Table 5.
Performance comparison of the DNN-based model with external data sets.
| References | Level * (%) |
Training Data Set Size |
ACC | SE | SP | AUC |
|---|---|---|---|---|---|---|
| Liew et al. (entire data set) [12] | 0% | 114 (68+/46−) |
0.789 | 0.838 | 0.717 | 0.853 |
| 100% | 187 (105+/82−) |
0.642 | 0.724 | 0.537 | 0.742 | |
| valBLACK | 0% | 38 | 0.974 | 0.955 | 1.000 | 0.955 |
| (22+/16−) | (0.809) | (0.957) | (0.667) | (0.924) | ||
| 100% | 47 (23+/24−) |
0.830 | 0.957 | 0.708 | 0.937 | |
| valPAIR | 0% | 14 | 0.500 | 0.857 | 0.143 | 0.551 |
| (7+/7−) | (0.550) | (0.800) | (0.300) | (0.450) | ||
| 100% | 20 (10+/10−) |
0.450 | 0.700 | 0.200 | 0.525 | |
| valRANDOM | 0% | 62 | 0.742 | 0.769 | 0.696 | 0.836 |
| (39+/23−) | (0.750) | (0.819) | (0.646) | (0.595) | ||
| 100% | 120 (72+/48−) |
0.600 | 0.653 | 0.521 | 0.687 | |
| Zhang et al. [14] | 0% | 80 | 0.950 | 1.000 | 0.926 | 0.957 |
| (53+/27−) | (0.750) | (0.932) | (0.379) | (0.667) | ||
| 100% | 85 (57+/28−) |
0.941 | 0.982 | 0.857 | 0.952 | |
| Ai et al. [15] | 0% | 84 | 0.881 | 0.905 | 0.810 | 0.920 |
| (63+/21−) | (0.843) | (0.869) | (0.754) | (0.904) | ||
| 100% | 121 (94+/27−) |
0.893 | 0.904 | 0.852 | 0.911 | |
| Kotsampasakou et al. [13] | 0% | 151 | 0.636 | 0.595 | 0.687 | 0.672 |
| (84+/67−) | (0.600) | (0.670) | (0.520) | (0.642) | ||
| 100% | 973 (524+/449−) |
0.585 | 0.635 | 0.526 | 0.605 |
* Endurance level. ACC: accuracy; SE: sensitivity; SP: specificity; AUC: area under the receiver–operating characteristic curve. The data in parentheses are validation results from each reference.