Skip to main content
Insights into Imaging

Table 4 Prediction performance of the two developed models and the BCLC stage

From: Predictive models and early postoperative recurrence evaluation for hepatocellular carcinoma based on gadoxetic acid-enhanced MR imaging

Models

Training set (n = 120)

Internal validation set (n = 38)

External validation set (n = 51)

AUC

(95% CI)

Sensitivity

Specificity

AUC

(95% CI)

Sensitivity

Specificity

AUC

(95% CI)

Sensitivity

Specificity

RF-Boruta

0.820

(0.738, 0.902)*

66.67%

(24/36)

89.29%

(75/84)

0.857

(0.699, 0.999)**

80.00%

(8/10)

89.28%

(25/28)

0.805

(0.686, 0.924)^

41.67%

(10/24)

92.59%

(25/27)

Logistic-AIC

0.853

(0.775, 0.930)*

83.33%

(30/36)

79.76%

(67/84)

0.812

(0.669, 0.955)

70.00%

(7/10)

83.33%

(23/28)

0.789

(0.665, 0.913)^

75.00%

(18/24)

70.37%

(19/27)

BCLC stage

0.658

(0.553, 0.762)

63.89%

(23/36)

60.71%

(51/84)

0.643

(0.422, 0.863)

60.00%

(6/10)

57.14%

(16/28)

0.633

(0.547, 0.720)

20.83%

(5/24)

100.00%

(27/27)

  1. *:vs the BCLC stage in the training set, p < 0.001; **: vs the BCLC stage in the internal validation set, p = 0.038; ^: vs the BCLC stage in the external validation set, p < 0.05; RF-Boruta, the Random Forests model with Boruta algorithm; Logistic-AIC, the Logistic regression model with stepwise selection (Akaike information criterion); BCLC, Barcelona Clinic Liver Cancer stage
Back to article page