Table 3.

Studies assessing AI’s pathological diagnosis.

Study	Ishikawa T. et al., 2022, Japan [21]	Kurita Y et al., 2019, Japan, [31]	Hashimoto Y. et al., 2018, Japan [32]	Inoue H. et al., 2014, Japan [33]
AI type	CNN	ANN	ANN	GMM
Topic	MOSE in pancreatic diseases	analysis of cyst fluid, cytology and EUS characteristics in differentiating malignant from benign pancreatic cysts	ROSE in PDAC	AI automatic visual inspection method is proposed to assist ROSE
Study population	96 patients, 173 specimens	85 patients (59 surgical specimens, 26 EUS-guided FNA specimens)	500 images of cytology specimen (stained and in high definition)	\
Main results	Initial study: AI Ac 71.8% (vs. MOSE performed by EUS experts 81.6%). Using contrastive learning: AI Sn, Sp, Ac: 90.34%, 53.5%, 84.39%, (vs. 88.97%, 53.5%, 83.24% of EUS experts)	AI diagnostic ability in malignant cystic lesions: AUROC curve 0.966 (vs. 0.719 for CEA, 0.739 for cytology) AI Sn, Sp, Ac: 95.7%, 91.9%, 92.9% (vs. CEA Sn 60.9%, p = 0.021; cytology Sn 47.8% p = 0.001; CEA Ac 71.8%, p < 0.00; cytology Ac 85.9%, p = 0.210)	AI Sn, Sp, Ac at the first learning stage: 78%, 60% 69% AI Sn, Sp, Ac at the second learning stage: 80%, 80%, 80%	The AI method is reported as helpful for EUS-FNA in aiding ROSE, indicating areas highly likely to include tumor cells

Abbreviations: AI (Artificial Intelligence), CNN (Convolutional Neural Network), ANN (Artificial Neural Network), GMM (Gaussian Mixture Model), MOSE (Magnifying Endoscopy with Narrow Band Imaging) in pancreatic diseases, ROSE (Rapid On-Site Evaluation), PDAC (Pancreatic Ductal Adenocarcinoma), EUS (Endoscopic Ultrasound), AUROC (Area Under the Receiver Operating Characteristic Curve), CEA (Carcinoembryonic Antigen), and Sn (Sensitivity), Sp (Specificity), and Ac (Accuracy).