Volumetric texture features from higher-order images for diagnosis of colon lesions via CT colonography

Abstract

Purpose: Differentiation of colon lesions according to underlying pathology, e.g., neoplastic and non-neoplastic lesions, is of fundamental importance for patient management. Image intensity-based textural features have been recognized as useful biomarker for the differentiation task. In this paper, we introduce texture features from higher-order images, i.e., gradient and curvature images, beyond the intensity image, for that task.

Methods: Based on the Haralick texture analysis method, we introduce a virtual pathological model to explore the utility of texture features from high-order differentiations, i.e., gradient and curvature, of the image intensity distribution. The texture features were validated on a database consisting of 148 colon lesions, of which 35 are non-neoplastic lesions, using the support vector machine classifier and the merit of area under the curve (AUC) of the receiver operating characteristics.

Results: The AUC of classification was improved from 0.74 (using the image intensity alone) to 0.85 (by also considering the gradient and curvature images) in differentiating the neoplastic lesions from non-neoplastic ones, e.g., hyperplastic polyps from tubular adenomas, tubulovillous adenomas and adenocarcinomas.

Conclusions: The experimental results demonstrated that texture features from higher-order images can significantly improve the classification accuracy in pathological differentiation of colorectal lesions. The gain in differentiation capability shall increase the potential of computed tomography colonography for colorectal cancer screening by not only detecting polyps but also classifying them for optimal polyp management for the best outcome in personalized medicine.

Keywords: CT colonography; Colorectal lesions; Computer-aided diagnosis; Curvature; Gradient; Textural biomarker; Texture feature.

Fig. 1

Overview of our proposed 3D texture feature extraction scheme

Fig. 2

Illustration of 2D Haralick spatial information extraction model with d = 1 and d = 2 neighbor and four directions

Fig. 3

Illustration of 3D expansion of the Haralick model. The 3D frame shows the 26 neighbors (d = 1) of the red cube. Graph a shows the four directions in the middle layer. Graphs b, c show the rest nine directions in top layer. Graphs d–f show the direction number corresponding to a–c

Fig. 4

Illustration of semi-automatic volume extraction. The first column shows the locations of lesions in 2D slices. The second column shows the semi-automatically extracted region of interest in 2D view. The green line indicates the manually drawn boundary, and the red border indicates the interface between air and polyp after automatic air removal. The third column shows the 3D endoscopic views using the software of Viatronix. Row a shows a 7mm tubular adenoma located in sigmoid colon. Row b shows a 8mm hyperplastic polyp located in descending colon. Row c shows a 9mm tubular adenoma located in hepatic flexure. Row d shows a 20mm tubulovillous adenoma in ascending colon

Fig. 5

Distribution information of the database used in study. Graph a shows the histogram of linear size information in the pathology report. Graph b is the histogram of pathological phase of the 148 lesions

Fig. 6

Overview of our texture feature set evaluation scheme

Fig. 7

Regression scatter plot shows the linear relationship between the third root of extracted 3D volume voxel number and its corresponding linear size in the pathology report

Fig. 8

ROC curve with only lesion size information as decision value. Graph a shows the ROC curve based only on linear size differentiation. Graph b shows the ROC curve based only on 3D volume size differentiation

Fig. 9

Histogram of 3D volume size distribution within different groups, i.e., H against TA&TVA&A

Fig. 10

Averaged ROC curves from SVM classification results. The curves were conducted according to the horizontal axis, where the linear interpolation was employed when needed