Estimation of Cortical Density with Clinical CT
Osteoporosis - Assessment
Bone Biomechanics and Quality (Basic)
Poster Sessions, Presentation Number: MO0333
Session: Poster Session III & Poster Tours*
Monday, September 19, 2011 11:30 AM - 1:30 PM, San Diego Convention Center, Hall GH
* , UNITED KINGDOM, , UNITED KINGDOM, Andrew Gee, University of Cambridge
There is growing evidence that focal thinning of cortical bone in the proximal femur is critical in determining resistance to fracture. Accurately assessing cortices thinner than 2.5 mm in clinical MDCT is challenging, since they are corrupted by imaging blur. We recently proposed a model-fitting technique for accurate estimation down to 0.3 mm (Treece et al. Medical Image Analysis, 2010). The technique assumes a patient-specific, fixed value for true cortical density, estimated from CT data of thick cortices below the lesser trochanter. Using this technique, we have shown, for 65 osteoporotic women, precisely where in the proximal femur new bone is laid down following treatment with hPTH(1-34) (Poole et al. Plos One, 2011). We now aim to provide an improved density estimate which is more representative of the entire proximal femur, is available from incomplete CT data or patients with thin cortices, and has associated confidence intervals.
If the true cortical density is substantially constant, and the imaging blur is Gaussian, we can derive the expected distribution of peak cortical density with thickness (solid line in figure 1), which depends only on the imaging blur, true cortical density, and background density. We sample the peak cortical density, measured in clinical CT data, at many points over the proximal femur, and compare this with the expected distribution. Modelled blur, cortical and background density are adjusted to minimise the difference between measured and expected distributions, leading to mean and 95% confidence limits for the true cortical density. Figure 1 shows typical outcomes from three different subjects in the hPTH study. The measured and expected distribution are in close agreement in figure 1(a). Figure 1(b) is a more difficult case where the model accurately predicts a true cortical density which is not apparent in the measured data. In figure 1(c), the cortex is very thin and the density estimate is hence less reliable: this is reflected in the increased confidence limits of ± 25 HU. Nevertheless, using the new technique on the hPTH data, the worst case confidence limits were only ± 38 HU, and all cases were deemed to give acceptable estimates of cortical density. The resulting baseline population cortical bone mineral density was 1185.1 ± 67.9 mg/cm3 decreasing to 1133.6 ± 78.1 mg/cm3 (p << 0.001) after 24 months' treatment.
* Presenting Authors(s):
, UNITED KINGDOM