The Capacity of Generic Musculoskeletal Simulations to Predict Knee Joint Loading Using the CAMS-Knee Datasets
Zohreh Imani Nejad, Khalil Khalili, Seyyed Hamed Hosseini Nasab, Pascal Schütz, Philipp Damm, Adam Trepczynski, William R. Taylor, Colin R. Smith
Abstract
Abstract Musculoskeletal models enable non-invasive estimation of knee contact forces (KCFs) during functional movements. However, the redundant nature of the musculoskeletal system and uncertainty in model parameters necessitates that model predictions are critically evaluated. This study compared KCF and muscle activation patterns predicted using a scaled generic model and OpenSim static optimization tool against in vivo measurements from six patients in the CAMS-knee datasets during level walking and squatting. Generally, the total KCFs were under-predicted (RMS: 47.55%BW, R 2 : 0.92) throughout the gait cycle, but substiantially over-predicted (RMS: 105.7%BW, R 2 : 0.81) during squatting. To understand the underlying etiology of the errors, muscle activations were compared to electromyography (EMG) signals, and showed good agreement during level walking. For squatting, however, the muscle activations showed large descrepancies especially for the biceps femoris long head. Errors in the predicted KCF and muscle activation patterns were greatest during deep squat. Hence suggesting that the errors mainly originate from muscle represented at the hip and an associated muscle co-contraction at the knee. Furthermore, there were substaintial differences in the ranking of subjects and activities based on peak KCFs in the simulations versus measurements. Thus, future simulation study designs must account for subject-specific uncertainties in musculoskeletal predictions.