Prognostic role of artificial intelligence among patients with hepatocellular cancer: A systematic review

Table 1 Articles focused on the role of artificial intelligence in the prediction of survival

Ref.	Country/region	n	Research question/purpose	Method used	Key findings
Hamamoto et al[12], 1995	Japan	11	ANN for the prediction of survival after HCC resection.	ANN was trained with the data of 54 resected patients and then prospectively used.	The outcomes in the prospective cohort were successfully predicted in all the cases (10 successful, 1 died).
Ho et al[13], 2012	Taiwan	482	To validate the use of ANN model for predicting 1-, 3-, and 5-yr disease-free survival after hepatic resection, and to compare it with LR and decision tree model.	Training set: 80% of the cases; validation set: Remaining 20% of the cases.	The ANN model outperformed the other models in terms of prediction accuracy (AUC for 5-yr disease-free survival: 0.864 vs 0.627-0.736).
Shi et al[14], 2012	Taiwan	22926	ANN model for predicting in-hospital mortality in HCC surgery patients and to compare it with LR models.	This study analyzed administrative claims data obtained from the Taiwan Bureau of National Health Insurance.	Compared to the LR models, the ANN models had a better accuracy rate in 97.28% of cases, and a better ROC curve in 84.67% of cases.
Shi et al[15], 2012	Taiwan	22926	To validate the ANN models for predicting 5-yr mortality in HCC resected patients, and to compare them with LR models.	This study analyzed administrative claims data obtained from the Taiwan Bureau of National Health Insurance.	Compared to the LR models, the ANN models had a better accuracy rate in 96.57% of cases, and a better receiver operating characteristic curves in 88.51% of cases.
Chiu et al[16], 2013	Taiwan	434	To compare significant predictors of mortality for HCC resected patients between ANN and LR models, and to evaluate the predictive accuracy of ANN and LR in different survival year estimation models.	Training set: 80% of the cases; validation set: Remaining 20% of the cases.	The results indicated that ANN had double to triple numbers of significant predictors at 1-, 3-, and 5-yr survival models as compared with LR models. Scores of accuracy, sensitivity, specificity, and AUC using ANN were superior to those of LR.
Qiao et al[17], 2014	China	543; 182; 104	ANN for the prediction of survival in early HCC cases following partial hepatectomy.	Training set: 75% of the cases; internal validation set: Remaining 25% of the cases; external validation set.	In the training cohort, the AUC of the ANN was larger than that of the Cox model (0.855 vs 0.826, P = 0.0115). These findings were confirmed with the internal and external validation cohorts.
Liang et al[18], 2014	Taiwan	83	Use of support vector machine for the development of recurrence predictive models for HCC patients receiving RFA treatment.	Five feature selection methods including genetic algorithm, simulated annealing algorithm, random forests and hybrid methods were utilized.	The developed support vector machine-based predictive models using hybrid methods had averages of the sensitivity, specificity, and AUC as 67%, 86%, and 0.69.
R et al[19], 2019	India	152	To use artificial plant optimization algorithm to select optimal features and parameters of classifiers to improve the effectiveness and efficiency of prediction of HCC recurrence.	Different methods tested.	The sampling based multiple measurement artificial plant optimized random forest classifier with statistical measure showed the best results (balanced accuracy: 0.955).
Shan et al[20], 2019	China	156	Peritumoral radiomics for the prediction of early recurrence after HCC curative resection or ablation.	Training cohort (n = 109) and validation cohort (n = 47). Using CT images, two regions of interest were delineated around the lesion for feature extraction o tumoral radiomics and peritumoral radiomics.	In the validation cohort, the ROC curves, calibration curves and decision curves indicated that the CT-based peritumoral radiomics model had better calibration efficiency and provided greater clinical benefits.

ANN: Artificial neural network; HCC: Hepatocellular carcinoma; AUC: Area under the curve; LR: Logistic regression; RFA: Radiofrequency ablation; CT: Computed tomography; ROC: Receiver operating characteristic.