Disease Type | Study | Country | Patient Data | Data Type | AI Algorithm | Performance | Main Findings |
Hp Infection | Nakashima et al., 2018 [15] | Japan | 222 patients | White-Light Imaging (WLI), BLI-bright, and Linked Color Imaging (LCI) endoscopic images | DCNN | WLI AUC: 0.66, BLI-bright AUC: 0.96, LCI AUC: 0.95. | AI technology for image-enhanced endoscopy can be used in the diagnosis of Hp infection. |
Nakashima et al., 2020 [16] | Japan | 515 cases | LCI, WLI | DL | LCI-CAD: Uninfected AUC: 0.90, Currently infected AUC: 0.82, Eradicated AUC: 0.77. | The LCI-CAD system is superior to endoscopists using WLI images and shows consistent diagnostic accuracy with endoscopists using LCI images. | |
Gonçalves et al., 2022 [17] | Brazil | DeepHP Database (394,926 images) | Gastric mucosal histopathology Whole Slide Imaging (WSI) | CNN | Best CNN model (VGG16) Accuracy: 0.98 Specificity: 0.98 AUC: 0.99. | The CNN model can detect Hp infection and inflammation spectrum in gastric biopsies and provides the DeepHP database. | |
Liscia et al., 2022 [18] | Italy | 679 cases | Warthin-Starry (W-S) silver-stained gastric biopsies | DL | Performance metrics for deep learning classifier: Accuracy: 0.880, F1 score: 0.829. | Digital pathology (DP) and AI can reliably identify HP at a 20× resolution. | |
Xiaobin Song et al., 2021 [19] | China | - | Tongue image | Alexnet convolutional neural network | - | It is feasible to construct an Hp tongue image classification model using the Alexnet convolutional neural network. | |
Barrett’s Esophagus and Early Esophageal Cancer | Gao Jingwen et al., 2022 [21] | China | 481 images | Endoscopic esophageal cardia white light images | CNNs | Best model: (EfficientNet model) Accuracy of 0.898, precision of 0.892, recall of 0.906, AUC of 0.946. | Secondary pre-training model outperforms single pre-training model. |