Table 2.
A performance summary of the image−only CNNs and image−only radiomics models in training and testing cohorts in predicting ALN metastasis of breast cancer.
| Model | Dataset | AUC | ACC (%) | SEN (%) | SPE (%) | PPV (%) | NPV (%) | |
|---|---|---|---|---|---|---|---|---|
| Image−only CNN | Intra | Training | 0.937 (0.908, 0.968) |
84.6 − |
95.7 (80.4, 98.9) |
80.3 (76.6, 95.4) |
65.2 (60.7, 87.8) |
98.0 (92.5, 99.5) |
| Testing | 0.748 (0.622, 0.874) |
71.8 | 76.0 (48.0, 92.0) |
70.5 (60.3, 94.9) |
45.2 (39.2, 77.8) |
90.2 (84.0, 96.7) |
||
| Peri | Training | 0.944 (0.920, 0.969) |
87.0 − |
95.7 (89.1, 100.0) |
83.7 (78.7, 91.6) |
69.3 (63.5, 81.1) |
98.0 (95.3, 100.0) |
|
| Testing | 0.775 (0.665, 0.886) |
72.8 − |
80.0 (56.0, 92.0) |
70.5 (64.1, 92.3) |
46.5 (40.9, 72.7) |
91.7 (85.3, 96.8) |
||
| Cmb | Training | 0.957 (0.926, 0.989) |
93.7 − |
92.6 (86.2, 97.9) |
94.1 (90.8, 99.6) |
86.1 (80.0, 98.8) |
97.0 (94.5, 99.1) |
|
| Testing | 0.912 (0.834, 99.0) |
89.3 − |
85.7 (67.9, 96.4) |
90.7 (84.0, 100.0) |
77.4 (66.7, 100.0) |
94.4 (89.0, 98.6) |
||
| Image−only radiomics |
Intra | Training | 0.913 (0.870, 0.956) |
87.9 − |
84.8 (75.0, 91.3) |
89.1 (86.6, 95.8) |
75.0 (70.9, 88.6) |
93.8 (90.6, 96.4) |
| Testing | 0.693 (0.573, 0.812) |
68.9 − |
56.0 (32.0, 100.0) |
73.1 (28.2, 100.0) |
40.0 (30.0, 100.0) |
83.8 (81.0, 100.0) |
||
| Peri | Training | 0.920 (0.882, 0.958) |
87.3 − |
82.6 (72.8, 93.5) |
89.1 (80.3, 96.7) |
74.5 (63.6, 89.9) |
93.0 (90.2, 97.0) |
|
| Testing | 0.724 (0.609, 0.839) |
70.9 − |
64.0 (48.0, 100.0) |
73.1 (38.5, 91.0) |
43.2 (32.4, 66.7) |
86.4 (83.5, 100.0) |
||
| Cmb | Training | 0.940 (0.908, 0.973) |
87.1 − |
92.3 (81.3, 96.7) |
85.2 (82.3, 95.9) |
70.0 (66.1, 88.1) |
96.7 (92.8, 98.6) |
|
| Testing | 0.886 (0.831, 0.942) |
83.3 − |
87.5 (72.5, 97.5) |
81.8 (75.5, 91.8) |
63.6 (55.7, 78.7) |
94.7 (90.0, 98.9) |
||
ACC, AUC, SEN, SPE, PPT, and NPV are short for accuracy, area under the receiver operating characteristic curve, sensitivity, specificity, positive prediction value, and negative prediction value, respectively. Intra, Peri and Cmb indicate the intratumoral model, the peritumoral model and the combined−region model, respectively. Statistical quantifications were demonstrated with 95% confidential interval (CI), when applicable.