Table 2 Results of univariate (A, B) and multivariate analysis (C) using three classification models: logistic regression analysis, naive Bayes classifier, and k-NN

From: Predicting breast cancer response to neoadjuvant chemotherapy using pretreatment diffuse optical spectroscopic texture analysis

A

      

Significant univariate feature

Classifier/Model

%Sn

%Sp

AUC

P -value

Statistical power ( n2)

 

Logistic regression

60.0

60.0

0.726

  

Hb-homogeneity

Naive Bayes

82.0

82.0

0.799

0.030

71.8 (14)

 

k-NN

61.5

67.5

0.577

  
 

Logistic regression

70.0

70.0

0.756

  

HbO2-correlation

Naive Bayes

80.0

81.0

0.778

0.024

78.9 (11)

 

k-NN

66.5

74.5

0.602

  
 

Logistic regression

60.0

60.0

0.657

  

HbT-homogeneity

Naive Bayes

84.0

85.0

0.813

0.047

79.9 (11)

 

k-NN

74.0

47.0

0.552

  
 

Logistic regression

60.0

63.0

0.670

  

St-contrast

Naive Bayes

79.5

82.0

0.779

0.044

73.5 (13)

 

k-NN

70.5

64.5

0.582

  
 

Logistic regression

70.0

63.0

0.715

  

StO2-contrast

Naive Bayes

83.0

85.5

0.803

0.044

85.6 (enough)

 

k-NN

70.0

66.5

0.610

  

B

    

Univariate features

Classifier/Model

%Sn

%Sp

%Acc

HbO2-correlation

Logistic regression

70.0

70.0

70.0

HbO2-homogeneity

Naive Bayes

86.5

89.0

87.8

HbO2-contrast

k-NN

81.0

73.0

77.0

C

    

Multivariate features

Classifier/Model

%Sn

%Sp

%Acc

HbO2-correlation+Hb-homogeneity

Logistic regression

80.0

78.0

79.5

Hb-contrast+HbO2-homogeneity

Naive Bayes

78.0

81.0

79.5

Hb-correlation+HbO2-contrast

k-NN

79.5

76.0

77.8

  1. Abbreviations: %Acc=accuracy; AUC=area under curve; Hb=deoxy-haemoglobin; HbO2=oxy-haemoglobin; HbT=total haemoglobin; k-NN=k-nearest neighbour; Sn=sensitivity; Sp=specificity; St=oxygen desaturation; StO2=tumour oxygen saturation.
  2. Bold values indicate best classifiers. The last column in Table 2A reports the percentage of the statistical power. The numbers inside parentheses in this column indicate the number of non-responders (n2) required in this study to achieve a statistical power of minimum 80% in case that the number of responders (n1) is fixed at 27.
Back to article page