Table 2 Diagnostic performance of the best performing machine learning model in the training set and the test set.

From: Differentiation of recurrent glioblastoma from radiation necrosis using diffusion radiomics with machine learning model development and external validation

Sequence

Feature selection

No. of selected features

Classification

Training set

Test set

AUC (95% CI)

Accuracy (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

P-value*

AUC (95% CI)

Accuracy (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

P-value*

ADC

LASSO

18

SVM

0.90 (0.84–0.95)

80.5 (77.4–83.6)

78.3 (64.2–92.4)

82.9 (74.7–91.1)

Reference

0.80 (0.65–0.95)

78.0 (62.4–89.4)

66.7 (41.0–86.7)

87.0 (66.5–97.2)

Reference

T2

LASSO

21

SVM

0.86 (0.80- 0.91)

77.1 (74.1–80.1)

80.7 (70.8–90.6)

73.1 (66.0–80.2)

0.346

0.65 (0.48–0.82)

61.0 (44.5–75.8)

44.4 (21.5–69.2)

73.9 (51.6–89.9)

0.186

T1C

MI

30

SVM

0.91 (0.86–0.95)

87.4 (84.5–90.3)

90.7 (83.0–98.4)

84.3 (78.2–90.4)

0.798

0.66 (0.49–0.83)

53.7 (37.4–69.3)

11.1 (1.4–34.7)

87.0 (66.4–97.2)

0.217

ADC + T2 + T1C

LASSO

35

SVM

0.93 (0.89–0.97)

85.2 (82.0–88.4)

79.8 (71.2–88.4)

90.5 (83.0–98.0)

0.405

0.66 (0.49–0.84)

63.4 (46.9–77.9)

38.9 (17.3–64.3)

82.6 (61.2–95.0)

0.217

  1. All training set performance was calculated on SMOTE generated datasets.
  2. CI confidence interval, LASSO least absolute shrinkage and selection operator, MI mutual information, SMOTE synthetic minority over-sampling technique, SVM support vector machine, T1C postcontrast T1WI, T2 T2WI.
  3. *P-value refers to the significance among the differences of the AUCs between the ADC radiomics model and the other models.
Back to article page