New texture features for improved differentiation of hyperplastic polyps from adenomas via computed tomography colonoscopy

Feature classification plays an important role in computer-aided diagnosis (CADx) of suspicious lesions. While many texture features have been extracted and applied for various clinical purposes, Haralick's feature extraction method is of great interest, because it gives a series of texture measures on the image intensity correlations among the image pixels across an image slice. Based on the Haralick's method, we proposed a new set of features for CADx of colonic polyps or differentiation of hyperplastic polyps from adenomas. We evaluated this new feature set by means of random forest (RF) classifiers on a database of 153 polyps, including 116 adenomas and 37 hyperplastic polyps. The classification results were documented quantitatively by the Receiver Operating Characteristics (ROC) analysis and the merit of area under the ROC curve (AUC), which are well-established evaluation criteria to various classifiers. Experimental results demonstrated that the new feature set significantly improved the CADx performance for colonic polyps.