Table 2.
Performance of Resnet-101 trained with imbalanced data, oversampling, and undersampling (without and with Data Augmentation (DA)) in classifying Luminal A, Luminal B1, Luminal B2, HER2, and TN versus non-respective classes, in terms of accuracy. * p < 0.05: There is a significant difference between groups. In the line of average values, the best value obtained is highlighted in bold.
| Accuracy (%): Mean (stdDev) | ||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Imbalanced Data | Oversampling | Undersampling | ||||||||||||||||||||||
| without DA | with DA | without DA | with DA | without DA | with DA | |||||||||||||||||||
| Luminal A vs. Non-Luminal A | 72.28 (1.80) | 67.14 (2.04) | 70.73 (2.72) | 73.72 (0.74) | 51.78 (3.15) | 60.00 (5.93) | ||||||||||||||||||
| Luminal B1 vs. Non-Luminal B1 | 74.08 (1.36) | 71.21 (1.81) | 73.95 (0.41) | 74.91 (1.99) | 55.14 (4.89) | 55.14 (7.22) | ||||||||||||||||||
| Luminal B2 vs. Non-Luminal B2 | 79.33 (2.71) | 78.14 (0.95) | 79.69 (2.27) | 80.76 (1.66) | 58.07 (2.59) | 61.17 (2.38) | ||||||||||||||||||
| HER2 vs. Non-HER2 | 88.29 (1.66) | 89.79 (1.27) | 88.77 (1.36) | 88.65 (1.80) | 59.02 (5.91) | 62.84 (8.25) | ||||||||||||||||||
| TN vs. Non-TN | 77.42 (2.18) | 73.96 (2.90) | 76.70 (2.24) | 77.06 (0.62) | 59.13 (5.29) | 55.65 (4.52) | ||||||||||||||||||
| Average | 78.28 (1.94) | 76.05 (1.79) | 77.97 (1.80) | 79.02 (1.36) | 56.63 (4.37) | 58.96 (5.66) | ||||||||||||||||||
| * | ||||||||||||||||||||||||
| * | ||||||||||||||||||||||||
| * | ||||||||||||||||||||||||
| * | ||||||||||||||||||||||||