Validating subject-specific knee models from in vivo measurements

Despite the documented consequences of modeling decisions on the performance of computational models in orthopaedics and biomechanics, the influence of the input data has largely been ignored. Modeling the living knee is limited by methods to measure <i>in vivo</i> the quantities needed for ligament calibration; yet, this may be possible with new devices focused on non-invasive measurement of knee laxity. These devices offer measurements similar to those commonly obtained from cadaveric specimens but are limited by what can be practically and safely obtained from a living subject. Validation of models calibrated with <i>in vivo</i> data is crucial and increasingly important as personalized modeling becomes the basis for proposed digital twins, and <i>in silico</i> clinical trial workflows. To support our overall goal of building subject-specific models of the living knee, we aimed to show that subject-specific computational models calibrated using <i>in vivo</i> measurements would have accuracy comparable to models calibrated using <i>in vitro</i> measurements. Two cadaveric knee specimens were imaged using a combination of computed tomography (CT) and surface scans. Knee laxity measurements were made with a custom apparatus used for the living knee and from a robotic knee simulator. Models of the knees were built following previous methods and then calibrated with either laxity data from the <i>in vitro</i> robotic knee simulator (RKS) or from the <i>in vivo</i> knee laxity apparatus (KLA). Model performance was compared by simulation of various activities and found to be similar between models calibrated with laxity targets from the RKS and the KLA. Model predictions during simulated anterior-posterior laxity tests differed by less than 2.5 mm and within 2.6° and 2.8 mm during a simulated pivot shift. Still, differences in the predicted ligament loads and calibrated material properties emerged, highlighting a need for methods to include ligament load as part of the calibration process. Overall, the results showed that currently available methods of measuring knee laxity <i>in vivo</i> are sufficient to calibrate models comparable with existing <i>in vitro</i> techniques, and the workflows described here may provide a basis for modeling the living knee. The experimental data, models, results, and tools are publicly available.