| Meng et al., 2022 [27] | China | 1160 cases | WLI images and NBI images | YOLOv5 deep learning algorithm | After reference to CAD system, non-experts show significantly reduced differences in accuracy (88.2% vs. 93.2%, p < 0.001), sensitivity (87.6% vs. 92.3%, p = 0.013), and specificity (89.5% vs. 94.7%, p = 0.124) compared to experts. | The CAD system in WLI and NBI combination mode can improve the diagnostic performance of superficial ESCC. |
Zhao et al., 2022 [28] | China | 300 cases | NBI images | Google net model with Inception v3 image classification system | There is no statistically significant difference between AI-NBI diagnosis and doctor diagnosis in terms of sensitivity (90.0% vs. 92.0%), specificity (92.0% vs. 94.0%), and accuracy (91.0% vs. 93.0%) (P > 0.05). | AI-NBI can assist in the diagnosis of early esophageal cancer. | |
Chronic Atrophic Gastritis and Gastric Cancer | Zhang et al., 2020 [30] | China | 5470 images of the gastric antrum from 1699 patients | Conversion of white light endoscopic images to uncompressed BMP format | Trained CNN-CAG model using DenseNet 121 network architecture | The accuracy of the CNN system in diagnosing mild, moderate, and severe atrophic gastritis was 0.93, 0.95, and 0.99, respectively, indicating a higher detection rate for moderate and severe atrophic gastritis than for mild cases. | DenseNet network showed high performance in CAG identification. |