Investigating and improving the long-term precision of quantitative CT bone densitometry measurements by means of histogram analysis

This paper investigates the addition of histogram analysis to quantitative CT (QCT) bone density measurements, employing a simultaneous calibration phantom, to assess and improve its precision. Standardized scores (z- scores) that rank the precision of measurements of the annual change in trabecular bone density were used to sample the seven best and seven worst cases from a set of 40 patients. These were measured serially five times for 2 years. Regions of interest extracted from the image data included the entire vertebrate; segments of liver, muscle, and fat tissue; and a sample of Plexiglass within the calibration phantom. Distributions (histograms) of CT numbers were generated. Linear regression for each were analyzed as a function of time. Correlations to scanner parameters were investigated as possible sources of precision error, and changes in bone density were compared with the use of weighted and unweighted linear least squares fits