How does it work?

3D-DXA is a technology developed by Tristan Whitmarsh which generates a three-dimensional reconstruction of bone structures from Dual-energy X-ray Absorptiometry (DXA) images1,2. From these tomographic reconstructions, three-dimensional structural and densitometric measurements can be taken. The main difficulty with this technique is the limited number of views provided by DXA. This has been overcome by the use of a-priori information in the form of statistical bone models3. These statistical models are constructed from a large number of Quantitative Computed Tomography (QCT) scans and describe the variations in shape and density distribution of the input population. The reconstruction process finds the parameter values of the statistical model such that the projection of the model matches the DXA image. This results in an instance of the statistical model with a morphometry that closely resembles the patients’ bone.


  • The accuracy of the method is highly dependent on how well the statistical model describes the target population. A statistical model built from a Spanish population won’t be able to accurately reconstruct the bone of a British person and vice versa. Hence, for each patient an appropriate model should be available with corresponding demographics, age range and gender.
  • 3D-DXA can not measure the cortical thickness or density. A statistical model describes the most probable variations in shape and BMD distribution. A bone that deviates from normality can therefore not be reconstructed. In particular the presence of osteophytes, which are irregular in nature, can greatly disrupt the reconstruction process. Following this, 3D-DXA can not be used to investigate atypical disease patterns seen in pathologies such as hyperparathyroidism and Paget’s disease unless the statistical model contains a substantial amount of patients also affected by these diseases. The same is true for drug trials which can modify the bone density and distribution in a way that is not represented in a normal population.
  • A statistical model describes the changes in BMD distribution only globally, i.e. there is no single parameter that only describes the cortical thickness. This means that changes in cortical thickness, cortical density and trabecular density are all linked to each other. An increase in thickness might simply be the result of an increase in overall density. Furthermore, as the cortical thickness can be seen only on the perimeter of the bone within the DXA image, any misalignment or error in shape reconstruction will greatly affect the cortical thickness and density. It has been shown that a mean estimator produces better estimates of the cortical thickness than 3D-DXA. This evaluation has not been included in recent publications, despite some of the authors being fully aware of it.
  • Related to the above limitations, no focal changes can be detected considering the global nature of the statistical model. Thus, localised patterns of either the trabecular BMD or cortical parameters in a longitudinal or comparison study have no meaning and should not be reported.
  • The vertebrae can not be reconstructed from a single DXA image. There is simply not enough contrast, too much noise and too many overlapping structures.
  • 3D-DXA can not measure hip structural parameters as I explain here.


  1. T. Whitmarsh, et al., Reconstructing the 3D shape and bone mineral density distribution of the proximal femur from dual-energy X-ray absorptiometry, IEEE Transactions on Medical Imaging, vol. 30(12), pp. 2101-14, 2011. (here)
  2. T. Whitmarsh, et al., 3D Reconstruction of the Lumbar Vertebrae from Anteroposterior and Lateral Dual-energy X-ray Absorptiometry, Medical Image Analysis, vol. 17(4), pp. 475–487, 2013. (here)
  3. T. Whitmarsh, et al., A statistical model of shape and bone mineral density distribution of the proximal femur for fracture risk assessment, Medical Image Computing and Computer-Assisted Intervention, vol. 6892, pp. 393-400, 2011. (here)
  4. T. Whitmarsh (2012), 3D reconstruction of the proximal femur and lumbar vertebrae from dual-energy X-ray absorptiometry for osteoporotic fracture risk assessment. Ph.D thesis, Pompeu Fabra University, Spain. (here)


Results derived from 3D-DXA should not be used for diagnosis or treatment planning nor should they be used for research into disease progression or drug response. In doing so you may be held liable for medical malpractice or scientific misconduct.

Back to Top
3D-DXA by the inventor