Table 4.
Quantitative results obtained by URCTrans using different positive sampling ratios. The best result is highlighted in bold. The numbers of positive nodules predicted by the transformer model trained with different positive sampling ratios, where the input region is regarded as positive if the prediction score ≥ 0.5.
Positive ratio | 0.125 | 0.25 | 0.5 | 1 | 2 | 4 | 8 | CPM | #Predicted |
---|---|---|---|---|---|---|---|---|---|
0.1 | 0.795 | 0.841 | 0.894 | 0.924 | 0.933 | 0.947 | 0.969 | 0.900 | 158 |
0.2 | 0.826 | 0.879 | 0.917 | 0.947 | 0.962 | 0.977 | 0.985 | 0.926 | 217 |
0.3 | 0.765 | 0.856 | 0.917 | 0.932 | 0.969 | 0.977 | 0.985 | 0.915 | 247 |
0.4 | 0.788 | 0.841 | 0.879 | 0.917 | 0.962 | 0.962 | 0.962 | 0.902 | 369 |
0.5 | 0.795 | 0.841 | 0.909 | 0.947 | 0.947 | 0.970 | 0.970 | 0.911 | 409 |