Subject enrollment and allocation
The Institutional Review Board "Public Institutional Review Board Designated by Ministry of Health and Welfare" (P01-202,109–21-014) approved this study. A total of 1423 cases (1208 patients) who had undergone both CT and DXA in a single institution between May 6, 2012, to June 30, 2021, were initially corrected. Among them, we selected 780 cases (681 patients) with (1) abdominal CT or abdomen-pelvis CT with a complete first lumbar vertebra (L1) axial cut, (2) abdominal CT or abdomen-pelvis CT completely showing umbilical cord axial cut, and (3) less than a one-month gap between CT and DXA scan dates. Next, we excluded 70 cases (60 patients) with (1) a history of previous L1 vertebral body fracture, (2a) history of cement or metal artifacts of a previous fracture and refracture, and (3) difficulty in identifying trabecular bones. Thus, 710 cases (612 patients) were selected as shown in Fig. 1.
Additionally, we considered clinical BMD conditions that the samples whose BMD is less than 0.1 g/cm2 and greater than 1.5 g/cm2 were excluded (13 patients) as outliers. The samples were finally divided by normal and overweight/obese under the BMI, WC, or ICO standard as shown in Fig. 1.
Anthropometric measurements
A seca digital scale was used to measure the patient’s height and weight. Body mass index (BMI) was calculated as weight (kg) divided by the square height expressed in meters. BMI values in the range 19—25 kg/m2 were considered normal weight, while subjects with BMI ≥ 25 kg/m2 and ≥ 30 kg/m2 were the cut-off levels for overweight and obese subjects, respectively. Waist Circumference was measured at the abdominal-pelvis CT’s axial cut where the umbilical cord was visible along the axis of the lumbar vertebrae and cut-off > 88 cm was used as an indicator of central obesity. Index of central obesity (ICO) is a term defining the ratio of WC and height. The International Diabetes Federation (IDF) suggested that it better predict central obesity where ICO > 0.5- central obesity and ICO < 0.5—no central obesity.
Imaging protocols for CT and DXA
A Siemens (SOMATOM 128, Definition AS +) scanner (Siemens Healthcare, Forchheim, Germany) was used for CT scans, for every scan, the protocol was a single-energy CT with 120 kVp, 247 mA, dose modulation 0.6-mm collimation. The effective pitch was 0.8 and the reconstruction kernel was B60 (sharp). Reconstructed slice thicknesses were set at 5.0 mm and 3.0 mm for chest CT and lumbar spine CT, respectively. And the slice increments for abdomen and pelvis CT were 5.0 mm and 3.0 mm, respectively. For DXA scans, a standard device with a standard protocol (GE Lunar Prodigy, GE Healthcare) was used and reports were obtained using vendor-specific software (Physicians Report Writer DX, Hologic, Discovery Wi, USA).
Region of interest
Regions of interest (ROIs) for statistical measurement from the CT axial cuts of every patient, we selected one slice image such that it contains the maximum axial trabecular area of the bone. Of the many methods available for isolating ROIs, the thresholding method was chosen for this study. A 2-dimensional (2D) slice image was chosen from the CT axial cut of every patient wherein the 2D image contained the maximum axial trabecular area of the L1 spine body or the femoral neck. As shown in Fig. 2, we conducted texture analysis in a circular region covering most of the trabecular area.
Clinical BMD estimation using CT
Figure 2 illustrates the schematic flow of the BMD estimation. A total of 45 features were extracted from the ROIs, of which five features were intensity-based and extracted using a histogram, and 40 texture-based features were extracted using a GLCM matrix. One estimation clinical BMD was then computed from the features using conventional linear regression (LR).
Statistical analysis
Statistical analyses were performed using MATLAB 9.10 R2021a (MathWorks, Natick, Massachusetts, USA). The paired t-test was used to verify differences in clinical results and patient demographics between the two groups. Power analysis revealed an effect size of 0.5, the statistical significance of 0.05, and the statistical power of 0.90 for both groups. Mean absolute error was used to compare p-values between the two groups.