Table 4.
Comparisons using different combinations of hospitals for training DLRE to classify liver fibrosis stages in training and validation cohorts
| n (P) | AUC | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | LR+ | LR− | ||
| Cirrhosis (F4) | |||||||||
| Combo1 | T | 221 (27.6%) | 1.00 (0.99 to 1.00) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
– | 0.0 (0.0 to 0.0) |
| V | 177 (20.9%) | 0.95 (0.92 to 0.98) |
86.5 (81.4 to 91.6) |
94.3 (90.2 to 98.4) |
80.0 (73.8 to 86.4) |
96.4 (89.7 to 98.5) |
15.1 (14.8 to 15.4) |
0.1 (0.1 to 0.2) |
|
| Combo 2 | T | 221 (27.6%) | 1.00 (0.99 to 1.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
– | 0.0 (0.0 to 0.0) |
| V | 177 (20.9%) | 0.98 (0.96 to 0.99) |
93.6 (88.4 to 98.8) |
93.3 (89.4 to 97.2) |
84.6 (79.5 to 89.2) |
97.4 (92.6 to 99.2) |
13.9 (12.1 to 15.7) |
0.1 (0.1 to 0.1) |
|
| Combo 3 | T | 221 (27.6%) | 1.00 (0.99 to 1.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
– | 0.0 (0.0 to 0.0) |
| V | 177 (20.9%) | 0.97 (0.95 to 0.99) |
95.2 (92.1 to 98.3) |
92.4 (88.2 to 96.6) |
81.6 (73.8 to 88.4) |
98.2 (94.8 to 99.7) |
12.6 (117 to 13.5) |
0.1 (0.0 to 0.1) |
|
| Advanced fibrosis (≥F3) | |||||||||
| Combo 1 | T | 221 (59.7%) | 0.98 (0.96 to 0.99) |
94.7 (91.2 to 98.2) |
95.5 (91.7 to 99.3) |
96.9 (91.5 to 98.9) |
92.4 (87.5 to 96.5) |
21.1 (20.4 to 21.8) |
0.1 (0.1 to 0.1) |
| V | 177 (52.0%) | 0.97 (0.95 to 0.99) |
92.4 (88.7 to 96.1) |
95.3 (91.9 to 98.7) |
95.5 (90.5 to 97.8) |
92.1 (85.6 to 97.8) |
19.6 (18.3 to 20.8) |
0.1 (0.1 to 0.1) |
|
| Combo 2 | T | 221 (59.7%) | 0.98 (0.97 to 0.99) |
95.8 (92.7 to 98.9) |
93.9 (90.1 to 97.7) |
94.2 (87.5 to 97.8) |
95.5 (88.5 to 97.9) |
15.6 (14.9 to 16.3) |
0.05 (0.0 to 0.1) |
| V | 177 (52.0%) | 0.97 (0.95 to 0.98) |
96.2 (92.6 to 98.7) |
88.3 (83.5 to 93.1) |
93.6 (86.9 to 96.8) |
93.0 (96.4 to 96.4) |
8.3 (8.0 to 8.5) |
0.0 (0.0 to 0.1) |
|
| Combo 3 | T | 221 (59.7%) | 1.00 (0.99 to 1.00) |
96.9 (93.8 to 99.9) |
99.1 (95.8 to 99.9) |
99.2 (97.1 to 100.0) |
96.5 (92.5 to 99.7) |
106.5 (100.2 to 112.0) |
0.0 (0.0 to 0.0) |
| V | 177 (52.0%) | 0.97 (0.96 to 0.99) |
93.8 (89.4 to 98.2) |
92.2 (87.8 to 96.7) |
94.8 (89.5 to 98.6) |
90.8 (84.5 to 96.2) |
12.0 (10.9 to 13.2) |
0.1 (0.1 to 0.1) |
|
| Significance fibrosis (≥F2) | |||||||||
| Combo 1 | T | 221 (87.3%) | 1.00 (0.99 to 1.00) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
– | 0.0 (0.0 to 0.0) |
| V | 177 (79.1%) | 0.83 (0.79 to 0.87) |
84.3 (80.1 to 88.5) |
70.3 (66.8 to 73.9) |
91.5 (85.7 to 97.2) |
54.2 (49.5 to 59.2) |
2.8 (2.6 to 3.1) |
0.2 (0.2 to 0.2) |
|
| Combo 2 | T | 221 (87.3%) | 1.00 (0.99 to 1.00) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
– | 0.0 (0.0 to 0.0) |
| V | 177 (79.1%) | 0.86 (0.81 to 0.91) |
80.8 (76.2 to 85.4) |
80.0 (74.6 to 85.4) |
97.6 (92.8 to 99.6) |
29.3 (27.1 to 31.5) |
4.0 (3.8 to 4.2) |
0.2 (0.2 to 0.3) |
|
| Combo 3 | T | 221 (87.3%) | 1.00 (0.99 to 1.00) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
100.0 (99.0 to 100.0) |
– | 0.0 (0.0 to 0.0) |
| V | 177 (79.1%) | 0.82 (0.78 to 0.87) |
74.1 (69.8 to 78.4) |
77.8 (72.4 to 83.2) |
96.4 (90.5 to 98.9) |
27.5 (25.8 to 29.1) |
3.3 (3.2 to 3.5) |
0.3 (0.3 to 0.4) |
|
Statistical quantifications were demonstrated with 95% CI, when applicable.
AUCs obtained by three different combinations of hospitals were statistically compared with each other in each classification and each cohort (*P<0.05; **P<0.01; ***P<0.001).
AUC, area under the receiver operating characteristic curve; Combo, combination of hospitals for training and validation cohorts; DLRE, deep learning Radiomics of elastography; LR+, positive diagnostic likelihood ratio; LR−, negative diagnostic likelihood ratio; n, number of patients; NPV, negative predictive value; P, prevalence; PPV, positive predictive value; T, training cohort; V, validation cohort.