Role of artificial intelligence in hepatobiliary and pancreatic surgery

Table 1 Summary of the studies included in the review evaluating the role of artificial intelligence in hepatobiliary and pancreatic surgery

Ref.	Aim	No. of patients	Outcome
Preoperative imaging
Fang et al[18]	To compare the surgical outcomes of pre-operative planning based on 3D assisted surgery for HCC	116	Shorter operation time (P = 0.028), and reduced complications (P = 0.048) among surgeries performed based on 3D planning
Mise et al[30]	To assess how pre-operative VH influences the outcomes of liver surgery	1194	Better post-operative oncological outcomes for those in the VH group (P = 0.04)
Fang et al[33]	To assess the resectability of pancreatic and periampullary tumours by 3D visualization system	80	PPV, NPV, sensitivity, specificity, accuracy for resectability was 100% and was better than CT angiography (P < 0.05)
Intra-operative use
Okamoto et al[46]	To evaluate the utility of AR-based navigation surgery for pancreatectomy	19	Surface-rendering image corresponded to that of the actual organ
			Allowed safe dissection while preserving the adjacent vessels or organs
Ntourakis et al[49]	To investigate the potential of AR-based navigation to help locate and resect colorectal liver metastases	03	Allowed detection of all the lesions
Buchs et al[65]	To evaluate Stereotactic navigation technology for targeting hepatic tumors during robotic liver surgery	02	The augmented endoscopic view allows accurate assessment of resection margin and allowed better identification of vascular and biliary structures during parenchymal transection
Post-operative management and follow-up
Merath et al[71]	To assess ML algorithm to predict patient risk of developing complications following liver, pancreatic or colorectal surgery	15, 657	Good predictability of post-operative complication with C-statistic of 0.74, outperforming the ASA (0.58) and ACS-surgical risk (0.71) calculators
Mai et al[73]	To establish and validate an ANN model to predict severe PHLF in patients with HCC following hemi hepatectomy	357	The ANN model resulted in AUROC of 0.880 for the development set of and 0.876 for the validation set in predicting severe PHLF
Zhou et al[80]	To develop a CT-based radiomic signature and assess its ability to preoperatively predict the early recurrence of HCC	215	Adding a radiomics signature into conventional clinical variables can significantly improve the accuracy of the preoperative model in predicting early recurrence (P = 0.01)
Banerjee et al[82]	RVI was assessed for its ability to predict MVI and outcomes in patients with HCC who underwent surgical resection or liver transplant		The diagnostic accuracy, sensitivity, and specificity of RVI in predicting MVI was 89%, 76% and 94%, respectively. Positive RVI score was associated with lower OS (P < 0.001) and RFS (P = 0.001)

3D: 3-dimensional; HCC: Hepatocellular carcinoma; VH: Virtual hepatectomy; PPV: Positive predictive value; NPV: Negative predictive value; AR: augmented reality; ML: Machine learning; ANN: Artificial neural network; RVI: Radio genomic venous imaging; MVI: Microvascular invasion; PHLF: Post-hepatectomy liver failure; ASA: American Society of Anaesthesiologists; ACS: American College of Surgeons.