Table 1 (A) Percentage of features with significantly different distributions attributable to contrast enhancement, spatial resolution due to reconstruction kernel, and manufacturer in the original features and after applying standard ComBat, Nested ComBat, and NestedD (dropping with every iteration) ComBat in the CapTK features extracted from the Lung3 dataset. (B) Corresponding table for PyRadiomics features extracted from the Lung3 data. (C) Corresponding table for CapTK features extracted from the Radiogenomics dataset. (D) Corresponding table for PyRadiomics features extracted from the Radiogenomics dataset.

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

 

Original (%)

ComBat (%)

Nested (%)

NestedD (%)

A. Lung3/CAPTK

CE

10

16

5

3

Spatial resolution

18

21

23

19

Manufacturer

48

45

41

28

B. Lung3/PyRadiomics

CE

40

11

5

2

Spatial resolution

43

25

29

11

Manufacturer

61

28

27

12

C. Radiogenomics/CAPTK

CE

17

42

33

14

Spatial resolution

42

43

36

11

Manufacturer

20

51

38

15

D. Radiogenomics/PyRadiomics

CE

54

27

26

9

Spatial resolution

69

29

23

13

Manufacturer

44

36

44

20

  1. Tables contain the percentage of features out of the original number of features with detected significant (p < 0.05) differences in distribution for all batch effects.