Original Full Length ArticleHip fracture discrimination from dual-energy X-ray absorptiometry by statistical model registration
Highlights
► A method for improving the hip fracture risk estimation from DXA is proposed. ► The proximal femur is reconstructed from a DXA image using the registration of a statistical model. ► The statistical model describes the patient specific 3D shape and BMD distribution. ► The model parameters are analyzed for their hip fracture discrimination ability with respect to regular areal BMD measurements. ► The method allows for a 3D analysis whilst maintaining DXA as the standard low radiation dose modality.
Introduction
Fracture risk assessment currently relies on areal BMD measurements from dual-energy X-ray absorptiometry (DXA). Although the areal BMD has been shown to be strongly correlated with fracture incidence, this measure is limited by its two-dimensionality whilst the spatial distribution and geometry of the bone to a large extent determines the bone strength [1]. A volumetric image of the bone densities can be acquired using Quantitative Computed Tomography (QCT) and volumetric BMD and 3D structural measurements from QCT have been shown to be important parameters for determining the femoral strength [2], [3], [4] and deriving a fracture risk [5], [6], [7]. The acquisition of QCT scans, however, administers the patient with a relatively high dose of radiation compared to DXA. In addition, the high cost and limited access to CT scanners prevents this modality from being used in clinical routine and thus DXA remains the current clinical standard for bone density measurements and fracture risk assessment. Being able to extract information on the hip geometry and density distribution from the widely used DXA technology would therefore constitute a great advantage to hip fracture risk assessment.
Some work has already been done in reconstructing the 3D shape from DXA images [8]. This allows for the extraction of 3D geometric parameters which, combined with areal BMD measurements, were shown to improve the fracture load prediction over areal BMD alone [9].
Although these methods incorporate the 3D shape, also the spatial distribution of the bone determines its strength and thus the risk of fracture. Several methods have been proposed to acquire a 3D reconstruction of both the 3D shape and the density distribution from a regular DXA image by the deformation of a template [10] or by incorporating a statistical model in an intensity-based registration process [11], [12]. From these reconstructions the same parameters can be extracted as from QCT. However, these parameters are still limited descriptors of the shape and spatial distribution of the bone.
Several authors propose the use of Finite Element Analysis (FEA) of bones from CT scans [13], [14], [15] or from the reconstruction from DXA [10] to determine their mechanical behaviour. Although FEA has been shown to accurately estimate the resistance to specific loading conditions, their direct relation to the risk of fracture has not yet been fully established.
In [16] a method was presented for a more detailed analysis of the spatial distribution of the bone from CT using deformable registrations and image similarities. This method was shown to be able to accurately discriminate between fracture patients and controls and was used to identify the regions within the femur which are most strongly associated with hip fracture [17].
Recently, statistical models as pioneered by Cootes et al. [18] have received a great deal of interest as a means to analyze the complex morphometry of organs for the diagnosis of diseases and detection of symptoms. Both in [19] and [20] the femoral shape was analyzed from planar radiographs using statistical shape modelling, and the association of the modes of variation with the fracture incidence was examined. In [21] this method was extended by including an analysis of the trabecular bone structure, which significantly improved the hip fracture discrimination ability. These methods, however, are still limited by a two-dimensional analysis from radiographic projections.
In a recent work, Li et al. [22] constructed a statistical model of the volumetric density distribution of the proximal femur and analyzed the model parameters for their hip fracture discrimination power. Here, only the density distribution was analyzed whilst also the shape determines the femoral strength and shape parameters have been shown to be independent hip fracture discriminators [2].
In [23] a statistical model of shape and appearance was presented and the parameters of this model were, in other work, analyzed for their ability to predict the fracture load of the proximal femur [24]. Although the fracture load gives a measure of the bone strength, it does not directly relate to the risk of fracture.
Previously, we developed a statistical model of both the 3D shape and BMD distribution of the proximal femur for fracture risk assessment [25]. This model is constructed from a large dataset of QCT volumes using an intensity based registration process. The parameters of this model describe the global shape and spatial distribution of the bone and were shown to correctly represent the shape and density variations which determine the fracture risk. In other work this model was shown to be able to accurately reconstruct the 3D shape and BMD distribution of the proximal femur by registering it onto a single DXA image, whereby the proximal femur was reconstructed with a mean shape accuracy of 1.1 mm and a global BMD distribution error of 4.9% [11].
In this study we aim to evaluate the parameters of the statistical model resulting from the registration onto DXA for their ability to improve the hip fracture discrimination from DXA with respect to regular areal BMD measurements.
Section snippets
CT dataset for statistical model construction
For the construction of the statistical model a dataset of 80 CT scans of the pelvic region was collected at the CETIR Medical Center (Barcelona, Spain) using the Philips Gemini GXL 16 system (Philips Healthcare, Best, The Netherlands). This dataset was supplemented by a set of 80 CT scans collected at the Department for Trauma Surgery of the Medical University Innsbruck (Innsbruck, Austria), using the GE LightSpeed VCT Multi Slice CT device (GE Healthcare, Madison, WI, USA). All CT scans were
Results
First, the relationships of the individual model parameters with the hip fracture incidence are examined. In Table 2, Table 3 the ORs and AUC values are given for the shape and density model parameters individually. The 1st mode of variation of the shape model corresponding to the scale (Fig. 1) has a strong association with the fracture incidence with an OR of 0.655 (95% CI 0.490–0.875, p = 0.004). Also for the 2nd shape model parameter, which describes the neck shaft angle, there is a
Discussion
In this work a method to construct and register a statistical model of shape and density distribution onto a DXA image was presented and evaluated for its ability to discriminate between a fracture and control group. The model parameters resulting from the reconstructions have been shown to improve the discrimination ability with respect to only DXA derived areal BMD measurements.
Fig. 5 illustrates the differences in shape and density distribution that determine the fracture risk according to a
Acknowledgments
The authors gratefully acknowledge J. Malouf Sierra (Departamento de Medicina Interna, Unidad de Metabolismo Mineral, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain) and S. Di Gregorio (CETIR Centre Mèdic, Barcelona, Spain) for technical support and data acquisition.
The research leading to these results has received funding from: the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement nr 269909, MySPINE project; the ERDF Operational Programme of Catalonia
References (32)
- et al.
Discrimination between cases of hip fracture and controls is improved by hip structural analysis compared to areal bone mineral density. An ex vivo study of the femoral neck
Bone
(2004) - et al.
Assessment of the strength of proximal femur in vitro: relationship to femoral bone mineral density and femoral geometry
Bone
(1997) - et al.
Volumetric quantitative computed tomography of the proximal femur: precision and relation to bone strength
Bone
(1997) - et al.
Analysis of hip geometry by clinical CT for the assessment of hip fracture risk in elderly Japanese women
Bone
(2010) - et al.
Distribution of cortical bone in the femoral neck and hip fracture: a prospective case–control analysis of 143 incident hip fractures; the AGES-REYKJAVIK Study
Bone
(2011) - et al.
Assessment of femoral neck strength by 3-dimensional X-ray absorptiometry
J Clin Densitom
(2006) - et al.
Validation of an automated method of three-dimensional finite element modelling of bone
J Biomed Eng
(1993) - et al.
Bone fracture risk estimation based on image similarity
Bone
(2009) - et al.
Identify fracture-critical regions inside the proximal femur using statistical parametric mapping
Bone
(2009) - et al.
Proximal femoral density and geometry measurements by quantitative computed tomography: association with hip fracture
Bone
(2007)
Volumetric quantitative computed tomography of the proximal femur: relationships linking geometric and densitometric variables to bone strength. Role for compact bone
Osteoporosis Int
In vivo discrimination of hip fracture with quantitative computed tomography: results from the prospective European femur fracture study (EFFECT)
J Bone Miner Res
Three-dimensional X-ray absorptiometry (3D-XA): a method for reconstruction of human bones using a dual X-ray absorptiometry device
Osteoporos Int
Estimation of 3D shape, internal density and mechanics of proximal femur by combining bone mineral density images with shape and density templates
Biomech Model Mechanobiol
Reconstructing the 3D shape and bone mineral density distribution of the proximal femur from dual-energy X-ray absorptiometry
IEEE Trans Med Imaging
Volumetric DXA (VXA): a new method to extract 3D information from multiple in vivo DXA images
J Bone Miner Res
Cited by (28)
Finite element analysis informed variable selection for femoral fracture risk prediction
2021, Journal of the Mechanical Behavior of Biomedical MaterialsCombining shape and intensity dxa-based statistical approaches for osteoporotic HIP fracture risk assessment
2020, Computers in Biology and MedicineCitation Excerpt :Statistical Shape Models (SSMs) have been promisingly adopted to examine associations between the global morphometry variations in the femoral shape of an input cohort and the fracture incidence [17,18]. Similarly, the bone main BMD distribution variations have also been explored through Statistical Intensity Models (SIMs) for fracture discrimination [19–22]. In the afore-mentioned cases, the construction of the statistical models relied on Principal Component Analysis (PCA), aiming to retrospectively identify those geometric and densitometric attributes which are significantly different in fracture and non-fracture cohorts [17,18].
Enhancing hip fracture risk prediction by statistical modeling and texture analysis on DXA images
2020, Medical Engineering and PhysicsSignificant morphological change in osteoarthritic hips identified over 6–12 months using statistical shape modelling
2018, Osteoarthritis and CartilageProceedings of the 2016 Santa Fe Bone Symposium: New Concepts in the Management of Osteoporosis and Metabolic Bone Diseases
2017, Journal of Clinical DensitometryStatistical Shape and Appearance Models for Bone Quality Assessment
2017, Statistical Shape and Deformation Analysis: Methods, Implementation and Applications