Table 1.
Summary of the studies describing the recent developments in Artificial Intelligence systems.
| Study (Year) |
Study Type | Objective | Data type | AI | Results | Application | Limitations |
|---|---|---|---|---|---|---|---|
| Mohan et al., 2022 |
Meta-analysis 11 studies 2001–2020 |
Overall performance of AI in: -Diagnosis and characterization of solid PL -Differentiate PC from non-neoplastic tissue -Differentiate malignancy from CP -Diagnosis of PC |
-EUS elastography -EUS images -EUS—videos -CEH-EUS |
Fractal-based quantitative analysis NN algorithm SVM |
-Overall accuracy 85.8% -Sens 91.8% -Spec 84.6% -PPV 87.4% -NPV 91.4% -Heterogeneity 57% |
Superior diagnostic results with the combination of AI and newer core-biopsy needles in EUS evaluation of solid masses | -Heterogeneity -Absence of prospective data |
| Yu et al., 2022 |
Case report | Guiding punction of pancreatic masses by differentiating cancerous, non-cancerous, and necrotic regions | N/A | Deep CNN | Improving the diagnostic accuracy of EUS FNA | ||
| Zhang et al., 2020 |
Cross-over 8 participants |
BP-MASTER® -Test the performance of classifying the previously learned stations of pancreatic EUS -Pancreatic tissue and blood vessel segmentation |
-EUS images -EUS videos |
Deep learning | -Classification accuracy 86% -Comparable accuracy between endoscopists and AI -Improvement of trainee’s accuracy for classification and segmentation |
Shortening the pancreatic EUS learning curve Improving EUS quality control |
-Duodenal bulb station non studied |
| Goyal et al., 2022 | Systematic review 11 studies |
Study the effectiveness of AI with EUS in the diagnosis of pancreatic cancer -Differentiating PC from CP -Differentiating malignant from benign IPMNs |
-Retrospective EUS images and videos -Real time collected EUS images |
ANN CNN SVM |
Performance in recognition of pancreatic malignancy: -Sens 83–100% -Spec 50–99% -Accuracy 80–97.5% -PC vs. CP ANN -Sens 88–100% -Spec 50–94% -SVM -Sens 96% -Spec 93% -Accuracy 94% CNN -Sens 90% -Spec 75% Benign vs. malignant IPMNs CNN -Sens 95.7% -Spec 92.6% -Accuracy 94% |
-Improvement of pancreatic malignancy recognition even in presence of chronic pancreatitis -SVM method simpler and highly performant |
|
| Ishikawa et al., 2022 |
Retrospective | -Study the usefulness of AI in predicting the EUS-FNB sample quality for histopathological examination | -Stereomicroscopic images of EUS-FNB specimens | CNN and deep learning Contrastive learning |
-AI evaluation using contrastive learning is comparable to MOSE performed by EUS experts -Diagnostic accuracy with deep learning not as high as MOSE performed by experts |
Increasing the objectivity of the evaluation | Small sample size |
AI artificial intelligence, ANN artificial neuronal networks, CEH-EUS contrast enhanced EUS, CNN convolutional neural networks, CP chronic pancreatitis, EUS endoscopic ultrasound, FNA fine needle aspiration, FNB fine needle biopsy, IPMN intraductal papillary mucinous neoplasm, MOSE macroscopic on-site evaluation, NN neural network, NPV negative predictive value, PC pancreatic cancer, PL pancreatic lesions, PPV positive predictive value, Sens sensitivity, Spec specificity, SVM Support vector machine.