Multi-scale characterizations of colon polyps via computed tomographic colonography

Abstract

Texture features have played an essential role in the field of medical imaging for computer-aided diagnosis. The gray-level co-occurrence matrix (GLCM)-based texture descriptor has emerged to become one of the most successful feature sets for these applications. This study aims to increase the potential of these features by introducing multi-scale analysis into the construction of GLCM texture descriptor. In this study, we first introduce a new parameter - stride, to explore the definition of GLCM. Then we propose three multi-scaling GLCM models according to its three parameters, (1) learning model by multiple displacements, (2) learning model by multiple strides (LMS), and (3) learning model by multiple angles. These models increase the texture information by introducing more texture patterns and mitigate direction sparsity and dense sampling problems presented in the traditional Haralick model. To further analyze the three parameters, we test the three models by performing classification on a dataset of 63 large polyp masses obtained from computed tomography colonoscopy consisting of 32 adenocarcinomas and 31 benign adenomas. Finally, the proposed methods are compared to several typical GLCM-texture descriptors and one deep learning model. LMS obtains the highest performance and enhances the prediction power to 0.9450 with standard deviation 0.0285 by area under the curve of receiver operating characteristics score which is a significant improvement.

Keywords: Colon cancer; Computed tomographic colonography; Polyp characterization; Texture feature.

Fig. 1

Polyp heterogeneity and texture in computed tomographic colonography. The green curves are their boundaries plotted by radiologists. The air in polyp is labeled by red color

Fig. 2

Illustration of co-occurrence matrix of two-dimensional images. a: Two-dimensional GLCM calculation; b: Three-dimensional GLCM calculation; c: A GLCM example when angle(θ) = 0°, displacement = 1. GLCM: Gray-level co-occurrence matrix

Fig. 3

Calculation of multi-scale gray level co-occurrence matrix where d represents displacement, θ is the angle, s is stride (or scale), p₀ is the concerned point

Fig. 4

MSGLCM calculation by displacement samplings: (a) displacement = 1, (b) displacement = 2, (c) displacement = 3

Fig. 5

ASGLCM calculation on one slice of a volume, containing 13 directions in the 3D space: (a) stride = 2, displacement = 1, (b) stride = 2, displacement = 2, (c) stride = 2, displacement = 3, (d) stride = 3, displacement = 1, (e) stride = 3, displacement = 2, (f) stride = 3, displacement = 3

Fig. 6

Two cases of multiple angle sampling: (a) multiple angle sampling with duplicates, (b) multiple angle sampling without duplicates

Fig. 7

AUC score trends with the stride increasing for LMD + LMS. “*” indicates the best result position of each curve. AUC: Area under the curve of receiver operating characteristic curve; LMS: Learning model by multiple strides; LMD: Learning model by multiple displacements

Fig. 8

Receiver operating characteristic curves of six methods via random forest except VGG16. HF: Haralick feature; eHM: Extended Haralick measure; KLT: Karhunen-Loeve transform; LMD: Learning model by multiple displacements; LMS: Learning model by multiple strides; LMA: Learning model by multiple angles; CoLIAGe: Co-occurrence of local anisotropic gradient orientations