Table 2.
EUS-based models for evaluating PCLs.
| Study | Sample Size |
Model | Task | Accuracy | Comparisons |
|---|---|---|---|---|---|
| Schultz, 2022 [23] | 43 | CNN | Low- vs. high-grade IPMN dysplasia | Accuracy 99.6% | Higher accuracy than AGA, ACG, Fukuoka guidelines |
| Kuwahara, 2019 [62] | 50 | Deep learning | Evaluate malignant potential in IPMN images | Sensitivity 95.7%, Specificity 96.2%, Accuracy 94.0% | Human interpretation, 56% accuracy |
| Nguon, 2021 [63] |
109 | CNN | Differentiate between MCNs and SCAs | Accuracy 83% | |
| Machicado, 2021 [76] | 35 | CNN | Low- vs. high-grade BD-IPMN dysplasia | Accuracy 82% | Higher accuracy than AGA and Fukuoka guidelines |