Skip to main content

Table 3 Comparison of the predictive performance of the five models in predicting Ki-67 expression

From: Radiomic analysis of Gd-EOB-DTPA-enhanced MRI predicts Ki-67 expression in hepatocellular carcinoma

Models AUC (95% CI) Sensitivity (%) Specificity (%) Accuracy (%) PPV (%) NPV (%)
AP radiomics model
Training (n = 103) 0.873 (79.3–93.0) 92.5 (74/80) 78.3 (18/23) 89.3 (92/103) 93.7 (74/79) 75.0 (18/24)
Validation (n = 48) 0.813 (67.4–91.1) 81.3 (26/32) 81.3 (13/16) 81.3 (39/48) 89.7 (26/29) 68.4 (13/19)
Total (n = 151) 0.837 (76.8 89.2) 90.2 (101/112) 69.2 (27/39) 91.4 (138/151) 89.4 (101/113) 71.1 (27/38)
HBP radiomics model
Training (n = 103) 0.813 (72.4–88.3) 98.8 (79/80) 47.8 (11/23) 87.4 (90/103) 86.8 (79/91) 91.7 (11/12)
Validation (n = 48) 0.740 (59.3–85.6) 84.4 (27/32) 62.5 (10/16) 89.6 (43/48) 81.8 (27/33) 66.7 (10/15)
Total (n = 151) 0.793 (72.0-85.5) 87.5 (98/112) 59.0 (23/39) 80.1 (121/151) 86.0 (98/114) 62.2 (23/37)
T2W radiomics model
Training (n = 103) 0.889 (81.2–94.4) 72.5 (58/80) 95.7 (22/23) 77.7 (80/103) 98.3 (58/59) 50.0 (22/44)
Validation (n = 48) 0.698 (54.9–82.2) 90.6 (29/32) 43.8 (7/16) 75.0 (36/48) 76.3 (29/38) 70.0 (7/10)
Total (n = 151) 0.823 (75.2, 88.0) 67.9 (76/112) 68.4 (27/39) 68.2 (103/151) 86.4 (76/88) 42.9 (27/63)
Combined AP and HBP radiomics model     
Training (n = 103) 0.880 (0.802–0.936) 86.2 (69/80) 82.6 (19/23) 70.9 (73/103) 78.4 (69/88) 26.7 (4/15)
Validation (n = 48) 0.799 (0.658–0.901) 75.0 (24/32) 75.0 (12/16) 75.0 (36/48) 85.7 (24/28) 60.0 (12/20)
Total (n = 151) 0.852 (78.5, 90.4) 83.9 (94/112) 76.9 (30/39) 82.1 (124/151) 91.3 (94/103) 62.5 (30/48)
Combined modela
Training (n = 103) 0.922 (0.852–0.965) 98.7 (79/80) 78.3 (18/23) 94.2 (97/103) 94.0 (79/84) 94.7 (18/19)
Validation (n = 48) 0.863 (73.3–94.5) 90.6 (29/32) 75.0 (12/16) 85.4 (41/48) 87.9 (29/33) 80.0 (12/15)
Total (n = 151) 0.806 (73.4–86.6) 83.0 (93/112) 64.1 (25/39) 78.1 (118/151) 86.9 (93/107) 56.8 (25/44)
  1. AP alpha-fetoprotein, AUC area under receiver operating characteristic curve, HBP hepatobiliary phase, NPV negative predictive value, PPV positive predictive value
  2. aCombined model includes AP Rad-score and serum AFP level