Assessing Bone Mineral Density in Weight-Bearing Regions of the Body through Texture Analysis of Abdomen and Pelvis CT Hounsfield Unit.

Abstract

Objective: This study aimed to develop a novel method for opportunistically screening osteoporosis by measuring bone mineral density (BMD) from CT images. We addressed the limitations of commercially available software and introduced texture analysis using Hounsfield units (HU) as an alternative approach. Methods: A total of 458 samples (296 patients) were selected from a dataset of 1320 cases (782 patients) between 1 March 2013, and 30 August 2022. BMD measurements were obtained from the ilium, femoral neck, intertrochanteric region of both femurs, and L1-L5 and sacrum spine body. The region of interest (ROI) for each patient's CT scan was defined as the maximum trabecular area of the spine body, ilium, femoral neck, and femur intertrochanter. Using gray-level co-occurrence matrices, we extracted 45 texture features from each ROI. Linear regression analysis was employed to predict BMD, and the top five influential texture features were identified. Results: The linear regression (LR) model yielded correlation coefficients (R-squared values) for total lumbar BMD, total lumbar BMC, total femur BMD, total femur BMC, femur neck BMD, femur neck BMC, femur intertrochanter BMD, and femur intertrochanter BMC as follows: 0.643, 0.667, 0.63, 0.635, 0.631, 0.636, 0.68, and 0.68, respectively. Among the 45 texture features considered, the top five influential factors for BMD prediction were Entropy, autocorrelate_32, autocorrelate_32_volume, autocorrelate_64, and autocorrelate_64_volume.