Role of artificial intelligence in Barrett’s esophagus

Table 1 Summary of the original research studies included in the review

Ref.	Study objective	Diagnostic modality	Study type	Real time	Dysplasia inclusion	Test images, n	Diagnostic performance
Pan et al[20], 2021	BE segment identification	WLI, NBI	Retrospective	No	NA	443	IOU: GEJ 0.56, SCJ 0.82, GEJ+SCJ 0.66
Ebigbo et al[23], 2019	BE neoplasia detection	WLI, NBI	Retrospective	No	EAC	MICCAI: 100; Ausburg: 148	Sen 92%, Spec 100%; WLI: Sen 97%, Spec 88%; NBI: Sen 94%, Spec 80%
Ebigbo et al[24], 2020	BE neoplasia detection	WLI	Prospective	Yes	EAC	191	Sen 83.7%, Spec 100%
de Groof et al[25], 2019	BE neoplasia detection	WLI	Retrospective	No	HGD, EAC	60	Sen 95%, Spec 85%
de Groof et al[26], 2020	BE neoplasia detection	WLI	Retrospective	No	HGD, EAC	297; 80; 80	Sen 87.6%, Spec 88.6%; Sen 90%, Spec 87.5%; Sen 92.5%, Spec 82.5%
de Groof et al[27], 2020	BE neoplasia detection	WLI	Prospective	Yes	HGD, EAC	144	Sen 91%, Spec 89%
Abdelrahim et al[28], 2020	BE neoplasia detection	WLI	Retrospective	No	NA	251	Sen 93%, Spec 78%
Hashimoto et al[29], 2020	BE neoplasia detection	WLI, NBI	Retrospective	No	HGD, EAC	458	Sen 96.4%, Spec 94.2%
Struyvenberg et al[30], 2021	BE neoplasia detection	NBI	Retrospective	No	HGD, EAC	183 zoom images; 157 zoom videos	Sen 88%, Spec 78%; Sen 85%, Spec 83%
Ebigbo et al[41], 2021	BE cancer invasion	WLI	Retrospective	No	EAC (T1a, T2a)	230	Sen 77%, Spec 64%
Tomita et al[45], 2019	BE neoplasia histology detection	NA	Retrospective	No	LGD, HGD, EAC	123 WSI	Mean accuracy 83%

BE: Barrett’s esophagus; WLI: White-light images; NBI: Narrow-band images; LGD: Low grade dysplasia; HGD: High grade dysplasia; EAC: Esophageal adenocarcinoma; IOU: Intersection over union; GEJ: Gastro-esophageal junction; SCJ: Squamo-columnar junction; Sen: Sensitivity; Spec: Specificity; MICCAI: Medical Image Computing and Computer-Assisted Intervention; WSI: Whole-slide images; NA: Not available.