Histologic classification of colonic polyps based on fractal dimension analysis: comparison of results using support vector machine and logistic regression

Abstract

The aim of this study was to evaluate fractal analysis as a tool for differentiating between normal tissue and adenomatous polyp lesions. Images of colon samples from 140 patients were analyzed. There were 70 subjects in each of the normal and polyp groups. Two texture features based on fractal analysis were studied: fractal dimension (FD) and lacunarity (Lac), extracted using the overlapping box-counting method. The proposed classification models based on fractal analysis of normal colon and abnormal polyp images were performed using two classification methods: the support vector machine (SVM) and the logistic regression (LR). Several widely-recalled statistical metrics (accuracy, sensitivity, specificity and precision) were used to evaluate the global model performance. To avoid any overfitting problems, all models were evaluated using a 10-fold cross-validation. The SVM method showed better performance in detecting normal colon images than the LR method. As a result, the SVM method provided results with higher accuracy (ACC) and specificity than the LR method (ACCSVM=0.90 vs. ACCLR=0.75). These results give confidence for developing a practical automated analysis technique for detecting colon polyps.