Skip to main content
. 2020 May 28;10:680. doi: 10.3389/fonc.2020.00680

Table 3.

Diagnostic performance of three-way classifiers and radiologists in the malignancy validation set.

Value (95%CI) McNemar's P-Valuea Delong P-Value
Model E Model F Model G Reader consensus Model G vs. Reader Model E vs. Model G Model F vs. Model G
HCC
 AUC 0.879 (0.808, 0.949) 0.972 (0.938, 1.000) 0.951 (0.906, 0.997) 0.002 0.792
 Sensitivity, % 93.6 (82.5, 98.7) 95.7 (85.5, 99.5) 95.7 (85.5, 99.5) 89.1 (76.4, 96.4) 0.289
 Specificity, % 67.3 (52.9, 79.7) 96.2 (86.8, 99.5) 90.4 (79.0, 96.8) 90.4 (79.0, 96.8) 0.063
Metastatic malignancy
 AUC 0.814 (0.722, 0.907) 0.980 (0.947, 1.000) 0.985 (0.958, 1.000) 0.0002 0.403
 Sensitivity, % 59.5 (42.1, 75.3) 100 (90.5, 100) 94.6 (81.8, 99.3) 89.2 (74.6, 97.0) 0.688
 Specificity, % 93.6 (84.3, 98.2) 96.8 (88.8, 99.6) 100 (94.2, 100) 95.1 (86.3, 99.0) 1.000
Primary malignancy except HCC
 AUC 0.761 (0.613, 0.909) 0.989 (0.951, 1.000) 0.905 (0.801, 1.000) 0.008 0.081
 Sensitivity, % 53.3 (26.6, 78.7) 86.7 (59.5, 98.3) 73.3 (44.,9 92.2) 60.0 (32.3, 83.7) 0.688
 Specificity, % 95.2 (88.3, 98.7) 100 (95.7, 100) 96.4 (89.9, 99.3) 91.6 (83.4, 96.5) 0.250
a

P value was calculated between Model G (three-sequence images + clinical data) vs. Reader consensus using the McNemar's test.