Textural Analysis of Mandibular Trabecular Bone Based on Grey Level Co-occurrence Matrix Features in Correlation With Age and Gender of an Egyptian Population Sample: A Cross-sectional Study Using Cone Beam CT

Scientific background

Bone health of the elderly is a major health concern which has gained a lot of attention lately since elderly populations are growing worldwide(Vijayakumar and Büsselberg, 2016). It is expected for Egypt in particular to have 130 million inhabitant by 2050, out of which 30% will be above 50 years of age(Chen et al., 2013; Gheita and Hammam, 2018). Aging is commonly accompanied by musculoskeletal disorders such as osteoporosis and osteoarthritis as a result of disturbances occurring in bone remodeling (Gheno et al., 2012). These disorders greatly increase the risk of bone fractures, such as hip and vertebral fractures, and thus compromising the quality of life of these patients(Gheno et al., 2012). The prevalence of these musculoskeletal disorders is also affected by gender, studies claim that females are more prone to osteoporosis and osteoarthritis than males(on Osteoporosis and Prevention, 2001; Gheno et al., 2012).

Bone mineral density measured by dual x-ray absorptiometry (DXA) has been considered as the gold standard for assessing bone health for a long time(Link and Heilmeier, 2016). However, the national institute of health on its osteoporosis consensus conference pointed out that bone strength depends not only on its density but also on quality(on Osteoporosis and Prevention, 2001). Therefore, complementary diagnostic methods for assessing bone quality are needed to improve the prediction of individual fracture risk(Link and Heilmeier, 2016).

Various attempts has been made by researchers since then to develop a satisfactory method to assess bone quality, trabecular bone score (TBS) is one of the most successful trials. TBS is a texture parameter related to bone microarchitecture that uses conventional DXA images of the lumbar spine to extract the gray-level texture features providing skeletal information that is not captured from the standard BMD measurement. It has been used successfully as an adjunctive to DXA, and further research attempts are made to investigate its use independently to predict fracture risk.(Martineau, Silva and Leslie, 2017; Shevroja et al., 2017) The success of TBS has led the research community to examine the potential use of texture parameters to assess bone microarchitecture in different imaging modalities(Shevroja et al., 2017). Texture is one of the main visual pattern recognition techniques used by humans to identify objects or regions of interest in an image. It is an innate property of each object, containing important information about the structural arrangement of its surface and its relation with surroundings (Haralick, Shanmugam and Dinstein, 1973). Textural analysis is a type of quantitative image assessment based on relationships between pixel intensities. It is gaining a wide range of attention nowadays in the field of biomedical imaging since it can be beneficial in tissue characterization with various diagnostic applications(Gebejes and Huertas, 2013; Summers, 2017).

There are various methods for textural analysis, Grey level co-occurrence matrix (GLCM) is one of the popular and most commonly used for extraction of texture features(Gebejes and Huertas, 2013). It is a second order statistical method of characterizing the texture of an image by considering the frequency of occurring of pixel grey value pairs and their spatial relationship(Haralick, Shanmugam and Dinstein, 1973; Gebejes and Huertas, 2013).

Cone beam computed tomography (CBCT) is nowadays used routinely for multiple diagnostic procedures and is known to use significantly lower radiation dose than CT with higher spatial resolution and at lower cost, therefore could be a suitable candidate as a screening tool for bone quality(Scarfe, Farman and Sukovic, 2006; Scarfe and Farman, 2008). We hereby hypothesize that using grey level co-occurrence matrix derived texture features to assess bone quality can be applicable to CBCT scans, instead of the debatable use of grey levels to measure bone density which in many studies proved to be inadequate in comparison to the calibrated Hounsfield units of multi-detector CT(Pauwels et al., 2013; De Rosa et al., 2020; Gonçalves et al., 2020).

Therefore the aim of this study is to investigate if GLCM texture features are correlated with age-gender trabecular bone variations in different regions of interest on mandibular CBCT scans. This correlation will help in guiding the future selection of textural features and anatomical regions suitable for developing a potential screening method for age and sex related skeletal disorders.

PECO Question:

P: Mandibular trabecular bone of adult Egyptians E: Females (18-40 OR above 40) C: Males (18-40 OR above 40) O: Grey level co-occurrence matrix textural features

Research question:

What is the correlation between the age and sex of Egyptian population and GLCM derived texture features of mandibular trabecular bone?

Statement of the problem:

Bone health of elderly is a major public health concern. Although diagnosisof disorders affecting bone health requires assessing bone quality as well as density, there is still a shortage in diagnostic techniques that could examine bone quality successfully at considerably low radiation dose and low cost.

Specific objectives:

To detect the presence of a correlation between age-sex of an individual and the texture features of mandibular trabecular bone derived from grey level co-occurrence matrix in a sample of adult Egyptian population.

Hypothesis:

Null: There is no difference in trabecular textural features between different age groups or between the two sexes Alternative: There is a difference in trabecular textural features between different age groups or between the two sexes

Methodology Variables

Independent Variables:

1. Age ( 18-40 OR above 40)
2. Sex (Males OR Females)

Dependent Variables:

Eleven Texture features derived from Haralick's work (Haralick, Shanmugam and Dinstein, 1973) Feature Description Contrast Represents the amount of local variation of gray shades Inverse difference moment Homogeneity of the distribution of gray shades on the image Angular Second moment Measurement of image uniformity Correlation Linear measure dependence of gray shades between neighboring pixels Sum of squares Measurement of the dispersion(related to average) of gray shade distribution Entropy Degree of disorder between pixels in the image Sum of average Mean of the distribution of the sum of gray shades Sum of variance Dispersion around the mean of the sum distribution of gray shades Sum of entropy Disorganization of the sum distribution of gray shades Difference of variance Dispersion of the gray shade difference Difference of entropy Disorganization of the gray shade difference

Data sources\Measurements

CBCT image acquisition:

Mandibular CBCT scans will be retrieved retrospectively from the database registry of oral and maxillofacial radiology department, faculty of dentistry, Cairo University. The included CBCT images shall be obtained using Planmeca ProMax® 3D Mid CBCT machine, at resolution of 0.4 voxel, 20*10 cm FOV (single mandibular arch), operating at 8 mA and 90 Kvp.

All CBCT images will be obtained in DICOM format (Digital Imaging Communication in Medicine) and viewed by Romexis software for analysis by a senior oral radiologist. Images with artifacts and poor visual quality, such as beam hardening artifacts, partial volume effect, aliasing artifacts, ring artifacts, low sharpness images resulting from patient movements during scanning, will be excluded.

Eligible CBCT scans will be divided into 4 groups according to age and sex of the patient (Males younger than 40, Males older than 40, Females younger than 40, Females older than 40). Groups will be assigned a letter A, B, C, or D in a random order, so that only the assistant supervisor will be aware of the demographic data of each group but will be concealed from the principal investigator who will perform the textural analysis.

Corrected sagittal cuts of the three regions of interest (anterior, premolar, molar) will be prepared from the eligible scans, then images will be saved in a BMP format.

Segmentation and texture analysis:

The BMP images of the corrected sagittal cuts of the cases will be imported to MaZda software (Technical University of Lodz, Institute of Electronics, Poland). A round region of interest will be manually cropped apical to the tooth, so that its size will be big enough to include the maximum trabecular bone present with no or little non-trabecular tissue.

GLCM textural features will be calculated by the software. All parameters will be calculated for two distances between pixels (d1 = 1, d2 = 2) and four sweeping directions (horizontal, diagonal, vertical and anti-diagonal) corresponding to (φ = 0°,45°,90°, 135°, respectively).The mean of the values at different positions will be taken as a single reading representing the texture feature value for the area of interest.

Handling of quantitative variables:

Eleven texture features will be calculated for 3 regions of interest on each eligible CBCT scan. Texture values will be handled as mean and standard deviation, and comparison between different age\gender groups will be done for selection of significant features. Correlation analysis will be done to test the correlation between age and gender and selected texture features.