Artificial intelligence in pancreatic disease

Table 1 Recent publications using artificial intelligence and radiomics in pancreatic disease

Ref.	Year	Disease	Number	Training/testing	Modality	Design	Feature selection	Results
PADC detection
Chu et al[18]	2019	PDAC vs normal	190:190	255/125	CT	Retrospective	RF	Accuracy: 99.2%; AUC: 0.99
Liu et al[19]	2020	PDAC vs normal	370:320	PDAC: 295/256; Normal: 75/64	CT	Retrospective	CNN	Accuracy: 98.6-98.9%; AUC: 0.997-0.999
Li et al[21]	2020	LN metastasis	159	118/41	CT	Retrospective	LASSO	Combined model; AUC: Training/test = 0.944/0.912
Bian et al[22]	2019	LN metastasis	225	-	CT	Retrospective	LASSO	The arterial rad-score is associated with the risk of LN metastasis.
Hui et al[25]	2020	R0 vs R1 after PD	34:52	-	CT	Retrospective	SVM	AUC: 0.8614 Accuracy: 84.88%
Bian et al[26]	2020	SMV margin (R0 vs R1) after PD	127:54	-	CT	Retrospective	LASSO	AUC: 0.75
Zhang et al[28]	2018	POPF after PD	117	80/37	CT	Retrospective	LASSO	AUC: Training/test 0.8248/0.7609
Xie et al[32]	2020	PFS and OS	220	147/73	CT	Retrospective	LASSO	Rad-score is better than clinical model and TNM system
Cozzi et al[33]	2019	OS and local control after SBRT	100	60/40	CT	Retrospective	Elastic net regularization, Cox regression models	Identify low and high-risk groups
IPMN
Chakraborty et al[41]	2018	Low risk vs high risk	103		CT	Retrospective	RF, SVM	AUC: 0.77
Corral et al[42]	2019	Normal pancreas, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma	139 (31:48:20:40)	-	MRI	Retrospective	Deep learning	AUC: 0.78
PNET
Liang et al[49]	2019	Grade 1 vs 2/3	137	86/51	CT	Retrospective	LASSO	AUC: Training/test = 0.907/0.891
Gu et al[50]	2019	Grade 1 vs 2/3	138	104/34	CT	Retrospective	MRMR, RF	AUC: Training/test = 0.974/0.902
Bian et al[51]	2020	Grade 1 vs 2/3 (non-functional)	139	97/42	MRI	Retrospective	LASSO and LDA	AUC: Training/test = 0.851/0.736
Other pancreatic lesions
Park et al[54]	2020	AIP vs PDAC	85: 93	60/29: 60/33	CT	Retrospective	RF	Accuracy: 95.2%; AUC: 0.975
Zhang et al[55]	2019	AIP vs PDAC	45: 66	-	PET/CT	Retrospective	RF, adaptive boosting, SVM	Accuracy: 85%; AUC: 0.93
Ren et al[56]	2019	MFP vs PDAC	79: 30	69/40	CT	Retrospective	Mann-Whitney U test, MRMR	AUC: 0.98
Mashayekhi et al[57]	2020	Functional abdominal pain, recurrent acute pancreatitis, chronic pancreatitis	20:19:17	-	CT	Retrospective	Isomap and SVM	Accuracy: 82.1%
Yang et al[58]	2019	Serous vs mucinous cystadenoma	53: 25	4:1	CT	Retrospective	RF	Accuracy: 83%; AUC: 0.75

AIP: Autoimmune pancreatitis; AUC: Area under receiver operating characteristic curve; CNN: Convolutional neural network; DFS: Disease-free survival; IPMN: Intraductal papillary mucinous neoplasms; LASSO: Least absolute shrinkage and selection operator; LDA: Linear discriminative analysis; LN: Lymph node; MFP: Mass-forming pancreatitis; MRMR: Minimum redundancy maximum relevance; OS: Overall survival; PD: Pancreaticoduodenectomy; PDAC: Pancreatic ductal adenocarcinoma; PET: Positron emission tomography; PFS: Progression-free survival; PNET: Pancreatic neuroendocrine tumor; POPF: Postoperative pancreatic fistula; RF: Random forest; SBRT: Stereotactic body radiation therapy; SVM: Support vector machine; SMV: Superior mesenteric vein.