Diagnostic Value of Machine Learning-Based Quantitative Texture Analysis in Differentiating Benign and Malignant Thyroid Nodules

<div>A total of seven texture features were selected for the final model: one histogram (HistPerc 99), one HOG (HogO8b2), four GRLM (GrlmHRLNonUni, GrlmHMGLevNonUni, GrlmNRLNonUni, and GrlmZRLNonUni), and one GLCM (GlcmZ3AngScMom). The information gain attribute evaluator identified that GrlmZRLNonuni was the most important feature in the final model followed by HogO8b2 and GrlmNRLNonUni. The formula of the information gain attribute evaluator was InfoGain(Class, Attribute) = <i>H</i>(Class) − <i>H</i>(Class | Attribute), where <i>H</i> represents the amount of information in a unit called bits and ranges in value between 0 and 1.</div>

Journal of Oncology

fig3

Figure 3

Figure 3: Diagnostic Value of Machine Learning-Based Quantitative Texture Analysis in Differentiating Benign and Malignant Thyroid Nodules 