Table 5 Comparison of the efficacy of ResNet, DenseNet, Vision Transformer and Swin Transformer in identifying benign and malignant ovarian tumors with or without CA125
From: Machine learning models in evaluating the malignancy risk of ovarian tumors: a comparative study
AUC | Sensitivity | Specificity | NPV | PPV | Youden index | Cutoff value | DOR | P value | |
|---|---|---|---|---|---|---|---|---|---|
ResNet | 0.91(0.85–0.95 | 82.1(60.7–88.9) | 93.4(87.5–97.1) | 99.6(99.2–99.8) | 20.3(7.2–45.7) | 0.75 | > 0.58 | 46.92 | |
ResNet + CA125 | 0.90(0.84–0.94) | 82.1(66.5–92.5) | 93.4(87.5–97.1) | 99.6(99.2–99.8) | 20.3(7.2–45.7) | 0.75 | > 0.38 | 65.84 | 0.29 |
DenseNet | 0.91(0.86–0.95) | 84.6(69.5–94.1) | 92.6(86.5–96.6) | 99.7(99.3–99.8) | 26.0(8.1–58.4) | 0.77 | > 0.25 | 67.47 | |
DenseNet + CA125 | 0.91(0.85–0.95) | 84.6(69.5–94.1) | 95.9(90.7–98.7) | 99.7(99.3–99.8) | 29.6(8.4–65.9) | 0.81 | > 0.18 | 129.06 | 0.53 |
Vision Transformer | 0.87(0.81–0.92) | 84.6(69.5–94.1) | 81.2(73.1–87.7) | 99.6(99.2–99.8) | 8.4(4.5–15.1) | 0.66 | > 0.17 | 23.63 | |
Vision Transformer + CA125 | 0.87(0.81–0.92) | 84.6(69.5–94.1) | 79.5(71.3–86.3) | 99.6(99.2–99.8) | 7.8(4.3–13.7) | 0.64 | > 0.11 | 21.74 | 0.71 |
Swin Transformer | 0.92(0.87–0.96) | 87.2(72.6–95.7) | 94.3(88.5–97.7) | 99.7(99.4–99.9) | 23.7(7.9–52.7) | 0.81 | > 0.33 | 108.50 | |
Swin Transformer + CA125 | 0.93(0.88–0.97) | 87.2(72.6–95.7) | 94.3(88.5–97.7) | 99.7(99.4–99.9) | 23.7(7.9–52.7) | 0.81 | > 0.25 | 108.50 | 0.23 |
SA | 0.97(0.93–0.99) | 87.2(72.6–95.7) | 98.4(94.2–99.8) | 99.7(99.4–99.9) | 52.0(8.7–92.5) | 0.86 | > 3 | 409.08 |