Table 2 Performance metrics of different methods on the Camelyon-17-Refine dataset.
From: Comprehensive Benchmark Dataset for Pathological Lymph Node Metastasis in Breast Cancer Sections
Methods | Acc (%) | AUC (%) | F1 (%) | Recall (%) | Precision (%) | Kappa |
|---|---|---|---|---|---|---|
PLIP3 (WSIs pre-trained) | ||||||
Max-MIL | 77.8 ± 9.55 | 72.7 ± 11.09 | 38.7 ± 10.28 | 37.4 ± 10.63 | 41.0 ± 3.25 | 0.49 ± 0.32 |
Mean-MIL | 79.8 ± 0.24 | 77.7 ± 0.25 | 43.2 ± 0.18 | 44.8 ± 0.22 | 49.2 ± 0.57 | 0.59 ± 0.00 |
ABMIL19 | 81.8 ± 1.39 | 87.0 ± 1.18 | 52.9 ± 1.63 | 51.0 ± 1.48 | 51.5 ± 0.42 | 0.75 ± 0.02 |
Gate-ABMIL19 | 81.9 ± 2.04 | 87.5 ± 0.13 | 53.1 ± 2.83 | 51.3 ± 2.61 | 51.7 ± 1.01 | 0.75 ± 0.02 |
CLAM-SB25 | 82.8 ± 1.73 | 87.0 ± 0.76 | 54.2 ± 2.32 | 52.3 ± 1.95 | 52.0 ± 0.88 | 0.76 ± 0.02 |
CLAM-MB25 | 87.2 ± 0.80 | 89.3 ± 0.41 | 61.4 ± 2.12 | 59.6 ± 2.69 | 61.2 ± 7.85 | 0.82 ± 0.01 |
DSMIL34 | 86.2 ± 1.32 | 87.8 ± 0.86 | 56.4 ± 2.37 | 56.2 ± 1.95 | 56.5 ± 1.79 | 0.76 ± 0.02 |
TransMIL24 | 83.8 ± 1.00 | 88.9 ± 1.15 | 58.9 ± 5.70 | 56.9 ± 4.14 | 56.7 ± 2.78 | 0.71 ± 0.02 |
DTFD35 | 84.3 ± 1.81 | 86.7 ± 0.95 | 52.1 ± 1.79 | 51.2 ± 1.12 | 51.3 ± 0.19 | 0.77 ± 0.01 |
AMD-MIL33 | 86.4 ± 0.97 | 89.0 ± 0.67 | 61.9 ± 3.01 | 59.6 ± 3.07 | 58.4 ± 2.66 | 0.78 ± 0.03 |
WiKG36 | 87.1 ± 1.20 | 88.2 ± 1.64 | 57.9 ± 2.76 | 56.6 ± 2.13 | 55.8 ± 1.86 | 0.80 ± 0.03 |
FR-MIL37 | 80.6 ± 1.59 | 87.0 ± 5.90 | 57.5 ± 6.19 | 55.3 ± 4.36 | 54.8 ± 3.49 | 0.63 ± 0.02 |
UNI4 (WSIs pre-trained) | ||||||
Max-MIL | 79.6 ± 10.07 | 77.9 ± 11.95 | 41.3 ± 10.87 | 39.1 ± 9.87 | 41.1 ± 2.10 | 0.55 ± 0.35 |
Mean-MIL | 76.5 ± 2.02 | 81.6 ± 0.26 | 46.4 ± 0.70 | 46.8 ± 1.09 | 49.4 ± 0.83 | 0.49 ± 0.06 |
ABMIL19 | 82.1 ± 1.60 | 92.3 ± 0.64 | 67.7 ± 1.51 | 61.9 ± 1.47 | 59.6 ± 1.29 | 0.72 ± 0.02 |
Gate-ABMIL19 | 81.0 ± 0.86 | 92.2 ± 0.28 | 65.7 ± 4.13 | 59.7 ± 3.14 | 57.9 ± 2.48 | 0.68 ± 0.05 |
CLAM-SB25 | 81.3 ± 1.07 | 93.1 ± 0.22 | 64.8 ± 6.08 | 58.5 ± 4.25 | 56.5 ± 3.85 | 0.73 ± 0.07 |
CLAM-MB25 | 85.0 ± 0.68 | 95.9 ± 0.21 | 74.5 ± 8.42 | 65.6 ± 3.72 | 63.4 ± 3.06 | 0.70 ± 0.08 |
DSMIL34 | 85.9 ± 3.17 | 93.1 ± 2.26 | 62.6 ± 5.28 | 59.9 ± 3.16 | 59.0 ± 1.35 | 0.75 ± 0.05 |
TransMIL24 | 88.5 ± 0.44 | 95.2 ± 0.93 | 70.4 ± 1.88 | 65.7 ± 1.97 | 66.1 ± 6.04 | 0.78 ± 0.06 |
DTFD35 | 82.6 ± 0.56 | 94.5 ± 0.31 | 61.0 ± 4.13 | 56.5 ± 4.67 | 56.6 ± 5.77 | 0.78 ± 0.01 |
AMD-MIL33 | 86.0 ± 1.12 | 94.8 ± 0.13 | 73.6 ± 3.45 | 68.5 ± 1.14 | 66.8 ± 2.55 | 0.78 ± 0.03 |
WiKG36 | 83.1 ± 3.56 | 95.0 ± 0.42 | 73.3 ± 1.91 | 64.6 ± 3.19 | 62.6 ± 1.50 | 0.68 ± 0.03 |
FR-MIL37 | 85.0 ± 0.97 | 96.0 ± 0.46 | 78.3 ± 4.48 | 68.0 ± 1.51 | 65.4 ± 1.22 | 0.70 ± 0.09 |
Gigapath5 (WSIs pre-trained) | ||||||
Max-MIL | 83.7 ± 2.77 | 83.4 ± 3.83 | 47.3 ± 0.92 | 45.9 ± 0.76 | 44.9 ± 1.69 | 0.75 ± 0.03 |
Mean-MIL | 76.1 ± 4.24 | 81.3 ± 0.55 | 49.7 ± 3.29 | 49.5 ± 0.91 | 52.1 ± 1.32 | 0.51 ± 0.06 |
ABMIL19 | 81.4 ± 0.49 | 91.8 ± 0.70 | 66.6 ± 2.55 | 61.6 ± 2.21 | 59.8 ± 2.55 | 0.75 ± 0.03 |
Gate-ABMIL19 | 81.4 ± 1.73 | 92.7 ± 0.73 | 70.6 ± 1.69 | 63.2 ± 2.80 | 61.1 ± 2.50 | 0.72 ± 0.07 |
CLAM-SB25 | 78.7 ± 2.33 | 92.6 ± 0.38 | 59.9 ± 2.54 | 54.8 ± 1.68 | 55.2 ± 4.44 | 0.72 ± 0.08 |
CLAM-MB25 | 84.4 ± 1.92 | 96.5 ± 0.44 | 81.1 ± 3.75 | 68.5 ± 0.39 | 65.6 ± 0.30 | 0.64 ± 0.09 |
DSMIL34 | 86.2 ± 0.85 | 93.5 ± 0.61 | 68.8 ± 4.85 | 65.2 ± 3.73 | 63.8 ± 3.42 | 0.75 ± 0.06 |
TransMIL24 | 86.2 ± 0.76 | 95.6 ± 0.36 | 73.4 ± 4.51 | 67.9 ± 4.69 | 66.5 ± 4.56 | 0.80 ± 0.02 |
DTFD35 | 80.9 ± 1.18 | 93.4 ± 0.81 | 58.1 ± 2.52 | 52.7 ± 1.53 | 53.5 ± 1.32 | 0.76 ± 0.02 |
AMD-MIL33 | 84.6 ± 1.20 | 95.0 ± 0.41 | 72.9 ± 1.31 | 66.4 ± 1.12 | 64.9 ± 2.85 | 0.75 ± 0.07 |
WiKG36 | 83.8 ± 1.41 | 94.4 ± 0.52 | 73.8 ± 2.83 | 66.0 ± 2.31 | 63.8 ± 1.39 | 0.74 ± 0.04 |
FR-MIL37 | 85.9 ± 0.88 | 95.4 ± 0.89 | 79.7 ± 6.78 | 69.2 ± 2.62 | 65.7 ± 1.82 | 0.72 ± 0.04 |
