Table 5. The classification results of HSIL− vs. HSIL+.
Model | AUC | Accuracy | Sensitivity | Specificity | PPV | NPV |
---|---|---|---|---|---|---|
Clinical only | 0.700 (0.676–0.723) | 0.675 (0.650–0.698) | 0.724 (0.679–0.766) | 0.655 (0.626–0.684) | 0.451 (0.414–0.490) | 0.858 (0.833–0.881) |
CNN only | 0.887 (0.870–0.903) | 0.797 (0.776–0.818) | 0.802 (0.761–0.839) | 0.796 (0.770–0.819) | 0.606 (0.564–0.647) | 0.911 (0.891–0.928) |
CNN + clinical | 0.900 (0.884–0.915) | 0.807 (0.786–0.826) | 0.823 (0.783–0.858) | 0.800 (0.775–0.823) | 0.618 (0.576–0.658) | 0.920 (0.901–0.937) |
Data were presented with 95% CIs. HSIL, high-grade squamous intraepithelial lesions; AUC, receiver operating characteristic curve; PPV, positive predictive value; NPV, negative predictive value; CI, confidence interval; CNN, convolutional neural network.