Automation of a DXA-based finite element tool for clinical assessment of hip fracture risk

https://doi.org/10.1016/j.cmpb.2017.11.020Get rights and content

Highlights

  • DXA-based finite element model for assessing hip fracture risk in sideways fall.

  • Completely automated computer program with hip DXA as the only input.

  • Greatly improved short-term precision after automation.

  • Better performance than femoral BMD in discrimination test.

ABSTRACT

Finite element analysis of medical images is a promising tool for assessing hip fracture risk. Although a number of finite element models have been developed for this purpose, none of them have been routinely used in clinic. The main reason is that the computer programs that implement the finite element models have not been completely automated, and heavy training is required before clinicians can effectively use them. By using information embedded in clinical dual energy X-ray absorptiometry (DXA), we completely automated a DXA-based finite element (FE) model that we previously developed for predicting hip fracture risk. The automated FE tool can be run as a standalone computer program with the subject's raw hip DXA image as input. The automated FE tool had greatly improved short-term precision compared with the semi-automated version. To validate the automated FE tool, a clinical cohort consisting of 100 prior hip fracture cases and 300 matched controls was obtained from a local community clinical center. Both the automated FE tool and femoral bone mineral density (BMD) were applied to discriminate the fracture cases from the controls. Femoral BMD is the gold standard reference recommended by the World Health Organization for screening osteoporosis and for assessing hip fracture risk. The accuracy was measured by the area under ROC curve (AUC) and odds ratio (OR). Compared with femoral BMD (AUC = 0.71, OR = 2.07), the automated FE tool had a considerably improved accuracy (AUC = 0.78, OR = 2.61 at the trochanter). This work made a large step toward applying our DXA-based FE model as a routine clinical tool for the assessment of hip fracture risk. Furthermore, the automated computer program can be embedded into a web-site as an internet application.

Introduction

Fall-induced hip fracture is a common health risk among elderly people, especially for those who have osteoporosis [1], [2], [3], [4]. Osteoporosis can significantly reduce bone strength, it is a ‘silent’ disease as it has no symptom until first fracture. Therefore, subjects of high fracture risk must be diagnosed by clinicians using a reliable assessment tool. A number of tools for assessing fracture risk have been developed from either population-based statistical models [5], [6], [7], [8], [9] or biomechanical models [10], [11], [12], [13], [14], [15], [16]. The existing clinical tools are almost exclusively based on statistical models. T-score calculated from femoral bone mineral density (BMD) is the gold standard reference for screening osteoporosis and for assessing fracture risk [17], [18]. However, extensive clinical studies showed that the majority of patients who sustain low-trauma fractures have T-scores within the WHO (World Health Organization) safe range [19], [20], [21], [22]. Therefore, researchers were motivated to develop assessment tools to consider multiple clinical risk factors [23]. Among these tools FRAX (Fracture Risk Assessment Tool) is the most popular one, which is a web-based calculator to predict an individual's 10-year probability of hip fracture. However, FRAX still has limited accuracy even though 12 clinical risk factors are considered [5], [23], [24], [25]. The main reasons are: the considered clinical factors are biomechanically dependent and some important biomechanical variables such as fall-induced impact force are missing [26].

In view of the limitations of population-based tools, researchers have turned their attention to biomechanical approaches. Theoretically, biomechanical modeling has the potential to more accurately predict an individual's fracture risk than statistical approaches, as biomechanical models are based on well-established mechanical principles. A large number of finite element (FE) models have been developed to determine bone strength and to assess fracture risk. Most of the FE models are constructed from QCT (quantitative computed tomography) [11], [16], [27], [28], [29], [30], [31], [32], [33], [34], since QCT contains actual material and geometry information required to construct a three-dimensional finite element model. However, QCT uses high dosage of radiation and it is not recommended for routine clinical examination [35]. To meet the current clinical needs, DXA-based finite element models [36], [37] have also been developed, although construction of accurate finite element models from DXA images is more challenging and assumptions must be introduced.

Nevertheless, none of the finite element models have been used routinely in clinic. Indeed, there are still unresolved technical issues in the finite element models [27], for example, the discrepancy among different bone material models [38], the lack of a generally accepted risk threshold for clinical intervention and the inconsistency among risk measurements produced by different finite element models. However, even after the technical issues are all resolved, there is still an obstacle for clinical application, that is, the computer programs that implement the finite element models have not been completely automated. User intervention that often requires the knowledge of finite element analysis (FEA) does not appeal to clinicians. We developed a DXA-based FE model for the prediction of hip fracture risk [37], [39]. In this study, we completely automate the computer program with simple input, so that the automated DXA-based FE tool can be used by clinicians with a minimum training and no knowledge of FEA is required.

Section snippets

Automation of DXA-based finite element (Fe) model

The procedure of our DXA-based finite element analysis [37], [39] is shown in Fig. 1. The only input required by the procedure is a raw hip DXA, for example, the enCore file (with extension of DFF) generated by GE Lunar. The procedure starts with a clinical hip DXA of the concerned subject and includes the following steps: (1) Clinical regions of interest (ROI) are identified; (2) Proximal femur is segmented from the DXA and a contour of the femur is obtained; (3) A finite element mesh is

Clinical study

Short-term precision and the ability of discriminating clinical fracture cases from matched controls are the two important performance indicators of clinical tools for assessing hip fracture risk. To study the performance of the automated DXA-based FE tool, clinical cases were acquired from St. Boniface Hospital in Winnipeg under an approval of human subject research ethics, with personal information such as patient name and residence address removed before acquisition. Each case had a unique

Results

The average processing time for each case was about 15 seconds when the cases were processed in batch. If processed cases by case, the processing time was about 17 seconds. Both are acceptable for clinical practice.

Short-term precision of FRI and BMD expressed by coefficient of variation (CV, %) are presented in Table 3. In general, BMD still has the best short-term precision; automated DXA-based FE tool greatly improved the precision compared with the previous semi-automated version.

In the

Discussion

The large improvement in short-term precision of our DXA-based FE tool is mainly attributed to the automation of femur contour segmentation. In the previous semi-automated version [37], manual operation is required in segmentation of femur contour, which introduces random error. Automated segmentation in the new version is able to eliminate the random error and thus improves the short-term precision. However, even the automated FE tool still has much lower short-term precision than BMD. There

Conclusion

In conclusion, this study made a step toward the application of our DXA-based FE model as a clinical tool for assessment of hip fracture risk. As the next step, we will apply our model in prospective studies and in validations of cross clinical centers and DXA scanners. The automated standalone computer program will be distributed to clinical centers. Furthermore, the computer program will be embedded into a web-site as an internet application, so that clinicians over the world will be able to

Conflict of Interest

All the three author declare there is no conflict of interest.

Acknowledgments

The reported research has been supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada (37098) and Research Manitoba (37807), which are gratefully acknowledged.

References (67)

  • E. Dall'Ara et al.

    Experimental validation of DXA-based finite element models for prediction of femoral strength

    J. Mech. Behav. Biomed. Mater.

    (2016)
  • B. Helgason et al.

    Mathematical relationships between bone density and mechanical properties: a literature review

    Clin.Biomech.

    (2008)
  • Y. Luo

    A nearest-nodes finite element method with local multivariate lagrange interpolation

    Finite Elements Anal. Design

    (2008)
  • Y. Luo

    Adaptive nearest-nodes finite element method guided by gradient of strain energy density

    Finite Elements Anal. Design

    (2009)
  • E.F. Morgan et al.

    Dependence of yield strain of human trabecular bone on anatomic site

    J. Biomech.

    (2001)
  • E.F. Morgan et al.

    Trabecular bone modulus–density relationships depend on anatomic site

    J. Biomech.

    (2003)
  • M.M. Juszczyk et al.

    The human proximal femur behaves linearly elastic up to failure under physiological loading conditions

    J. Biomech.

    (2011)
  • H. Burger et al.

    The association between age and bone mineral density in men and women aged 55 years and over: the Rotterdam study.

    Bone Miner

    (1994)
  • J. Carballido-Gamio et al.

    Structural patterns of the proximal femur in relation to age and hip fracture risk in women

    Bone

    (2013)
  • B. Abrahamsen et al.

    Excess mortality following hip fracture: a systematic epidemiological review

    Osteoporos. Int.

    (2009)
  • C.A. Brauer et al.

    Incidence and Mortality of Hip Fractures in the United States.

    J. Am. Med. Assoc.

    (2009)
  • G. Carrière

    Hip fracture outcomes in the household population

    (2007)
  • J.A. Kanis et al.

    Interpretation and use of FRAX in clinical practice

    Osteoporosis Int.

    (2011)
  • F.A. Trémollieres et al.

    Fracture risk prediction using BMD and clinical risk factors in early postmenopausal women: sensitivity of the WHO FRAX tool

    J. Bone Miner. Res.

    (2010)
  • G.S. Collins et al.

    Predicting risk of osteoporotic and hip fracture in the United Kingdom: prospective independent and external validation of QFractureScores.

    BMJ

    (2011)
  • J. Hippisley-Cox et al.

    Predicting risk of osteoporotic fracture in men and women in England and Wales: prospective derivation and validation of QFractureScores.

    BMJ

    (2009)
  • National Clinical Guideline Centre (UK). Osteoporosis: assessing the risk of fragility fracture (NICE clinical...
  • J.O. Den Buijs et al.

    Validated finite element models of the proximal femur using two-dimensional projected geometry and bone density

    Comput. Methods Program Biomed.

    (2010)
  • R.D. Carpenter

    Finite element analysis of the hip and spine based on quantitative computed tomography

    Current Osteoporos Rep.

    (2013)
  • D. Christen et al.

    Multiscale modelling and nonlinear finite element analysis as clinical tools for the assessment of fracture risk

    Philos. Trans. R. Soc. A

    (2010)
  • Z. Ferdous et al.

    Study of Hip Fracture Risk by DXA-based Patient-Specific Finite Element Model.

    Bio-Med. Mater. Eng.

    (2015)
  • P. Pulkkinen et al.

    BMD T-score discriminates trochanteric fractures from unfractured controls, whereas geometry discriminates cervical fracture cases from unfractured controls of similar BMD.

    Osteoporos Int.

    (2010)
  • E. Sornay-Rendu et al.

    Identification of osteopenic women at high risk of fracture: the OFELY study

    J. Bone. Miner. Res.

    (2005)
  • Cited by (19)

    • Stochastic failure analysis of proximal femur using an isogeometric analysis based nonlocal gradient-enhanced damage model

      2022, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      The finite element studies have evaluated femur bone's strength under different loading conditions for 2D [8–10] and 3D [11,12] models. These studies focused on identifying the regions of high-stress concentration and determining the strength of the bone using either stress-based [9,10] or strain-based criteria [11,12]. However, these studies lacked details of the actual fracture process, i.e., the crack path, structural response, etc.

    • Risk assessment of resurfacing implant loosening and femur fracture under low-energy impacts taking into account degenerative changes in bone tissues. Computer simulation

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      According to these data, the values of the elastic modulus and strength may be derived. This approach is widely used in finite element modeling to select model inputs and predict fracture risks [6–13]. To study the mechanical behavior of the system “bone-endoprosthesis”, its geometric model was created.

    • Computer-aided osteoporosis detection from DXA imaging

      2019, Computer Methods and Programs in Biomedicine
    • Numerical modelling of hip fracture patterns in human femur

      2019, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      The synergy of these techniques allows to obtain more reliable results enhancing the knowledge about femur fracture biomechanics [4]. These models can be a useful tool for clinical assessment of hip fracture risk [5]. In the literature, the mechanical behaviour of femur has been addressed under the point of view of elastic response, fracture load and fracture morphology.

    View all citing articles on Scopus
    View full text