Abstract
BACKGROUND: Bone mineral content (BMC) values in certain bones and changes in BMC over time are key features for diagnosing osteoporosis. This study examined those features using morphometric texture analysis in chest computational tomography (CT) by comparing a dual-energy X-ray absorptiometry (DXA)-based BMC. An accessible approach for screening osteoporosis was suggested by accessing BMC using only Hounsfield units (HU). METHODOLOGY: The study included a total of 510 cases (255 patients) acquired between May 6, 2012, and June 30, 2020, at a single institution. Two cases were associated with two chest CT scans from one patient with a scan interval of over two years, and each scan was followed soon after by a DXA scan. Axial cuts of the first lumbar vertebra in CT and DXA-based L1 BMC values were corrected for each case. The maximum trabecular area was selected from the L1 spine body, and 45 texture features were extracted from the region using gray-level co-occurrence matrices. A regression model was employed to estimate the absolute BMC value in each case using 45 features. Also, an additional regression model was used to estimate the change in BMC between two scans for each patient using 90 features from the corresponding cases. RESULTS: The correlation coefficient (CC) and mean absolute error (MAE) between estimates and DXA references were obtained for the evaluation of regressors. In the case of the BMC estimation, CC and MAE were 0.754 and 1.641 (g). In the case of the estimation of change in BMC, CC and MAE were 0.680 and 0.528 (g). CONCLUSION: The modality using morphometric texture analysis with CT HUs can indirectly help screening osteoporosis because it provides estimates of BMC and BMC change that show moderate positive correlations with DXA measures.