Fractal analysis and texture features of vascular networks observed in colonoscopy images as novel image classifiers for clinical metabolic syndrome

摘要

In the study, we aim to analyze the texture characteristics of colonic vascular networks observed on colonoscopy, in order to test and set up biomedical imaging markers for clinical metabolic syndrome. We retrospectively selected clear colonoscopic images from 30 clinical metabolic syndrome subjects and 30 healthy control subjects. Regions of interest containing vascular networks of major trunk and primary branches were adopted. White light reflection points were detected and eliminated. With the wavelet transform, the fractal dimensions in different subbands are determined. In addition, six more statistical features are combined with the fractal dimension in the support vector machine (SVM) to differentiate two different groups of subjects. Experimental results show that 93.94%-96.97% classification accuracy can be achieved when using the combination of the fractal dimension and other different statistical feature.