Table 2 Performance of four different classifiers with three different feature selection methods in the modeling set

From: An oral cavity squamous cell carcinoma quantitative histomorphometric-based image classifier of nuclear morphology can risk stratify patients for disease-specific survival

Classifier

Feature selection

AUC

Accuracy

Specificity

Sensitivity

LDA

WRST

0.74±0.07

0.83±0.03

0.91±0.10

0.67±0.04

 

MRMR

0.83±0.05

0.79±0.07

0.87±0.08

0.65±0.09

 

RF

0.77±0.05

0.81±0.03

0.91±0.10

0.62±0.06

QDA

WRST

0.87±0.02

0.88±0.02

0.93±0.10

0.78±0.04

 

MRMR

0.81±0.04

0.84±0.05

0.88±0.14

0.76±0.04

 

RF

0.83±0.06

0.85±0.3

0.91±0.15

0.72±0.06

RF

WRST

0.81±0.05

0.77±0.04

0.87±0.06

0.59±0.02

 

MRMR

0.84±0.03

0.81±0.04

0.87±0.06

0.68±0.06

 

RF

0.78±0.04

0.74±0.04

0.83±0.13

0.58±0.05

SVM

WRST

0.86±0.02

0.82±0.03

0.93±0.06

0.62±0.04

 

MRMR

0.79±0.07

0.72±0.04

0.90±0.15

0.35±0.06

 

RF

0.84±0.02

0.79±0.02

0.92±0.08

0.53±0.03

  1. Abbreviations: AUC, area under curve; LDA/QDA, linear/quadratic discriminant analysis; MRMR, minimum redundancy, maximum relevance feature selection method; RF, random forest; SVM, support vector machine; WRST, Wilcoxon rank sum test.
  2. The best performance in each metric/column is shown in bold.
Back to article page