Reproducibility in simulation-based prediction of natural knee mechanics

点击翻译按钮获取中文摘要

PROJECT SUMMARY Simulations of knee biomechanics are increasingly utilized to develop individualized and actionable knowledge regarding healthy homeostasis, impact of musculoskeletal diseases, and risk of injury. Computational modeling is also appealing to predict performance and safety of implants, and informs surgical tissue reconstruction and rehabilitation strategies. This is a natural result of the ubiquity of modeling and simulation in biomedicine, and it is motivated by the knee being a primary concern in musculoskeletal healthcare. However, start-to-end and turn- key examples of mechanistic simulations, to accelerate biomedical discovery and to routinely inform clinical care, are scarce. In knee biomechanics specifically, the compelling promise of modeling and simulation has not been fulfilled. Delivery of computational models and their intended utilization require many operations that collectively define the modeling and simulation workflow. Variations in the modeler’s choices and the ambiguity of their implementation introduce uncertainties that impact reproducibility across the modeling and simulation lifecycle, from intermediate products to end-point simulation results. Reproducibility is essential to the broader credibility of modeling practices. Its absence within an acceptable threshold, as dictated by the model’s intended use, is a significant barrier for adoption of simulation. Scalable uses of knee models are also impeded by the burden of modeling and simulation activities and a lack of specialized guidance. In the past award period, we documented the art of modeling and simulation in knee biomechanics. Starting with the same data and with the intent to simulate the same model use cases, five teams demonstrated how each team’s processes vary from others. This resulted in differences in anatomical and mechanical representations of tissue structures and, most importantly, simulation predictions. In the proposed project, we first aim to determine consensus workflows that represent good practices in computational knee biomechanics. Consensus will be established specific to a diverse set of model contexts of use as credibility frameworks already acknowledge the dependence of the depth and intensity of modeling activities on the intended use of models. Context of use also dictates the fidelity of the model, the amount of data to build the model and validate its predictions, computing resources, and subject- specificity. Consensus building will be facilitated by the Delphi method, which will provide the means for a structured and iterative outreach to the knee modeling experts worldwide. In the second aim, the reproducibility and accuracy of context relevant consensus workflows will be tested against required resolution of simulation outputs. Five teams will develop models and conduct simulations using the same data, with the same intent, and this time, relying on the same modeling and simulation workflow. Documented and disseminated consensus workflows, including outcomes and by-products of use cases, will unify, and possibly standardize, the highly fragmented ecosystem of knee modeling approaches. Subsequently, they will enable multi-site, large scale, and dependable in silico trials in knee biomechanics.