Calibration methods for the screening of osteoporosis on lumbosacral spine CT scans

Osteoporosis, a common skeletal disorder causing reduced bone strength and fracture risk, is typically diagnosed via areal Bone Mineral Density (aBMD) assessment. This thesis retrospectively evaluates asynchronous and phantomless calibration methods for opportunistic volumetric BMD (vBMD) estimation on standard CT scans of the lumbosacral spine. Calibration methods were tested on 26 at-risk patients (mean age 61.54 years, 46.2% males, 53.8% females) who underwent non-contrast spine CT (120 kVp, 1.25 mm slice thickness, 0.625 mm interval, standard kernel). vBMD (mg/cc) was measured in L4 using ITK-SNAP. Methods included asynchronous calibration with a potassium phosphate dibasic phantom and internal calibration with three or five reference ROIs (air, adipose, muscle, cortical bone, blood). ESP phantom calibration was set as the gold standard. Measurements were performed in CHA and K₂HPO₄ units using the open-source Ogo library (Michalski, Boyd et al.). Linear regression showed significant correlations (p < 0.05) between all methods and ESP. The K₂HPO₄ phantom showed the highest concordance (slope 0.82, intercept 9.90, R² = 0.998). Internal calibration showed lower concordance, with slopes from 0.58 to 0.73 and increasing intercepts (15.38 using three ROIs, R² = 0.972; 47.72 using five ROIs, R² = 0.907). Internal calibration with five ROIs overestimated vBMD, particularly in low-density vertebrae. In conclusion, asynchronous calibration using different phantoms is highly concordant, whereas phantomless methods, though promising for opportunistic osteoporosis screening, require further refinement for clinical reliability.