Pierre Pottecher, Alexander Mühlberg, Oleg Museyko, Valerie Bousson, Klaus Engelke, Jean-Denis Laredo, Hôpital Lariboisière, Paris, France, Inst. of Medical Physics, University of Erlangen-Nuremberg, Germany, BioClinica, Inc., Princeton USA and Germany
BMD and propensity to falls are important risk factors for fractures of the proximal femur. Muscle morphology and function may be closely associated with the risk of fall. Here we compared the performance of CT based 3D-descriptors characterizing the spatial muscle-lipid distribution with bone parameters to discriminate hip fractures in a cross-sectional study.
MATERIALS AND METHODS
Two regions of the proximal femur were used for the analysis: an upper shaft (US) and a proximal femur posterior (PF) region (Fig. 1).
- Segmentation of the skin
- Segmentation of the intrafascia (IF) VOI
- Segmentation of subcutaneous adipose tissue (SAT) (Fig. 1)
- A gaussian-mixture-model (GMM) of the CT value distribution and SAT and water as calibration fixpoints are used to define adipose tissue (AT) and muscle tissue (MT) VOI (Fig. 2)
- The MT VOI is further divided into a ‘high density muscle’ (HDM) VOI containing ‘pure muscle’ and VOIs containing different fat muscle concentration ratios. These are not further evaluated here.
- EML is defined as IMAT not spatially connected with SAT
2. Structural analysis
- The 3D-descriptors listed in Tab. 1 were determined for the various VOIs
3. Fracture Discrimination
- 91 patients of the EFFECT-Study (Bousson et al. JBMR 2011): 36 patients with acute proximal femur fx, age 73.1 ± 9.3y; 55 controls, age 79.8 ±11.1y
- 3D-QCT: 120kV, 170mAs, slice thickness 1 or 1.25mm, pitch 1
- Analysis of the contralateral (fractured patients) or left leg (controls)
4. Statistical Analysis
- Univariate analysis: General linear model (GLM) adjusted for age and BMI.
- Multivariate analysis: Best subset analysis with bayesian information criterion to find the best muscle-lipid models
- Area-Under-Curve (AUC) and odd‘s ratio adjusted for age and BMI
- Structural descriptors are sensitive to noise, which depends on body size. Therefore all results were adjusted to BMI.
- Best subset models constructed from the set of significant (p<0.01) descriptors are shown in Tab.2.
- In the EFFECT cohort muscle-lipid models were numerically superior to the best bone model (Tab.3) and parameters in the US region were superior to those of the PF region. The combination of the muscle-lipid and bone model in the US region achieved the best discrimination.
- Discriminative power of models including volume or/and density of muscle, SAT or adipose tissue were comparable to the bone model.
In the EFFECT cohort, parameters describing the muscle-lipid distribution within the fascia numerically discriminated hip fracture better than the best ‘BMD’ model, a combination of BMD and cortical thickness,. In the fractured subjects muscles were more porous and adipocytes less aggregated, independent of their absolute volume.
The work was in part supported by the FORMOSA project of the Bavarian Research Foundation, Germany