Is the plate to rod transition in trabecular bone loss a real phenomenon or a spurious result of a misused metric?

A latticework of bony trabeculae stiffens the internal spaces of many bones. Trabeculae transmit forces between joint surfaces and bone shafts and maintain the integrity of vertebrae. Bone loss is a normal physiological response to reduced mechanical load, occurring during bed rest, in sedentary lifestyles, and during the weightlessness of spaceflight. Bone loss is also a normal, undesirable, part of ageing which occurs particularly rapidly in women after menopause. Reduced bone mass is strongly associated with increased fracture risk. Peak bone mass is obtained in humans in early adulthood. Strategies to increase peak bone mass and to slow its physiological loss are of great interest to ensure healthy ageing.

In addition to bone mass (bone volume fraction, BV/TV), bone architecture is thought to play a crucial role in trabecular bone's force transmission and fracture resistance. Bone architecture is defined by standard measurements such as trabecular thickness (Tb.Th), spacing (Tb.Sp), interconnectedness (Conn.D), anisotropy (DA) and rod-plate geometry (structure model index, SMI). The bulk behaviour of trabecular bone in physiological loading is almost entirely determined by BV/TV: this is intuitive because the more bone matrix there is per unit volume, the greater the overall volume can resist applied load. Once a higher failure load has been reached, however, the behaviour of trabecular elements dominates collapse of the overall structure. Each trabecula might fail by buckling, shear or crumpling. Bone biologists believe that plate-like trabeculae resist higher loads than rod-like trabeculae and that plate-like trabeculae convert to rod-like trabeculae during bone loss, creating a 'double-whammy': reduced bulk resistance to load due to decreased BV/TV and reduced resistance to trabecular element failure due to rod-like geometry.

Whether a plate-to-rod transition actually occurs in bone loss is now in serious doubt, because the paradigm is based on flawed SMI measurements. I recently showed that variation in SMI during bone loss is dominated by increasing amounts of concave portions of the bone surface, which the SMI theory assumes are not there or have a negligible contribution. There is a high correlation between SMI and BV/TV even when the underlying architecture has not changed. SMI is a fundamentally broken measurement that does not do what bone bioscientists use it for. Over 850 papers have cited SMI and a common interpretation when seeing BV/TV and SMI results together is that bone loss is associated with a plate-to-rod transition, however, this is merely an artefact of SMI's design that obscures any true relationship. Statements linking bone loss and plate-rod transition have been repeated often enough to have become an accepted dogma in the bone field.

This project aims to overturn the dogma there is a plate to rod transition in bone loss using 3D X-ray microtomography (XMT) image data sets archived from previous experiments. To measure rods and plates independent of BV/TV, the PDRA will validate a new method, ellipsoid factor (EF), that defines rods and plates by the shape of the biggest ellipsoid that fits within each bony region. Rods can fit long, javelin-shaped ellipsoids, while plates can fit flat, discus-shaped ellipsoids. The PDRA will use image data collected on RVC's XMT instrument, images from my project partner Phil Salmon at Bruker microCT, and from other collaborators worldwide, and aim to increase our current pool of ~150 images from 3 studies to ~1000 images from ~20 studies. Having identified rods and plates in bone samples, the PDRA will correlate rods and plates with mechanical behaviour using finite elements analysis in the physiological load range for bulk behaviour and overload range for trabecular element failure behaviour. The primary benefit of the work is improved understanding of true geometric and mechanical changes in bone loss, along with the new EF method and free software.

Physiological bone loss is believed to be associated with a transition from plate-like to rod-like trabeculae. This belief is based on a flawed metric published in 1997, structure model index (SMI), which artefactually links bone volume fraction (BV/TV) to rod/plate geometry due to the unexpected influence of concave surface curvature. Such a plate-rod transition may exist, but SMI cannot measure it. Despite this, SMI has been cited over 850 times in published studies and the plate-rod transition in bone loss has become 'common wisdom' in the field. Because BV/TV dictates the majority of trabecular bone's mechanical behaviour, and SMI is collinear with BV/TV, bone biologists may have inadvertently linked rod/plate bone architecture with mechanical behaviour. This project will test the hypothesis that there is a real correlation between BV/TV and rod-plate geometry using a new unbiased metric, ellipsoid factor (EF), which measures rods and plates using axis ratios of maximally-fitting ellipsoids fit to 3D images of trabecular bone. Before use in biological hypothesis testing, EF must be further validated and its performance improved. X-ray microtomography image data will be obtained from prior studies where bone loss is a prominent feature, from RVC and collaborators' collections. The project will also quantify the mechanical consequences of any such plate-rod change under normal loading conditions and under mechanical overload (fracture) conditions using computational finite elements analysis.

The research will result in tangible outcomes:
-further evidence that SMI is flawed and must no longer be used to measure rods and plates in trabecular bone, changing bone bioscience practice
-reliable data on the plate-rod transition, if it occurs
-correlations between an unbiased measure of trabecular bone shape and mechanical behaviour, informing image-based predictors of fracture risk
-open source and commercial software implementing the ellipsoid factor algorithm

The primary impact of this proposal is to end two decades of misuse of a flawed bone measurement, SMI, which has been cited over 850 times in the bone literature, and which continues to be cited and misused. The project will develop a replacement for SMI, ellipsoid factor, which is an algorithm that produces unbiased measurement of rods and plates in trabecular bone. The beneficiaries of this research are bone bioscientists who will benefit by changing their practice to stop using a broken metric (SMI) and better understand the true rod/plate nature of trabecular bone and its mechanical implications.

A very broad range of specimens could be analysed with ellipsoid factor: useful information would be found in porous continua from very many fields of endeavour of relevance to the UK. In diagnostic imaging, the structure of trabecular bone contributes to fragility fractures, particularly in the elderly. In materials science, the mechanical performance of structural foams is important for the spongy motors of soft robots; in the oil, gas and water industries the pores in rock determine rate of flow during extraction; in food technology the structure of edible foams (cakes, bread, espresso, ice cream) is of great importance to their palatability.

Tangible impact will be made by providing ellipsoid factor algorithms in both open-source and proprietary software forms to academic and industrial beneficiaries. The open source version will be distributed via the internet as a Java plugin for the widely used ImageJ image analysis program. It will run on any operating system (Windows, Mac OS X and Linux) and on any suitable 3D image data. Making the ImageJ plugin free to download means that anyone with a tomographic dataset, from any instrument (including benchtop and synchrotron sources), will be able to run an ellipsoid factor analysis with a very low barrier to adoption. By making the code open source it will be both freely available, and very importantly, free to study. Providing code leads to impact in difficult-to-predict fields: the ellipsoid code has already been used by Foster + Partners' Specialist Modelling Group to speed up architectural design iteration time by 50% on a structural optimisation project. Scientists and engineers are free to examine and improve the code and to submit their improvements back to us via the Git distributed source code management tool. The open source version will be a reference implementation against which optimised versions in other languages may be developed and tested. To aid in testing and reimplementation efforts, a suite of standard test images will be produced and published for free download and hardened for research use by providing digital object identifiers (DOIs, for persistent identification) and hash sums (e.g. MD5 or SHA-1, for unique identification).

With my project partner Bruker microCT the PDRA will produce and publish commercially, under a Collaboration Agreement licensing scheme, an optimised C++ implementation of ellipsoid factor as a plugin for Bruker's CTAn analysis software. This will be provided for a fee to users of Bruker's SkyScan X-ray microtomography imaging devices, so that these users can readily include an ellipsoid factor analysis in their automated morphology pipeline. We may develop intellectual property worth patenting, in particular those aspects that relate to specific execution optimisations. The IP and commercial product may be popular enough to develop a new spinout company that works on optimisations or performs analysis as a service, or consults on the development of third party implementations. This would result in inward investment to the UK as the development activities would be based at (or near) the RVC, and licensing fee revenues accruing to my RVC research group.

The PDRA will benefit from being employed on the project by learning technical skills, developing their bone knowledge and improving industry and academic career options.