Table 2.
Diagnostic performance of 2D and 3D CNN models and participated reading clinicians on surgery and internal test sets
| Metrics | Accuracy | Precision | Sensitivity | Specificity | F1-score | Youden-Index |
|---|---|---|---|---|---|---|
| Surgery test set | ||||||
| 2D CNN | 0.87 (60/69) | 0.75 (21/28) | 0.913 (21/23) | 0.848 (39/46) | 0.824 | 0.761 |
| 3D CNN | 0.71 (49/69) | 0.54 (17/31) | 0.739 (17/23) | 0.696 (32/46) | 0.624 | 0.435 |
| Senior surgeon 1 and 2 | 0.891 (123/138) | 0.782 (43/55) | 0.935 (43/46) | 0.870 (80/92) | 0.852 | 0.805 |
| Junior surgeon 3 and 4 | 0.761 (105/138) | 0.592 (42/71) | 0.913 (42/46) | 0.685 (63/92) | 0.718 | 0.598 |
| Senior radiologist 5 and 6 | 0.862 (119/138) | 0.729 (43/59) | 0.935 (43/46) | 0.826 (76/92) | 0.819 | 0.761 |
| Junior radiologist 7 and 8 | 0.775 (107/138) | 0.612 (41/67) | 0.891 (41/46) | 0.717 (66/92) | 0.726 | 0.608 |
| Internal test set | ||||||
| 2D CNN | 0.818 (117/143) | 0.72 (39/54) | 0.78 (39/50) | 0.839 (78/93) | 0.75 | 0.619 |
| 3D CNN | 0.783 (112/143) | 0.679 (36/53) | 0.72 (36/50) | 0.817 (76/93) | 0.699 | 0.537 |
| Senior surgeon 1 and 2 | 0.857 (245/286) | 0.748 (89/119) | 0.89 (89/100) | 0.839 (156/186) | 0.813 | 0.729 |
| Junior surgeon 3 and 4 | 0.801 (229/286) | 0.669 (85/127) | 0.85 (85/100) | 0.753 (140/186) | 0.749 | 0.603 |
| Senior radiologist 5 and 6 | 0.839 (240/286) | 0.725 (87/120) | 0.87 (87/100) | 0.823 (153/186) | 0.791 | 0.693 |
| Junior radiologist 7 and 8 | 0.759 (217/286) | 0.637 (72/113) | 0.74 (74/100) | 0.801 (149/186) | 0.685 | 0.541 |