Table 3 C-statistics and 95% confidence intervals (CI) for fivefold cross-validated Cox proportional hazard models built from harmonized data, and log-rank p-values for Kaplan–Meier curve separation. ComBat (CE) indicates data was harmonized by contrast enhancement with ComBat.

From: Generalized ComBat harmonization methods for radiomic features with multi-modal distributions and multiple batch effects

 

5fold CV c-statistic

95% CI

Log-rank p-value

Lung3/CAPTK

Original

0.59

[0.53, 0.65]

0.0004

ComBat (CE)

0.60

[0.54, 0.64]

0.0029

Nested

0.63

[0.59, 0.67]

0.0025

NestedD

0.63

[0.58, 0.68]

0.0058

GMM

0.50

[0.42, 0.56]

0.011

GMM + ComBat (CE)

0.50

[0.42, 0.56]

0.036

Lung3/PyRadiomics

Original

0.62

[0.57, 0.66]

0.061

ComBat (CE)

0.61

[0.57, 0.66]

0.029

Nested

0.62

[0.56, 0.67]

0.022

NestedD

0.64

[0.58, 0.69]

0.0062

GMM

0.59

[0.53, 0.65]

0.01

GMM + ComBat (CE)

0.58

[0.52, 0.63]

0.016

Radiogenomics/CAPTK

Original

0.55

[0.52,0.63]

0.071

ComBat (CE)

0.55

[0.50,0.59]

0.056

Nested

0.56

[0.52,0.63]

0.016

NestedD

0.54

[0.53,0.65]

0.074

GMM

0.56

[0.51,0.64]

0.02

GMM + ComBat (CE)

0.58

[0.53,0.64]

0.071

Radiogenomics/PyRadiomics

Original

0.57

[0.52,0.63]

0.17

ComBat (CE)

0.63

[0.53,0.67]

0.0003

Nested

0.62

[0.53,0.69]

0.004

NestedD

0.61

[0.53,0.68]

0.0078

GMM

0.61

[0.51,0.66]

0.0012

GMM + ComBat (CE)

0.63

[0.52,0.68]

0.0002

Back to article page