Skip to main content

Table 7 Performance of five different machine learning methods for classifying 206 solid breast lesion on ultrasound images

From: Performance of machine learning software to classify breast lesions using BI-RADS radiomic features on ultrasound images

Method Features Sensitivity Specificity AUC
Point estimate (%) 95% CI Point estimate (%) 95% CI Point estimate 95% CI
Decision tree {6, 1, 2, 8} 70.6 0.5889–0.8008 75.0 0.6231–0.8448 0.744 0.677–0.774
Multilayer perceptron {5, 2, 1, 3} 66.2 0.5462–0.7612 71.7 0.5843–0.8203 0.806 0.677–0.839
Random forest {4, 8, 5, 2} 72.7 0.5983–0.8181 75.9 0.593–0.811 0.811 0.710–0.892
Linear discriminant analysis {5, 2} 76.0 0.6212–0.8345 69.8 0.6156–0.8316 0.818 0.6667–0.9444
Support vector machine {4, 2, 10, 8, 1} 71.4 0.6479–0.8616 76.9 0.6148–0.8228 0.840 0.6667–0.9762
  1. {1}: area difference with equivalent ellipse; {2}: orientation; {3}: average of difference vector; {4}: number of peaks on the distance vector; {5}: average of distance vector; {6}: area difference between the convex hull and tumour; {7}: echogenicity; {8}: entropy; {9}: shadow; {10}: lesion size. CI Confidence interval