Gray-level co-occurrence matrix (GLCM)

The GLCM is a mathematical method used for statistical 2D texture analysis proposed by Haralick et al.³⁴ GLCM corresponds to a directional pattern counter with a specific distance d and angle θ between neighboring pixel pairs for the grayscale image. GLCM computation was performed in 4 directions: 0, 90, 180, and 270 with a distance parameter of 1. The corresponding displacement vectors are [0 1], [−1 1], [−1 0], and [−1 −1].³⁵ In the proposed work, an averaged four-direction value was used for feature extraction to avoid dependence on the direction.³⁶ Five different GLCM features were quantified: energy or angular second moment (ASM), entropy (ENT), homogeneity or inverse difference moment (IDM), correlation (COR), and contrast (CON):

Angular second moment (ASM), also known as energy, maybe an indicator of grey level uniformity or homogeneity (i.e., similarity).²⁵ When the image is homogeneous, the ASM will have a high value.³²

Entropy (ENT) is a measure of textural disorder within the analyzed structure.²⁵ A homogeneous image will result in a lower entropy value.³²

Inverse difference moment (IDM) may be an indirect parameter of textural homogeneity (i.e., similarity) and is associated with pixel pairs.²⁵ Higher values indicate a homogeneous image.³²

Correlation (COR) explains the relationship between a pixel and its neighbor over the whole image. Higher values can be obtained for similar gray-level regions.³²

Contrast (CON) is a measure of the intensity variation between a pixel and its neighbor over the whole image.³⁵ The greater the variation (i.e., gray level dispersion) in an image, the greater the contrast.

These parameters summarize important information about the structural arrangement of surfaces by discerning likelihoods that pixels have the same or different gray-level values as their neighbors and distances between pixel pairs of equal intensity. A full description of the GLCM process can be found in Watanabe et al.²⁵ Healthy skeletal muscle is largely homogenous in texture.³⁷ A proposed schematic of how image texture (homo/heterogeneity) corresponds to these parameters can be found in Figure 2.

Image examples taken from scans of patients included in the current analysis—the image used as an example as high homogeneity was taken from the scan with the lowest entropy and highest ASM; the image used as an example as high heterogeneity was taken from the scan with the highest entropy and lowest ASM.

ASM = energy or angular second moment; IDM = homogeneity or inverse difference moment.

Full analysis was repeated for n = 5 patients to provide reliability data for the assessor (JA). No differences were seen for each GLCM parameter when re-assesed (p = .178 to .708) with intra-class correlation coeffiecnts between .277 and .951. Data can be found in Supplemental Material 1.