Table 3 The top features across 500 bootstrap iterations for the model with the highest AUC. The correlation coefficient is the Rank-Biserial correlation coefficient. The AUC was obtained by normalizing the feature values between 0 and 1 and using them as the confidences, and the sensitivity and specificity were recorded at the operating point corresponding to the upper left corner of the curve.

From: Distinguishing recurrence from radiation-induced lung injury at the time of RECIST progressive disease on post-SABR CT scans using radiomics

Name

Equation

Offset†

Correlation coefficient

p-value

AUC

Sensitivity (%)

Specificity (%)

Skewness

\(\frac{{mean\left(I\right)}^{3}}{{stdev\left(I\right)}^{3}}\)

–

 − 0.26

0.032

0.70*

65

82

GLCM max

\({\text{max}}\left(GLCM\right)\)

0.38

0.0015

0.56

36

98

GLCM joint energy

\(\sum {GLCM}^{2}\)

0.30

0.013

0.53

32

93

Mean

\(mean\left(I\right)\)

–

0.31

0.011

0.68

68

80

  1. *As skewness is anti-correlated with outcome, the outcomes were reversed for the AUC calculation.
  2. \(I\): all voxel intensity values within the ROI. GLCM: Grey Level Co-occurrence Matrix.
  3. †The offset refers to the direction of the GLCM calculation. In the diagrams provided in this column, the center grey squares denote the current pixel of interest while the black squares denote the neighbors of that pixel that are being considered.
  4. Significant values are in italic.
Back to article page