A feasibility study of high order texture features with application to pathological diagnosis of colon lesions for CT Colonography

Abstract

Differentiation of colon lesions into different pathological phases, e.g., neoplastic and non-neoplastic, is of fundamental importance for patient management. Image intensity-based textural features have been recognized as a useful biomarker for the differentiation task. In this paper, we introduce high order texture features, beyond the intensity, such as gradient and curvature, for that task. We expand the 2D Haralick model to 3D and extract texture feature from high order 3D images. These texture features from image intensity, gradient, and curvature were validated on a database, which consists of 148 lesions where 35 are non-neoplastic and 113 are neoplastic lesion, using the well-known support vector machine (SVM) classifier and the merit of area under the ROC curve (AUC). The AUC of classification was improved from 0.74 (by the use of the image intensity-alone) to 0.85 (by also considering the gradient and curvature images) in differentiating the non-neoplastic lesions from neoplastic ones, e.g., the hyperplastic polyps from the tubular adenoma, tubulovillous adenoma and adenocarcinoma lesions. The experimental results demonstrated that texture features from high order images can significantly improve the classification accuracy in differentiating benign from malignant colon lesions.