| Tang et al., 2022 [34] | China | 21,785 NBI (narrow-band imaging) images and 20 videos | NBI | NBI AI system | The AI system achieved better diagnostic performance (accuracy: 79.5%, 95% CI: 77.8% - 81.0%) compared to senior (93.2%, 95% CI: 90.0% - 94.9%) and junior (85.9%, 95% CI: 84.2% - 87.4%) endoscopists. | The NBI AI system outperformed endoscopists and has the potential to improve diagnostic rates in real-time diagnosis by physicians. |
Bang et al., 2021 [35] | South Korea | 2703 patients (967 ESD [endoscopic submucosal dissection], 1736 surgery) | Each dataset was written in .csv file format | ML classifier | XGBoost classifier demonstrated the best performance with an accuracy of 93.4%, recall of 99.0%, and F1 score of 95.7%. | XGBoost classifier showed high recognition capability for curative resection in undifferentiated early gastric cancer, considering variables such as patient age, sex, endoscopic lesion size, morphology, and presence/absence of ulcer. | |
Kuroda et al., 2022 [36] | Japan | 171 cases | Robot-assisted gastric resection using ultrasonic shears and conventional robot-assisted gastric resection using conventional forceps | Robot | The console time (310 minutes [interquartile range (IQR), 253 - 369 minutes] vs. 332 minutes [IQR, 294 - 429 minutes]; p = 0.022) and console time for gastric resection (222 minutes [IQR, 177 - 266 minutes] vs. 247 minutes [IQR, 208 - 321 minutes]; p = 0.004) were significantly shorter in the ultrasonic shears group compared to the conventional forceps group. | The ultrasonic shears robot showed better performance in robot-assisted gastric resection for gastric cancer. |