There is a critical need for safe innovation in total joint replacements to address the demands of an ageing yet increasingly active population. The development of robust implant designs requires consideration of uncertainties including patient related factors such as bone morphology but also activity related loads and the variability in the surgical procedure itself. Here we present an integrated framework considering these sources of variability and its application to assess the performance of the femoral component of a total hip replacement (THR).

The framework offers four key features. To consider variability in bone properties, an automated workflow for establishing statistical shape and intensity models (SSIM) was developed. Here, the inherent relationship between shape and bone density is captured and new meshes of the target bone structures are generated with specific morphology and density distributions. The second key feature is a virtual implantation capability including implant positioning, and bone resection. Implant positioning is performed using automatically identified bone features and flexibly defined rules reflecting surgical variability. Bone resection is performed according to manufacturer guidelines. Virtual implantation then occurs through Boolean operations to remove bone elements contained within the implant's volume. The third feature is the automatic application of loads at muscle attachment points or on the joint contact surfaces defined on the SSIM. The magnitude and orientation of the forces are derived from models of similar morphology for a range of activities from a database of musculoskeletal (MS) loads. The connection to this MS loading model allows the intricate link between morphology and muscle forces to be captured. Importantly, this model of the internal forces provides access to the spectrum of loading conditions across a patient population rather than just typical or average values. The final feature is an environment that allows finite element simulations to be run to assess the mechanics of the bone-implant construct and extract results for e.g. bone strains, interface mechanics and implant stresses. Results are automatically processed and mapped in an anatomically consistent manner and can be further exploited to establish surrogate models for efficient subsequent design optimization. To demonstrate the capability of the framework, it has been applied to the femoral component of a THR.

An SSIM was created from 102 segmented femurs capturing the heterogeneous bone density distributions. Cementless femoral stems were positioned such that for the optimal implantation the proximal shaft axis of the femurs coincided with the distal stem axis and the position of the native femoral head centre was restored. Here, the resection did not affect the greater trochanter and the implantations were clinically acceptable for 10000 virtual implantations performed to simulate variability in patient morphology and surgical variation. The MS database was established from musculoskeletal analyses run for a cohort of 17 THR subjects obtaining over 100,000 individual samples of 3D muscle and joint forces. An initial analysis of the mechanical performance in 7 bone-implant constructs showed levels of bone strains and implant stresses in general agreement with the literature.

Inter-subject variability is inherently present in patient anatomy and is apparent in differences in shape, size and relative alignment of the bony structures. Understanding the variability in patient anatomy is useful for distinguishing between pathologies and to assist in surgical planning. With the aim of supporting the development of stratified orthopaedic interventions, this work introduces an Articulated Statistical Shape Model (ASSM) of the lower limb. The model captures inter-subject variability and allows reconstructing ‘virtual’ knee joints of the lower limb shape while considering pose.

A training dataset consisting of 173 lower limbs from CT scans of 110 subjects (77 male, 33 female) was used to construct the ASSM of the lower limb. Each bone of the lower limb was segmented using ScanIP (Simpleware Ltd., UK), reconstructed into 3D surface meshes, and a SSM of each bone was created. A series of sizing and positioning procedures were carried out to ensure all the lower limbs were in full extension, had the same femoral length and that the femora were aligned with a coincident centre. All articulated lower limbs were represented as: (femur scale factor) × (full extension articulated lower limb + relative transformation of tibia, fibula and patella to femur). Articulated lower limbs were in full extension were used to construct a statistical shape model, representing the variance of lower limb morphology. Relative transformations of the tibia, fibula and patella versus the femur were used to form a statistical pose model. Principal component analysis (PCA) was used to extract the modes of changes in the model.

The first 30 modes of the shape model covered 90% of the variance in shape and the first 10 modes of the pose model covered 90% of the pose variance. The first mode captures changes of the femoral CCD angle and the varus/valgus alignment of the knee. The second mode represents the changes in the ratio of femur to tibia length. The third mode reflects change of femoral shaft diameter and patella size. The first mode characterising pose captures the medial/lateral translation between femur and tibia. The second mode represents variation in knee flexion. The third mode reflects variation in tibio-femoral joint space.

An articulated statistical modelling approach was developed to characterize inter-subject variability in lower limb morphology for a set of training specimens. This model can generate large sets of lower limbs to systematically study the effect of anatomical variability on joint replacement performance. Moreover, if a series of images of the lower limb during a dynamic activity are used as training data, this method can be applied to analyse variance of lower limb motion across a population.

Polymer foams have been used extensively in the testing and development of orthopaedic devices and for verification of computational models. Their use is often preferred over cadaver and animal models due to being relatively inexpensive and their consistent material properties. Successful validation of such models requires accurate material/mechanical data. The assumed range of compressive moduli, provided in the sawbones technical sheet, is 16 MPa to 1.15 GPa depending on the density of foam. In this investigation, we apply two non-contact measurement techniques (digital volume correlation (DVC) and optical surface extensometry) to assess the validity of these reported values. It is thought that such non-contact methods remove mechanical extensometer errors (slippage, misalignment) and restrict the effect of test-machine end-artifacts (friction, non-uniform loading, platen flexibility). This is because measurement is taken directly from the sample, and hence material property assessment should be more accurate. Use of DVC is advantageous as full field strain measurement is possible, however test time and cost is significantly higher than extensometry. Hence, the study also sought to assess the viability of optical extensometry for characterising porous materials.

Testing was conducted on five 20 mm cubic samples of 0.32g/cc (20 pcf) solid rigid polyurethane foam (SAWBONES^TM). The strain behaviour was characterised by incremental loading via an in situ loading rig. Loading was performed in 0.1 mm increments for 8 load steps with scans between loading steps. Full field strain measurement was performed on one sample by micro focus tomography (muvis centre, Southampton) and subsequent DVC (DaVis, Lavision). Calculation of Young's modulus and Poisson's ratio was then preformed through use of the virtual fields method. These results were subsequently corroborated by use of optical extensometry (MatchID). To account for heterogeneities, axial strain measurements were averaged from six points on the front and rear surfaces. A computationally derived correction factor was then applied to account for through volume strain variations. In each test compressive displacement was applied to 900N (∼2MPa) to remain within the linear elastic region.

Significant variability of individual strain measurements were observed from extensometry measurements on the same sample, indicating non-uniform loading did occur in all samples. However by averaging across multiple points linear loading profiles were identified. For all non-contact methods the calculated elastic moduli were found to range between 331–428 MPa whilst the approximated modulus based on cross head displacement was ∼210 MPa. The optical-extensometry gave a considerably higher modulus (p = 0.047) than the DVC results as only surface measurements were made. However, following computational based correction values converged within 6% of one another. Both the DVC and point-tracking results (p = 0.001) indicated substantially higher compressive modulus (137%) than the manufacturer provided properties.

This study demonstrates that methods of measuring displacement data on of cellular foams must be carefully considered, as artefacts can lead to significant errors of up to 137%, and such errors may falsely influence the design and validation of tested devices.