Table 2.
Studies assessing diagnostic capabilities of AI systems in EUS.
| Study | Hirai K. et al., 2022, Japan [19] |
Marya NB. et al., 2021, USA [24] |
Marya NB. et al., 2021, USA [25] |
Oh CK. et al., 2021, South Korea [26] |
Săftoiu A. et al., 2015, Denmark [29] |
Zhang MM. et al., 2010, China [27] |
|---|---|---|---|---|---|---|
| AI type | CNN | CNN | CNN | CNN | ANN | SVM |
| Topic | differential diagnosis of SELs (five-category: GIST, leiomyoma, schwannoma, NET, ectopic pancreas) | enhance the diagnosis of AIP | identify and classify FLLs | differential diagnosis of SELs (GISTs and leiomyomas) | differential diagnosis of PDAC and CP using CH-EUS and TIC analysis | differential diagnosis of PDAC from normal tissue (based on 29 pattern features) |
| Study population | 16,110 images, 631 examinations | 583 patients, 1,174,461 still images from videos | 256 patients, 210,685 still images from videos | 114 patients (with histologically confirmed gastric GIST), 376 still images | 167 patients with PDAC or CP | 216 patients (153 with PDAC, 63 without) |
| Main results | overall Ac: AI 86.1% vs. expert endoscopy 68.0% (p < 0.001); Sn, Sp, Ac of AI in differentiating GISTs from non-GIST: 98.8%, 67.6%, 89.3% (better than expert endoscopist: Sn, Ac p < 0001) |
AI processed 955 frames/sec. Sn and Sp for distinguishing AIP from PDAC: 90%, 93%. Sn and Sp for distinguishing AIP from all studied conditions (PDAC, CP, NP): 90%, 85% |
AI autonomously locates FLLs in 92.0% of videos. Sn and Sp in classifying malignant FLLs on random still images: 90%, 71%. Sn and Sp in classifying malignant FLLs on full-length videos: 100% and 80% |
AI Sn, Sp, Ac in per-image analysis: 95.6%, 82.1%, 91.2% Sn, Sp, Ac in per-patient analysis 100.0%, 85.7%, 96.3% (better than expert endoscopist: Sn, Ac p < 0001) |
AI Sn Sp PPV and NPV using TIC analysis on CH-EUS: 94.64%, 94.44%, 97.24%, 89.47% | AI Ac, Sn, Sp, PPV and NPV for the diagnosis of pancreatic cancer: 97.98%, 94.32%, 99.45%, 98.65%, 97.77% |
Abbreviations: AI (Artificial Intelligence), CNN (Convolutional Neural Network), ANN (Artificial Neural Network), SVM (Support Vector Machine), SELs (Subepithelial Lesions), GIST (Gastrointestinal Stromal Tumor), NET (Neuroendocrine Tumor), AIP (Autoimmune Pancreatitis), CP (Chronic Pancreatitis), CH-EUS (Contrast-Enhanced Endoscopic Ultrasound), TIC (Time-Intensity Curve), PDAC (Pancreatic Ductal Adenocarcinoma), FLLs (Focal Liver Lesions), NP (Normal Pancreas/Neuroendocrine Tumor), PPV (Positive Predictive Value), and NPV (Negative Predictive Value).