Table 4.

Lesion characterization algorithms.

Authors	Classes	Algorithm	Dataset	Image Modality	Results	AI Impact (Clinical Setting)
Wu et al., 2021 [58]	Invasion depth: mucosal and submucosal Differentiation Status: Differentiated and undifferentiated	ResNet-50	(Private Dataset) * 3407 images of gastric cancer 1131 differentiated type images 1086 undifferentiated type images	WLI Magnifying NBI	EGC invasion WLI: Accuracy: 88%; Sensitivity: 91%; Specificity: 85% EGC differentiation M-NBI: Accuracy of 86%; Sensitivity: 79%; Specificity: 89%	(Multi-center prospective controlled trial) EGC invasion Experts(n = 8): Sensitivity: 57%% Specificity: 76% Accuracy: 69% EGC invasion Seniors (n = 19): Sensitivity: 60% Specificity: 66% Accuracy: 64% EGC invasion Juniors (n = 19): Sensitivity: 61% Specificity: 61% Accuracy: 61% EGC invasion AI predictions: Sensitivity: 70% Specificity: 83% Accuracy 79% EGC differentiation Experts(n = 8): Sensitivity: 47% Specificity: 83% Accuracy 72% EGC differentiation Seniors (n = 19): Sensitivity: 53% Specificity: 74% Accuracy: 67% EGC differentiation Juniors (n = 19): Sensitivity: 56% Specificity: 60% Accuracy: 59% EGC differentiation AI predictions: Sensitivity: 50% Specificity: 80% Accuracy 71%
Yoon et al., 2019 [56]	T1a T1b Non-EGC	VGG-16	(Private Dataset) 1097 T1a-EGC 1005 T1b-EGC 9834 non-EGC	WLI	Specificity: 75% Sensitivity: 82%	N/A
Nagao et al., 2020 [63]	M-SM1 SM2 or deeper	ResNet-50	(Private Dataset) 10,589 M-SM1 images 6968 SM2 or deeper images	WLI NBI Indigo	WLI Accuracy: 95% NBI Accuracy: 94% Indigo Accuracy: 96%	N/A
Zhu et al., 2019 [64]	P0 (M or SM1) P1 (deeper than SM1)	ResNet-50	(Private Dataset) 545 P0 images 245 P1 images	WLI	Accuracy: 89% Sensitivity: 76% Specificity: 96%	CNN Accuracy: 89% Sensitivity:76% Specificity: 96% Experts(n = 8): Accuracy: 77% Sensitivity: 91% Specificity: 71% Junior (n = 9): Accuracy: 66% Sensitivity: 85% Specificity: 57%
Xu et al., 2021 [65]	GA IM	VGG-16	(Private Dataset) 2149 GA images 3049 IM images	Magnifying NBI Magnifying BLI	GA Accuracy: 90% Sensitivity: 90% Specificity: 91% IM Accuracy: 91% Sensitivity: 89% Specificity: 93%	(Multi-center Prospective blinded trial) GA classification CAD System: Accuracy: 87% Sensitivity: 87% Specificity: 86% Experts(n = 4): Accuracy: 85% Sensitivity: 91% Specificity: 72% Non experts (n = 5): Accuracy: 75% Sensitivity: 83% Specificity: 59% IM classification: CAD System: Accuracy: 89% Sensitivity: 90% Specificity: 86% Experts(n = 4): Accuracy: 82% Sensitivity: 83% Specificity: 81% Non experts (n = 5): Accuracy: 74% Sensitivity: 74% Specificity: 73%

* An external dataset of 1526 images has been used to test the performance of the model.