Book

Description

Recently, a radiomics technique using CT texture fea- tures was adopted to access estimated BMD. Unlike QCT, CT cannot analytically compute the quan- tity due to the use of single energy, thus it needs special post-processes using CT images in Hounsfield units (HU) to predict BMD. The processes manually segment a bone area of interest, extract the features using gray-level co-occurrence matrices (GLCM), and optimally weigh the features using simple machine-learning tech- niques. In this study, we take advantage of the estimation method to indirectly identify that obesity influences the clinical BMD. One presumption is that the features extracted from the only bone areas in CT are rarely dependent on obesity like QCT. Then, if obesity causes the discrepancy between real BMD and clinical BMD, it will bring different statistics of estimation errors between the obesity group and the normal group. BMI and WC and the Index of Central Obesity (ICO) are adopted to investigate the impact of central obesity on the BMD estimation. In this study, a texture-based technique in CT was conducted to create a linear regression model to confirm the error and correlation between DXA measures and DXA-equivalent estimates.