Artificial Intelligence in hepatology, liver surgery and transplantation: Emerging applications and frontiers of research

Table 1 Review of articles where artificial intelligence has been studied in the context of non-alcoholic liver disease

Ref.	Dataset	Number	ML algorithms	Problem	Performance measures
Byra et al[15], 2018	Department of Internal Medicine, Hypertension and Vascular Diseases, Medical University of Warsaw, Poland	55	Deep CNN	Automatically diagnose the amount of fat in the liver from US images	AUROC, Delong statistical test, lasso regression method, Spearman correlation coefficient, Meng test
Perveen et al[16], 2018	CPCSSN	667907	Decision tree	Classification, NAFLD progression risk	Micro- and Macro-average of Precision, Recall and F-measure, MCC, AUROC
Ma et al[17], 2018	First Affiliated Hospital, College of Medicine, Zhejiang University, China	10508	Several, Weka open source software	Classification, feature selection	Accuracy, specificity, precision, recall (i.e. sensitivity), and the F-measure
Vanderbeck et al[18], 2014	Medical College of Wisconsin, Milwaukee, United States	59	SVM	Automated assessment of histological features of NAFLD	Precision rate, recall rate, and AUROC
Meffert et al[68], 2014	SHIP	4222	Boosting algorithm, discrimination and calibration plots	Scoring system for hepatic steatosis risk	Discrimination (AUROC) and calibration
Sowa et al[69], 2014	University Hospital Essen	82	Logistic regression, decision trees, SVM, RF	Distinguish NAFLD from ALD	Sensitivity, specificity, and accuracy
Kuppili et al[70], 2017	Instituto Superior Tecnico, University of Lisbon, Portugal	63	Extreme Learning Machine- SLFFNN	Stratification of FLD disease in US liver images	AUROC, reliability and stability analysis
Sorino et al[71], 2020	MICOL cohort	2970	SVM	Stratify NAFLD risk to reduce need for imaging	Accuracy, variance, calculated confidence limits (95%), the weight of each model (as a %) and the number of ultrasound examinations it could avoid
Wu et al[72], 2019	New Taipei City Municipal Hospital Banqiao Branch	577	ANN, NB, RF, LR	Diagnosis and risk stratification in NAFLD	Accuracy, sensitivity, specificity

ALD: Alcoholic liver disease; ANN: Artificial neural network; AUROC: Area under the receiver operating characteristic; CNN: Convolutional neural network; CPCSSN: Canadian Primary Care Sentinel Surveillance Network; FLD: Fatty liver disease; LR: Logistic regression; MCC: Matthews correlation coefficient; MICOL: Multi-centre Italian study on cholelithiasis; ML: Machine learning; NAFLD: Non-alcoholic fatty liver disease; NB: Naïve Bayes; RF: Random forest; SHIP: Study of Health in Pomerania; SLFFNN: Single-layer feed-forward neural network; SVM: Support vector machine; US: Ultrasound.

Table 2 Review of recently published studies where artificial intelligence-based algorithms have been applied to liver transplantation

Ref.	Dataset	Number	ML algorithms	Problem	Performance measures
Bertsimas et al[62], 2019	STAR dataset	-	OCT	Predict 3 mo waitlist mortality-OPOM	ROC curve
Cruz-Ramírez et al[63], 2013	Spanish multi-center study	-	Radial basis function NN	Improve donor-recipient matching using rule-based allocation—MPENSGA 2 algorithm	Accuracy, minimum sensitivity, ROC curve, RMSE, Cohen’s kappa
Briceño et al[64], 2014	Spanish multi-center study	1003	Neural Net Evolutionary Programming	Improve equity in donor-recipient matching	Multiple regression analysis, simple logistic regression analysis, ROC curve
Ayllón et al[73], 2018	King’s College Hospital,United Kingdom + MADR-E, Spain	1437	ANN	Classification, end-point (3 mo, 1 yr)	ROC curve
Wadhwani et al[74], 2019	UNOS	1482	RF	Classification, end-point (3 yr)	Chi-square test, t-test, Wilcoxon rank sum test
Dorado-Moreno et al[75], 2017	King’s College Hospital, United Kingdom + MADR-E, Spain	1492	Ordinal ANN	Ordinal classification, fourclasses	MAE and the MZE, accuracy, GMS, AMAE
Guijo-Rubio et al[76], 2019	UNOS	39095	Cox, SVM, GB	Survival time	C-index, ROC curve, concordance index ipcw
Lee et al[77], 2018	Seoul National University Hospital	1211	Several ML methods compared, GBM found to be best	Prediction of AKI after liver transplant	ROC curve, accuracy
Lau et al[78], 2017	Austin Hospital, Melbourne, Australia	180	RF, ANN, logistic regression	Predict 30-d risk of graft failure	ROC curve

AKI: Acute kidney injury; AMAE: Average mean absolute error; ANN: Artificial neural network; c-index: Concordance index; GB: Gradient boosting; GBM: Gradient boosting machine; GMS: Geometric mean of the sensitivities; MADR-E: Model for Allocation of Donor and Recipient in España; MAE: Mean absolute error; MPENSG-A: Memetic Pareto evolutionary non-dominated sorting genetic algorithm; ML: Machine learning; MZE: Mean zero-one error; NN: Neural network; OCT: Optimal classification tree; OPOM: Optimized prediction of mortality; RF: Random forest; RMSE: Root mean squared error; ROC: Receiver operating characteristic; STAR: Standard Transplant Analysis and Research; SVM: Support vector machine; UNOS: United Network for Organ Sharing.