Modelling Every Wear: In Silico, In Vitro, and In Vivo Approaches to Predict Wear of Knee Implants
Loading...
Author / Producer
Date
2023
Publication Type
Doctoral Thesis
ETH Bibliography
yes
Citations
Altmetric
Data
Rights / License
Abstract
Total Knee Arthroplasty is a common surgical procedure for managing arthritis, providing pain relief and improved functionality to patients through implantation of a joint replacement. However, wear of the polyethylene (PE) inlay of knee implants poses clinical challenges, including the potential for implant failure and revision surgeries. Not only understanding, but also the ability to predict wear outcomes in knee implants is crucial for guiding implant design, regulatory approval, and clinical decision making. In vitro wear simulations are realistic, but slow and laborious, while in silico modelling is fast and accessible, but limited by model sensitivity and lack of input data. In this work, we applied these complementary approaches to "modelling every wear": We fitted statistical models to in vivo wear outcomes, performed in vitro testing, and built state-of-the-art computational wear models. Across these domains, we provide novel tools and realistic data, striving towards accurate predictions of in vivo PE wear.
To provide realistic and practically usable input data for experimental testing, we generated a representative summary of the CAMS-Knee datasets, the largest collection of knee implant kinematics and tibiofemoral contact loads measured in vivo for six subjects and five activities. The loads of the created standardized subject “Stan” are similar to earlier datasets, but valuably complemented by synchronized kinematics. Compared to the ISO 14243 standard loads, Stan’s loads are up to +56% higher, while the kinematics exhibit markedly different curve shapes. Application of Stan’s kinematics and the ISO standard loads to a knee simulator wear test revealed not only visibly different wear locations on the articulating surface, but also approximately three times higher wear rates for Stan’s boundary conditions. While further testing under Stan’s conditions is necessary to substantiate this, these initial results indicate that the ISO standards may not be fully representative of in vivo loading and damage.
In a different experiment, we performed an array of pin-on-disk tests to quantify the influence of contact pressure and cross-shear, i.e. multidirectional sliding, on the mechanism and volume of PE wear. Wear was found to strongly increase when going from unidirectional to multidirectional sliding and, contrary to the classical Archard law, not proportionally increase with increasing contact pressure, but increase less at higher pressures. This was due to the formation of hardened protrusions on the PE surface at higher pressures, which afford some protection of the surface. To the wear results, an empirical model of PE wear as a function of cross-shear and contact pressure was fitted to serve as input data for computational wear models.
The empirical PE wear model was then used as input to a computational wear prediction algorithm based on finite-element models of the implant under physiological load and motion. This algorithm can model the change in surface geometry due to nonlinear wear and long-term plastic creep of the PE, the following change in contact mechanics, and the resulting interaction with subsequent wear damage. Verification and validation were carried out against the knee simulator test with the Stan and ISO boundary conditions. While a force-controlled ISO model was unable to mirror the bench test kinematics and thus wear rate, displacement-controlled models accurately predicted experimental wear rates for both ISO and Stan boundary conditions. This analysis confirmed that in silico wear models are very sensitive to even small errors in relative tibiofemoral motion, and that accurate reproduction of in vivo joint kinematics is crucial while some deviation of the contact loads may be permissible.
If in vivo contact kinematics can be modelled, realistic computational predictions of implant wear are thus possible. To this end, we developed a novel technique to control the load and kinematic boundary conditions applied to computational models of joints. In the combined Load- and Displacement-Controlled method with Springs (LDCS), joint contact loads are applied to the model components directly. Then, motions are applied in the same directions as the loads, which would be not possible with conventional control methods. This is achieved by applying the motions through nonlinear springs, which mediate a balance between the applied loads and kinematics and also prevent propagation of any measurement noise present in the joint kinematics to the contact locations. The LDCS method reduced load and motion errors by a factor of two or more compared to conventional approaches and thus presents clear advantages for modelling tibiofemoral contact and wear.
Lastly, we analysed inlays retrieved from revision surgery to quantify in vivo wear, not only providing comparative data for in silico and in vitro wear simulations, but also providing direct evidence of the effect of clinical parameters like choice of implant design and mechanical axis limb alignment on personal wear outcomes. To this end, we validated and employed a surface reconstruction approach to obtain the distribution and volume of wear on the articulating surface of the polyethylene inlays. We found that rotating-platform inlays experienced a significant 39% lower wear rates in situ than fixed-bearing implants, likely due to the rotational freedom which mediates alignment of a rotating inlay to the femoral component, thus reducing contact stresses and motion. Limb alignment, on the other hand, was only non-significantly related to overall wear rates, though there was a slight trend of varus-aligned specimens showing more wear damage. However, limb alignment did significantly alter the mediolateral distribution of wear, with varus alignment resulting in predominantly medial compartment wear. Our analysis also revealed considerable inter-patient variability, likely due to patient-specific factors like level of activity which were not included in our statistical model.
In summary, this research has both added to the wear prediction toolbox by providing validated computational tools and has presented novel in vivo data to enable and validate wear simulations. With Stan, we have provided input data for experimental and computational knee implant wear simulations, where we also found in silico wear models to be sensitive to errors in joint kinematics. With the LDCS method, we provided a modelling technique to drastically reduce such errors when replicating in vivo joint kinematics in a computational model. Lastly, investigating how various parameters affect PE wear, we found contact pressure and multidirectional sliding as well as limb alignment and implant design to influence wear outcomes.
Further experimental and computational studies aiming to truly predict in vivo wear should attempt to account for the observed variability between patients and the sensitivity of models to changed input conditions by including multiple patients, activities, and repetitions in simulations of implant wear driven by accurate joint kinematics.
Permanent link
Publication status
published
External links
Editor
Contributors
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Organisational unit
03994 - Taylor, William R. / Taylor, William R.
Notes
Funding
Related publications and datasets
Has part: https://doi.org/10.3929/ethz-b-000556641
Has part: https://doi.org/10.3929/ethz-b-000578741
Has part: https://doi.org/10.3929/ethz-b-000611211