In vivo and in vitro precision for bone density measured by dual-energy X-ray absorption

Abstract

An investigation was made into some of the major sources of error influencing the bone mineral density (BMD) measurements of the lumbar vertebrae, the femoral neck and the greater trochanter. The effect on accuracy and reproducibility of the following parameters was investigated: influence of patient positioning, patient size, scan speed, the technique of scan analysis and the temporal variation in instrument performance.

The in vitro precision, both long-term and short-term, was assessed using aluminium phantoms supplied by the manufacturer. For the spine phantom, the precision expressed as a percentage coefficient of variation (%CV) was found to be 0.4% (10 scans) in the short term and 0.55% (15 scans) in the long term. Measured precision (short-term) for the three regions of the femur phantom analysed by the software was 1.3% for the neck of femur, 1.7% for Ward's triangle and 0.6% for the trochanter. Long-term precision was 1.0%, 1.9% and 1.1% respectively. No statistically significant difference was found between long- and short-term results.

Short-term in vitro precision on a low density anthropological phantom was 4.1%, 4.2%, 2.4% and 0.61% for neck of femur, Ward's triangle, trochanter and spine respectively.

In vivo short-term precision for the lumbar spine (L2–L4), measured by scanning four normal volunteers five times in one session, was found to be 0.8±0.25%. In vivo precision for the femur, measured on seven volunteers was 1.6±0.8% for the neck of femur, 3.2±1.7% for Ward's triangle and 2.2±1.1% for the trochanter.

The variation in density results caused solely by analysis was investigated by analysing ten randomly selected normal femoral scans five times on five different days. Again results are expressed as a coefficient of variation and were 1.3±0.8% for the neck of femur, 2.4±1% for Ward's triangle and 0.8±0.7% for the trochanter, suggesting that most of the variation in the femur density results can be attributed to the analysis procedure.