3D Imaging Bone Quality: Bench to Bedside

Abstract Introduction Measuring the health of bone is important for understanding the pathogenesis, progression, diagnosis and treatment outcomes for fragility. At present the most common method for measuring bone health in a clin-­‐ ical setting is to assess skeletal mass. The current gold standard is dual-­‐ energy X-­‐ray absorptiometry (DXA) which models bones as 2D objects and measures areal bone mineral density (BMD). However, BMD only accounts for 50% of bone strength and the technique ignores other important factors such as cortical geometry and trabecular architec-­‐


Introduction
Measuring the health of bone is important for understanding the pathogenesis, progression, diagnosis and treatment outcomes for fragility.At present the most common method for measuring bone health in a clin-ical setting is to assess skeletal mass.The current gold standard is dual-energy X--ray absorptiometry (DXA) which models bones as 2D objects and measures areal bone mineral density (BMD).However, BMD only accounts for 50% of bone strength and the technique ignores other important factors such as cortical geometry and trabecular architec-contributors.Consequently a new concept of 'bone quality' has devel-oped the material and structural basis of bone strength and fragility.As yet though, a suitable non--invasive method has not been developed for measuring quality in living patients.The aim of this paper is to discuss how bone quality might be visualised, setting.

Discussion
The most useful imaging techniques are likely to be clinical--CT and MRI.Both modalities have been used successfully to characterise bone macro--structure in 3D e.g.volume fraction and orientation.More recently in vivo systems with high 3D Imaging Bone Quality: Bench to Bedside RL Abel1 *, M Prime2 , A Jin 1 , JP Cobb 1 , R Bhattacharya 2 * resolution (~0.100-0.200mm) have been developed that can capture some aspects of bone micro--architec-ture. Alternatively 3D models created using clinical--CT and MRI can be used to virtually simulate loading on a computer and calculate bone mechan-ical properties.Analysed together these morphological and mechanical data sets might allow clinicians to provide screening programmes for osteoporosis and calculate individual fracture risk.Especially if applied as part of a holistic approach utilising patient meta--data on risk factors for metabolic bone disease (e.g.FRAX).As well as improve primary and secondary care by setting treat to target criteria for pharmacological therapies and planning surgical inter-ventions or following up treatment outcomes.

Conclusion
In the short to mid term the expense of 3D imaging and (in the case of CT) the risks associated with ionising radiation are going to restrict image resolution.Therefore, in order

Bone Quality
Research into bone fragility is impeded because there is no accurate, precise and inexpensive method for measuring bone strength-the ability to resist fracture 1 .For many years the most widely used technique for esti-mating bone strength has been densi-tometry, which measures bone mineral density (BMD).A variety of imaging techniques have been employed to measure BMD including dual--energy X--ray absorptiometry or DXA 2 , ultrasound 3 and peripheral computed--tomography or pQCT 4 .Originally it was thought that bone strength was almost entirely explained by density 5 .However clin-ical observations did not support the data, pharmaceutical trials revealed that anti--resorbtive therapies (such as bisphosphonates) reduced frac-tures to a greater degree than predicted from increases in BMD: see 6 and references therein.This was because densitometry failed to take into account the importance of cortical geometry and trabecular architecture for bone strength.Many research articles have since shown that BMD accounts for only about 40-50% 7 of the in vitro compressive strength of a bone whilst structure can account for as much as 30-40% 8 .Following these discoveries, the mate-rial (i.e.density) and structural (i.e.non--density) factors were combined into a new understanding of bone strength-termed bone quality, oper-mechanical basis of bone strength 1,9 .Quality is an amalgamation of all the factors that determine how well the skeleton can resist fracturing, such as micro--architecture, accumulated microscopic damage, the quality of collagen, the size of mineral crystals and the rate of bone turnover 10 .

Aims and objectives
Although the concept of bone quality provides a framework for summa-rising and explaining the determi-nants of bone strength a metric, method or protocol for measuring bone quality has been elusive.At present there are no satisfactory clin-ical means to assess bone quality.Such a protocol would be very useful for screening, monitoring and treating Critical Review bone fragility.Therefore, the aim of this paper is to discuss how bone quality might be visualised, quanti--

Discussion
Imaging bone quality non-invasively Non--invasive 3D imaging techniques can provide structural information about bone, beyond simple densitom-etry 11 .The obvious candidates for non--invasive imaging of bone quality are CT 12-13 and MRI 14 .CT is a radio-graphic imaging technique that maps tissue density distribution, as meas-ured by X--ray transmission (Figure 1).waves to produce an image that is dependent on the distribution of hydrogen in the body.Each modality creates a 3D computerised model made of voxels (the three--dimensional equivalent of a pixel), each assigned a grey value based on the tissues repre-sented within.
The main factor limiting the useful-ness of CT and MRI is spatial resolu--tion-i.e. the ability to resolve two objects of similar density/hydrogen content respectively that are situated close to one another.Resolution is largely determined by the size of the voxels (Figure 2).Typically the reso-lution of a 3D scan is between 2 to 5 times greater than the voxel size 15 .In vivo CT scanners produce scans with smaller voxels than MRI and there-fore have the potential to create higher resolution images of bone structure.However, the resolution of CT scans is also dependent upon the energy of the X--ray beam and is there-fore limited by dose (Figure 1).
The most common in vivo CT systems are volumetric scanners (vQCT) such as the whole body scan-ners typically found in hospitals.The smallest voxels are usually around 0.3×0.3×1.0 mm (pixel length × width × slice).Hence the systems can be used to visualise cortical geometry and trabecular density distribution at the macro--scopic level.Individual trabeculae cannot be visualised because the elements (<0.250 mm)

Critical Review
disappear inside the voxels.A phenomenon referred to as partial volume averaging which results from materials of different density occu-pying a single voxel and thus being represented by an averaged grey value (Figure 3).More recently though high--resolution (hrCT) systems have been developed than can produce voxels that are 0.090×0.090×0.200mm (e.g.XtremeCT, Scanco, Switzerland).The systems can image trabecular micro-architecture but the trade off is hrCT systems are generally restricted to imaging only the periphery of the body such as wrists and ankles.
Likewise hospital MRI scanners typically scan voxels approximately 0.500 3 mm, but high--resolution (hrMRI) systems that can achieve 0.100 3 to 0.200 3 mm are in develop-ment [16][17][18] .Unlike clinical--CT systems (which are limited by the energy and therefore the path length of the x--rays) the hrMRI systems are not restricted to peripheral regions of the body.MRI is not ideally suited to imaging bone though because scanners map the distribution of water on the body and hard tissues have a relatively low water content (Figure 4).Conse-quently the MRI signal for trabecular bone itself is not visualised as such and trabeculae appear as a signal void surrounded by high--intensity fatty bone marrow 19 .It is possible to visu-alise the bone more clearly by simply inverting the image grey scale.

Quantifying bone quality non-invasively
Currently the key to measuring bone strength in vivo using either CT or MRI is to get a handle on the meaning of the voxel grey values.This has been attempted in two ways.The distribu-tion of grey values has been used to quantify the macro--and even some aspects of microstructure that are correlated with mechanical proper-ties 20 .An alternative approach has been to measure mechanical properties more directly by using 3D image data to create computer models for 'virtual' mechanical testing 21 .Experimental mechanical testing can be used to vali-date computer--modelled measure of mechanical properties (Figure 5) Baum and colleagues [22][23] attempted to estimate the mechanical strength of bone using low--res in vivo CT and MRI scans of the proximal femur and distal radius respectively.Cadaveric femora were clinical--CT scanned at 0.190×0.190×0.500mm voxel size, whilst the radii were hrMRI imaged at 0.156×0.156×0.300mm.Hence the in plane pixel size was small enough to visualise the largest trabeculae but resolution was ultimately limited by the slice thickness.In both studies trabecular macro--structure was char-acterised by measuring bone [volume] fraction in 3D.Microstruc-ture was analysed in 2D by applying traditional histomorphometric tech-niques such as the medial intercept length method to calculate trabecular thickness, number and separation 24 .Given the large size of the voxels in comparison to individual trabeculae the measurements are usually referred to as 'apparent' because the scans cannot actually resolve the Critical Review strength respectively for CT and MRI based data were apparent trabecular separation (r 2 =0.511) and bone volume fraction (r 2 =0.548), which were only moderate correlations.However, by also including measures of bone mineral content collected using DXA scans (e.g. Figure 6) the authors were able to improve r 2 =0.760 for CT and r 2 =0.7744 for MRI.Hence apparent trabecular morphology alone was only able to explain 50-55% of the variation in bone strength, but the inclusion of areal bone mineral density increased this to as much as 77%.
These results suggest that low--reso-lution (i.e.> 0.3 mm voxel size) 3D scanned data were not useful for predicting bone strength.However, this may be due to the particular measure-ment techniques.The 2D histomorpho-metric measures that were applied in 2D are known be inaccurate in compar-ison to 3D data, even when collected at higher resolution.Micro--CT imaging would have been more useful for imaging trabecular micro--architecture (Figure 7).More importantly the increase in explained variation with the inclusion of mineral content suggests that using the 3D image data it was not possible to tease apart the volume of bone and its mineral content.Due to volume averaging the voxels blurred out the trabeculae, thus it was possible to get the same grey value representing either a large volume of bone with low mineral content or vice versa.Essen-tially any successful voxel based measure might need to be able to sepa-rate the effects of bone volume and mineral density.This may only be achievable at much smaller voxel size e.g.0.020-0.200mm.A study that vali-dates low resolution measures of struc-ture and density distribution against high resolution is therefore required.
Given that low--resolution analyses of bone structure alone were not able to strongly predict bone strength, it may be necessary to measure mechan-ical properties more directly.For example, using micro--CT scans to  elements (Figure 1).Femoral strength was experimentally measured using a side impact test to simulate a lateral fall on the greater trochanter.The forearms were biomechanically tested in a fall simulation using a uniaxial testing machine and the maximum failure load (i.e.ultimate strength) was recorded.Multiple regression models were used to determine which variables best predicted bone trabecular structural measures with femoral and radial bone strength amounted to between r = 0.428 and r = 0.740.The single best predictors of Critical Review create computer models of bones and estimating whole bone strength by simulating loading.Several researchers have used voxel based 25 modelling to predict the compressive strength of bone (see Figure 8).Crawford and colleagues 26 examined clinically--CT scanned verte-brae with a large voxel size of 0.674 3 mm.The scans were used to create 3D models of the bones that could be 'virtually' loaded on the computer a mesh that described variation in bone volume and mineral density.Importantly the models were constructed from the CT scans using automated algorithms programmed by the authors.After scanning, the ultimate compressive strength of the vertebrae was measured experimen-tally using a mechanical testing rig.The authors reported that the model predicted 86% of the variation in compressive strength.Thus it appears as though computer modelling could be used to accurately quantify bone strength non--invasively.Fracture loads can be predicted more accu-rately using 3D computer modelling than DXA data [27][28] .Furthermore, given that the scans were very low resolu-tion and therefore quick to collect and given that the mechanical modelling was automated it is entirely feasible to use the method in a clinical setting.
Several studies have also used in vivo models for analysing bone mechanical properties [29][30][31] .Unlike the CT based studies described above the models were not validated using physical mechanical test data.Since the voxel sizes were large (0.410 to 1.0 mm slice thickness) such a step would be neces-sary.To date only one in vivo MRI based study has attempted to corrobo-rate the computer modelled mechan-ical properties, comparing values measured using hrMRI and micro--CT as the gold standard 32 .Cadaveric distal tibia were MRI scanned at 0.160 3 mm voxel size but the resolution of the  Micro--CT scans of (A) osteoporotic and (B) osteoarthritic femoral heads were compared using BoneJ which can display thickness as a heat map which is easy to understand.For example, note that the osteoarthritic trabeculae are thicker and better connected than the osteoporotic elements.

Critical Review
micro--CT data was higher 0.250 3 mm.Trabecular stiffness and elastic moduli (i.e.ability to withstand a load without deforming) were computed.Stiffness measures calculated using the two modalities were highly correlated (r 2 =0.96) whilst elastic moduli were not (r 2 =0.58).The authors concluded that in vivo MRI scans could probably be used to measure some mechanical properties accurately.Further testing and validation, preferably against experimental data is required to deter-mine what information can be obtained, and which measures would be the most useful.

Applying bone quality in a clinical setting
In order to guide research into metrics for bone quality it is necessary to consider the end clinical uses.A better understanding of bone quality could monitoring of pharmacological treat-ments and surgical interventions of patients with fragile bone 33 .The holy grail of would be a predictive test for osteoporotic fracture risk.For example, the 10--year probability of fracture is the most desirable meas-urement to determine intervention thresholds 34 .As yet though there are no studies demonstrating prospective fracture risk prediction 35 .

Predicting 10-year fragility fracture risk
Osteoporotic or fragility fractures due to poor bone density are esti-mated to affect 200 million people worldwide 36 and 300,000 patients in the UK alone 37 .Yet the condition is substantially under diagnosed and under treated [38][39][40][41] .Furthermore the situation is getting worse.A study based in Canada revealed that between 1996 and 2002, the number of patients diagnosed with osteopo-rosis and receiving treatment increased from 6.1% to 12.3%, but then steadily declined to 5.9% by 2008 42 .As UK life expectancy increases and the population ages the number of fractures is expected to rise dramatically 43 .Recently there has been move by the WHO to set up tools such as FRAX (www.shef.ac.uk/FRAX/).The computer driven system uses algorithms to process patient probability of an osteoporotic frac-family history and lifestyle as well as a DXA scan (Figure 6).Recent studies have shown that although the system is reasonably accurate the algorithms tend to underestimate the risk of fracture in women, particularly those in the most at risk group over 65 years [44][45] .Improved metrics for bone quality collected from 3D CT scans could increase the predictive power mechanical properties based on element models for loading bones virtually.

Monitoring pharmacological treatments
Aftercare in osteoporotic fracture also focuses on improving bone quality to prevent further fractures through various pharmacological means (e.g.calcium, vitamin D and more recently bisphosphonates).After identifying patients with fragile bones repeat clinical CT scans could be used to monitor disease progression and/or monitor pharmacological treatment outcomes.For example, bisphospho-nates are highly effective in the treat-ment of osteoporosis.Numerous large clinical trials have demonstrated their Critical Review increasing bone mass and mineral density and reducing fracture risk.Consequently bisphosphonate therapy, in particular alendronate, has become the mainstay of bone fragility treatment since 1995.However, treat-stress fractures after long--term treat-ment e.g.5-10 years 46 .At least in part due to accumulation, propagation and merging of micro--cracks (Figure 9).Many studies have demonstrated that bone mass plateaus after 3-5 years of therapy 47,48 but, if treatment ceases, there can be a slight loss 49 .Therefore CT based measures of fragility frac-ture risk could potentially be used to set treat to target criteria for bisphos-phonate therapies, monitor progres-sion, identify the time point at which the effect of the drug starts to slow or increase stress fracture risk (Figure 9) and implement treatment holidays.Bisphosphonates are known to stay active in the body for up to 7-10 years after treatment 50 but holidays will help minimise the risk of fracture complications.

Informing surgical interventions
When osteoporotic fractures do occur ---culty in obtaining secure implant obtained before interventions could be used to inform implant and surgical choices.For example, neck of only if the bone is strong enough to hold the screws.Reduced cortical and cancellous bone mass decreases the ability of screw threads to gain purchase, which hugely decreases pullout strength and results in increased implant failure 51 .Surgeons could assess a patient's suitability for based assessment of bone quality.When screws will not hold, other implant designs such as total hips with acetabular and femoral compo-nents are usually more appropriate.
-tion the components need to osseoin-tegrate but osteoporosis alters the biomechanical properties of bone, making tissue stiffer and more brittle.Consequently the load transmitted at the bone--implant interface can often exceed the strain tolerance of osteo-porotic bone 51 causing micro--damage that leads to micro--fracture, resorp-tion of bone, implant loosening and subsequent implant failure 36 which could happen within months after surgery.In the long term there is potential for disease or patient matched implants to be built that replicate the biomechanical proper---ysis.In the short term there is a need to develop a follow up protocol to identify patients that exhibit resorb-tion before the implants fail, perhaps using 3D imaging data.Those patients exhibiting such failure potential would need to be restricted in terms of loading the bone--implant construct early and have earlier repeat surgical intervention and augmentation of -back of CT in this respect is that the x--rays cannot penetrate metal, which introduces streak artefacts and noise, blurring the image and making anal-ysis of bone shape and mechanics -able because the use of a magnetic implants from being scanned.
Clinical--CT scans collected preop-eratively that describe 3D variation in bone quality around a fracture site could be used by surgeons to select the most appropriate implant and the locations at which screws or nails osseointegration.Patients with enough healthy bone tissue should prevent excessive rigidity that may in

Critical Review
turn delay bone healing.While regions of poor bone quality will locking plates and screws to improve been developed for the common fragility fractures.The key change over conventional devices was the coupling of the screw to the plate, achieved by conically shaped threads in the screw head matching threads in the plate, which allows the screw to effectively bolt into the plate.The singular stable screws prevent load concentration at a single bone--screw interface by distributing load more evenly 52 .Similarly intramedullary nails and other relative stability tech-niques, such as dynamic hip screws, have been successfully employed to treat complex proximal femoral frac-tures in the elderly 53 .Buttressing a fracture by applying force at 90 degrees to the axis of a potential deformity (thereby providing a construct that resists axial load) is an effective method in metaphyseal osteoporotic fractures because it reduces strain at the bone implant interface 53 .

Conclusion
The understanding of what consti-tutes bone quality and how it can be measured may lead to better predic-tions of fracture risk, as well as improved diagnosis, management, treatment, and monitoring of patients with fragile bone.Non--invasive methods are restricted to low--resolu-tion whole body scans or high--resolu-tion peripheral scans as yet.Clearly research needs to be directed to improving imaging technology.High-resolution CT and MRI systems are already becoming available on the market.However, radiation dosage, resolution scan time and cost are always going to be limiting factors no matter how much equipment improves.Accordingly it is essential -fying bone quality at the lowest possible resolution.Even at sub--optimal resolution bone quality could be characterised by measuring one or more of the structural and material aspects of bone, or by calculating the mechanical properties more directly using computer models.If the tech-nique is going to translate into a clin-ical setting it will also be necessary to create an automated computerised system.One that can collect a scan of key fracture sites, automatically virtually load the bone and provide relevant mechanical data is an imme-diate possibility.The data can then be used to inform primary and secondary care of patients.Mechanical proper-ties could be entered into a FRAX (or similar) system, in place of BMD meas-ures.Surgeons could utilise 3D maps of bone quality distribution to plan interventions, select implant type and the optimal location for screws.

Figure 1 :
Figure 1: Computed tomographic imaging modalities.Clinical--CT scanners use lower radiation levels and can scan whole bodies but the resolution is too low to visualise tissue level structures.Micro--and nano--CT can image individual trabeculae and even micro--cracks respectively, but the radiation is too high for in vivo scanning.

Figure 2 :
Figure 2: The effect of voxel size the accuracy of 3D models.With increasing voxel the spatial resolution of a scan decreases.From (A) 0.050 to (B) 0.100 mm the 3D structure can be clearly visualised.At (C) 0.200 mm the larger voxels start to miss some features (compare top left hand corners of models).Above (D) 0.300 and (E) 0.400 mm the architecture deteriorates.

Figure 3 :
Figure 3: Partial volume averaging.A micro--CT slice is comprised of voxels.Partial volume averaging occurs when materials of different density (i.e.bone and air) occupy the same voxel.The CT (grey) value assigned to each voxel

Figure 4 :
Figure 4: MRI cross section at the level of the femoral head, in which the bone mass and structure is not clearly visible.

Figure 5 :
Figure 5: Femoral head trabecular core (A) before and (B) after compression testing.The rig measures mechanical properties such as strength and stiffness which can be used to quantify bone quality and perhaps to calculate whole bone fracture risk.

Figure 6 :
Figure 6: (A) DXA scans measure (B) bone mineral density (BMD) but not structure.Since structure accounts for 40-50% of bone strength the BMD data is not highly correlated with bone mechanical properties or fracture risk.

Figure 7 :
Figure 7: Micro--CT scans are excellent for visualising, measuring and describing trabecular architecture e.g.thickness.Micro--CT scans of (A) osteoporotic and (B) osteoarthritic femoral heads were compared using BoneJ which can display thickness as a heat map which is easy to understand.For example, note that the osteoarthritic trabeculae are thicker and better connected than the osteoporotic elements.

Figure 8 :
Figure 8: Finite element analysis is a computerised version of mechanical testing.(A) A CT scan is imaged and used to (B) build a 3D computer model of the bone, or a volume of interest within.(C) The model is 'virtually' loaded, typically in compression or tension and (D) the stress/strain distribution is

Figure 9 :
Figure 9: Bone micro--cracks (white arrows) and are repaired by remodelling.Bisphosphonates supress turnover, particularly bone resorption, leading to increased bone mass.However, over--suppression leads to accumulation of micro--cracks.Micro--cracks can be imaged using (A) nano--CT scans and (B) thresholding the crack void (C) in 3D.(D) An FE analysis of the micro--crack, based on the scan, revealed high stress concentrations at the tip which could cause the crack to propagate.