Table 2.
Performance summary of radiologists and different models for prediction of ALN metastasis
| Methods | AUC | ACC(%) | SENS(%) | SPE(%) | PPV(%) | NPV(%) | P value | |
|---|---|---|---|---|---|---|---|---|
| Pooled Radiologists | 0.703 (0.672-0.735) | 72 (69-75) | 63 (58-69) | 77 (74-81) | 62 (57-67) | 78 (74-82) | <0.001* | |
| EVC1 | CLI | 0.769 (0.677-0.860) | 75 (66-83) | 53 (38-68) | 90 (80-96) | 77 (59-90) | 74 (63-83) | <0.001* |
| Signature | 0.886 (0.815-0.958) | 87 (79-92) | 87 (73-95) | 87 (76-94) | 81 (67-91) | 91 (81-96) | 0.169 | |
| DLRN | 0.914 (0.858-0.971) | 88 (80-93) | 87 (73-95) | 88 (78-95) | 83 (69-92) | 91 (81-97) | ||
| EVC2 | CLI | 0.783 (0.687-0.880) | 75 (64-83) | 69 (51-83) | 79 (65-89) | 69 (51-83) | 79 (65-89) | <0.001* |
| Signature | 0.854 (0.778-0.931) | 72 (62-81) | 100 (90-100) | 54 (39-68) | 59 (46-72) | 100 (88-100) | 0.011* | |
| DLRN | 0.929 (0.877-0.980) | 87 (79-94) | 86 (70-95) | 88 (77-96) | 83 (67-94) | 90 (79-97) | ||
| EVC3 | CLI | 0.700 (0.570-0.830) | 68 (55-79) | 62 (44-79) | 74 (55-88) | 71 (51-87) | 66 (48-81) | <0.001* |
| Signature | 0.917 (0.854-0.980) | 83 (71-91) | 91 (75-98) | 74 (55-88) | 78 (62-90) | 88 (70-98) | 0.188 | |
| DLRN | 0.952 (0.906-0.997) | 89 (78-95) | 81 (64-93) | 97 (83-100) | 96 (81-100) | 83 (67-94) | ||
EVC external validation cohort, CLI clinical model, DLRN deep learning radiomic nomogram, AUC the area under the receiver operating characteristic curve, ACC accuracy, SEN sensitivity, SPE specificity, PPV positive prediction value, NPV negative prediction value.
Statistical quantifications were demonstrated with 95% confidential interval (CI), when applicable. The P value indicates the comparison between AUCs of each method and the integrated DLRN by the DeLong test in different cohorts.