Table 3.
Comparison of the absolute performance of the ANN training methods in the breast ultrasound study.
| No regularizationa | Noise injection | Weight decay | ||
|---|---|---|---|---|
| 50 training cases | Average AUCb [95% CI] | 0.801 [0.795. 0.807] | 0.849 [0.844, 0.853] | 0.838 [0.833, 0.843] |
| AUC standard deviation [95% CI] | 0.065 [0.061, 0.069] | 0.056 [0.052, 0.060] | 0.058 [0.054, 0.062] | |
| 100 training cases | Average AUCb [95% CI] | 0.807 [0.803, 0.811] | 0.856 [0.853, 0.860] | 0.851 [0.847, 0.854] |
| AUC standard deviation [95% CI] | 0.047 [0.044, 0.050] | 0.039 [0.037, 0.042] | 0.040 [0.038, 0.043] |
a
These ANNs had five hidden nodes and were trained to 500 training iterations. The results were calculated at the 485th training iteration.
b
Performance measured by the .632+ bootstrap AUC values.