TABLE 3.

AI in the Diagnosis of Benign Esophageal Diseases

					Performance
References	AI Model	Study Type	Aim	Diagnostic Tool	Accuracy	Sensitivity	Specificity
Pace et al50	ML	Prospective	Distinction between GERD and non-GERD based on symptoms	Questionnaire	100%	—	—
Horowitz et al51	Data mining	Prospective	Distinction between GERD and non-GERD based on symptoms	Questionnaire	AUC=0.78	70%-75%	63%-78%
Pace et al52	ML	Prospective	Distinction between NERD and EE	Questionnaire	NERD 62.2% EE 70.9%	—	—
Rogers et al53	Decision tree analysis	Prospective	Automate extraction of pH-impedance metrics	pH-impedance tracings	88.5%	—	—
Rogers et al53	Decision tree analysis	Prospective	Predict response to GERD management	pH-impedance tracings	AUC=0.77	—	—
Gulati et al54	DL	Prospective	Endoscopic diagnosis of GERD	NF-NBI	AUC=0.83	67%	92%
Sallis et al55	ML	Prospective	Diagnosis of EoE based on mRNA transcripts from esophageal biopsies	Esophageal biopsies	AUC=0.98	91%	93%
Santos et al56	Multilayered back-propagation ANN	Prospective	Diagnosis of esophageal motility pattern	Stationary esophageal manometry tracings	82%	—	—
Lee et al57	DL	Retrospective	AI vs. endoscopist in differential diagnosis HSV vs. CMV esophagitis	WLE AI Endoscopists	100% 52.7%	100% —	100% —

AI indicates artificial intelligence; ANN, artificial neural network; AUC, area under the curve; CMV, cytomegalovirus; DL, deep learning; EE, erosive esophagitis; EoE, eosinophilic esophagitis; GERD, gastroesophageal reflux disease; HSV, herpes simplex virus; ML, machine learning; NERD, nonerosive reflux disease; NF-NBI, near-focus narrow-band imaging; WLE, white-light endoscopy.