Table 2 Model classification performances on internal validation (SZEH).
From: An ultra-wide-field fundus image dataset for intelligent diagnosis of intraocular tumors
Method | Accuracy | AUC | Precision | Sensitivity | F1 score | Specificity | Kappa |
|---|---|---|---|---|---|---|---|
ResNet50 | 87.96 ± 2.74 [84.65–91.35] | 95.56 ± 1.97 [93.12–98.08] | 82.70 ± 6.20 [75.00–90.4] | 66.30 ± 7.36 [57.11–75.49] | 71.10 ± 7.05 [62.28–79.9] | 95.51 ± 1.05 [94.26–96.7] | 75.63 ± 5.85 [68.4–82.8] |
ResNet101 | 83.79 ± 1.91 [81.44–86.16] | 90.16 ± 2.88 [86.6–93.8] | 62.59 ± 6.09 [55.03–70.17] | 49.48 ± 6.21 [41.8–57.2] | 51.92 ± 6.86 [43.33–60.47] | 94.01 ± 0.90 [92.88–95.12] | 65.92 ± 4.72 [60.06–71.74] |
ConvNeXt-T | 84.76 ± 1.81 [82.57–87.03] | 94.17 ± 1.82 [91.97–96.43] | 66.57 ± 8.62 [55.92–77.28] | 66.27 ± 6.55 [58.11–74.49] | 65.26 ± 7.56 [55.86–74.74] | 95.82 ± 0.15 [95.55–96.05] | 71.58 ± 2.99 [67.88–75.32] |
ViT-B | 91.46 ± 2.12 [88.89–94.11] | 96.87 ± 1.09 [95.53–98.27] | 82.87 ± 5.06 [76.57–89.23] | 81.37 ± 4.24 [76.19–86.61] | 81.42 ± 4.85 [75.44–87.36] | 97.60 ± 0.52 [96.98–98.22] | 84.14 ± 3.84 [79.38–88.82] |
