Skip to main content

Table 8 AUC, accuracy, sensitivity, and specificity values from comparative methods and our proposed method where HF represents Haralick feature

From: Vector textures derived from higher order derivative domains for classification of colorectal polyps

Method

AUC

Accuracy

Sensitivity

Specificity

eHF

0.876

0.807

0.858

0.757

HOG3D

0.804

0.713

0.726

0.700

CoLIAGe

0.923

0.836

0.839

0.833

VGG-16

0.833

0.740

0.709

0.771

T1

0.948

0.868

0.883

0.853

T2

0.950

0.868

0.823

0.913

T3

0.949

0.863

0.847

0.879

T4

0.962

0.922

0.884

0.960

  1. Overall, the T1, T2, T3 and T4 derived texture features of our method achieved much higher AUC values than the four typical methods. The ROC curves of our four angular texture feature extraction methods are shown Fig. 6, which provides a visual assessment on their performances with comparison to other four references of HF, HOG, CoLIAGe and VGG-16