately stratify the risk of PHLF.
Key Points:
Question Can PHLF be predicted by integrating clinical and MRI-derived volume and functional variables through deep learning analysis of gadoxetic acid-enhanced MRI? Findings Whole liver and remnant liver nomograms demonstrated strong predictive performance for PHLF with the optimism-corrected area under the curve of 0.78 and 0.81, respectively. Clinical relevance These nomograms can effectively stratify the risk of PHLF, providing a valuable tool for treatment decisions regarding hepatectomy for HCC.
