Skip to main content

Table 3 The prediction of prognostic risk factors, gene mutations, and immunoexpression outcomes in the independent test cohort

From: Exploring non-invasive precision treatment in non-small cell lung cancer patients through deep learning radiomics across imaging features and molecular phenotypes

 

Model

Classifier

AUC (95% CI)

Accuracy (%)

Precision (%)

Recall (%)

F1 score (%)

p-value

Prognostic risk factors

 

LVI

Radiomics-Score

NB

0.842 [0.832–0.852]

69.6

71.2

68.8

79.6

0.043

 

Deep-CT

MLP

0.826 [0.816–0.835]

75.1

75.9

74.5

68.8

0.027

 

Deep-RadScore

LDA

0.889 [0.881–0.897]

80.4

80.5

84.8

78.5

Baseline

PI

Radiomics-Score

SVM

0.853 [0.844–0.862]

75.4

76.3

69.8

78.0

0.041

 

Deep-CT

RF

0.840 [0.831–0.848]

78.7

77.8

79.4

70.6

0.045

 

Deep-RadScore

SVM

0.903 [0.896–0.910]

81.6

82.9

84.7

81.5

Baseline

T staging

Radiomics-Score

NB

0.827 [0.817–0.838]

65.6

71.5

72.0

65.6

0.035

 

Deep-CT

LDA

0.834 [0.825–0.844]

76.2

78.0

77.9

72.6

0.018

 

Deep-RadScore

RF

0.894 [0.886–0.901]

81.5

77.6

84.9

78.6

Baseline

Gene mutations

 

EGFR

Radiomics-Score

SVM

0.843 [0.834–0.853]

68.0

70.0

70.7

67.9

0.038

 

Deep-CT

NB

0.838 [0.829–0.847]

74.6

76.2

76.2

74.6

0.019

 

Deep-RadScore

RF

0.884 [0.876–0.892]

81.9

81.5

83.0

81.5

Baseline

KRAS

Radiomics-Score

MLP

0.841 [0.827–0.855]

78.3

68.7

76.2

69.7

0.039

 

Deep-CT

LR

0.830 [0.820–0.841]

69.9

66.7

70.2

66.6

0.017

 

Deep-RadScore

LDA

0.896 [0.886–0.906]

80.3

78.1

82.6

78.8

Baseline

ALK

Radiomics-Score

RF

0.823 [0.808–0.838]

72.8

71.4

79.7

74.2

0.002

 

Deep-CT

RF

0.821 [0.802–0.840]

68.3

66.5

73.5

62.2

0.0001

 

Deep-RadScore

RF

0.884 [0.873–0.895]

83.4

78.6

82.2

78.0

Baseline

TP53

Radiomics-Score

KNN

0.828 [0.817–0.838]

75.8

73.8

75.6

74.2

0.022

 

Deep-CT

SVM

0.833 [0.823–0.844]

73.1

73.7

73.3

73.0

0.048

 

Deep-RadScore

LDA

0.889 [0.880–0.898]

79.5

80.5

83.2

79.3

Baseline

PIK3CA

Radiomics-Score

KNN

0.839 [0.827–0.851]

81.7

72.3

74.3

74.4

0.045

 

Deep-CT

NB

0.824 [0.813–0.835]

73.9

77.8

70.3

70.4

0.0006

 

Deep-RadScore

LDA

0.896 [0.886–0.907]

82.6

86.2

79.2

76.8

Baseline

ROS1

Radiomics-Score

MLP

0.827 [0.806–0.849]

73.3

62.5

77.9

61.9

0.018

 

Deep-CT

RF

0.832 [0.815–0.849]

72.8

66.4

78.7

66.1

0.015

 

Deep-RadScore

NB

0.895 [0.886–0.905]

82.2

78.2

80.4

79.0

Baseline

Immunoexpression

 

PD-1/PD-L1

Radiomics-Score

NB

0.841 [0.824–0.858]

79.2

78.7

77.5

77.8

0.024

Deep-CT

LDA

0.839 [0.823–0.854]

78.4

79.3

79.2

78.4

0.007

 

Deep-RadScore

LR

0.893 [0.882–0.905]

82.1

81.5

81.9

81.6

Baseline

  1. p-value: DeLong test for the difference in AUC between the Deep-RadScore (baseline) model and the Radiomics-Score/Deep-CT models
  2. Abbreviations: CI: confidence intervals; SVM: support vector machine; KNN: k-nearest neighbors; RF: random forests; NB: naive Bayes classifier; LR: logistic regression; MLP: multilayer perceptron; LDA: linear discriminant analysis
Back to article page

Biomarker Research

ISSN: 2050-7771

Contact us