Skip to main content

Table 4 Robustness comparison of final models across all five folds and choice of best model

From: A radiographic, deep transfer learning framework, adapted to estimate lung opacities from chest x-rays

Model Architecture

Data Balancing Strategy

Segmentation Strategy

Precision Mean (stdev)

Recall

Mean (stdev)

F1 Mean (stdev)

MAE Mean (stdev)

MA HCS Mean

CheXNet-121

DOUBLE

NONE

65.89 (1.28)

66.30 (1.13)

65.82 (1.18)

0.3944 (0.0147)

0.1481

CheXNet-121

UNDER

NONE

65.36 (1.26)

65.47 (1.15)

65.06 (1.32)

0.3930 (0.0126)

0.1352

ResNet-50

UNDER

NONE

64.88 (1.10)

64.81 (0.82)

64.33 (0.85)

0.4099 (0.0147)

0.1830

ResNet-50

DOUBLE

SPINE

64.33 (1.21)

64.37 (0.99)

63.90 (1.19)

0.4149 (0.0157)

0.2057

VGG-16

DOUBLE

SPINE

62.02 (1.31)

61.18 (1.27)

61.44 (1.27)

0.4325 (0.0182)

0.1631