Musculoskeletal (MSK) modeling and simulation of human movement is needed to estimate biomechanical quantities which cannot be directly measured, including muscle and joint contact forces during wheelchair propulsion. When accurate representation of an individual's muscle strength and musculoskeletal anatomy is needed, a generic upper limb MSK model can be scaled, or a subject-specific model can be developed. A generic MSK model can be scaled with regard to strength to represent an individual nonimpaired adult or child based on limb lengths and muscle volumes, or other techniques. Developing a subjectspecific model requires MSK and physiologic data typically derived from medical imaging and/or cadaveric studies. Therefore, scaling a generic model is an attractive alternative relative to the time and expense of developing subject-specific models. In a previous study, a strength scaled model using limb lengths and muscle volumes predicted pediatric joint moments that were highly correlated with experimentally measured moment-generating capacity in shoulder flexion, extension, abduction, and adduction. The aim of this study is to investigate whether a generic adult upper limb model can be similarly scaled based on limb lengths and muscle volumes to well represent upper limb muscle strength of adults and children with paraplegia.
Our overall approach is to scale a generic model based on reported limb length and muscle volumes, and then compare the model predicted maximum joint moments to the experimentally measured maximum isometric moments from the same subjects. The participants in this study were of 3 cohorts: children with child on-set SCI, adults with child onset SCI, and adults with adults on-set SCI. Participants performed maximum isometric voluntary contractions (MVIC) while seated in a Biodex System 3 or BTE PrimusRS to measure maximum joint moments of five different shoulder joint postures and two elbow joint postures. Linear regression equations derived from nonimpaired individuals were used to predict segment length and total muscle volume from the individual’s height and body weight, respectively. The length scale factor was computed as the ratio of the predicted forearm length over the generic model’s forearm length; the length scale factor was applied to scale a generic upper limb model in OpenSim (version 4.3) with regard to limb length. The total shoulder muscle volume derived from the linear regressions were portioned into the upper limb muscle groups following the proportions of the generic MSK model. Maximum isometric force (MIF) was calculated using individual muscle volumes, specific tension, and optimal fiber length (OFL) from length-scaled models. The muscle strength scale factor was computed as the average ratios of calculated MIFs over the related MIFs from the model; the muscle strength scale factor was then applied to all the muscles in the model. Muscle driven simulations were conducted with three different sets of muscle activation: “primary”, “separated”, and “EMG”. Primary muscle activation controls consisted of 100% activation of all agonist muscles for each joint posture. Separated muscle activation controls were defined according to the description of primary, secondary, and minor muscle contributors during MVIC defined as 100%, 50%, and 10% activation, respectively. EMG muscle activation controls were based on the processed EMG data that were collected during the MVIC collection.
Linear mixed effect models (LMEMs) were used to compare the prediction of MVIC based on muscle activation input type (i.e., primary, separated, EMG). A type III analysis of variation (ANOVA) and intraclass correlation (ICC) were conducted on the LMEMs in order to understand the contributions of the fixed variables (e.g., sex, age, joint posture, body mass, and height) in MVIC production compared to the predicted MVIC and how much the total variation is contributed by the random effect of the participants. Pearson’s correlation and linear regression were also conducted on the predicted and measured MVICs separated by muscle activation input type and the seven different joint postures to understand the interactions between the predicted MVIC and the joint postures.
The results from the LMEMs showed that the EMG based muscle activation of the scaled MSK model was able to better predict the experimental MVIC with an estimated coefficient of 0.46, relative to the literature derived activation. Body mass, height, and age had estimated coefficient magnitudes closer to zero (-0.5 < 0< 0.5), while sex and posture were larger than one. From the type III ANOVA for the EMG informed LMEM, sex and posture had significant p-values at < 0.05 and <0.001, respectively. The significant p-values suggested that sex and posture were not well accounted for in the prediction of MVIC from this study. The ICC and marginal R² resulting from the EMG informed LMEM were 0.56 and 0.42, respectively, indicating that the intrinsic variability of the participant was greater than the identified fixed variables (ICC > Marginal R²). Overall, the strength scaling method using linear regression based on nonimpaired data did not predict the strength of individuals with paraplegia well. Thus, it is recommended adjusting model strength based on measured MVICs for individuals with paraplegia if MRI data is not available.