Abstract:Shoulder muscle forces estimated via modelling are typically indirectly validated against measurements of glenohumeral joint reaction forces (GHJ-RF). This validation study benchmarks the outcomes of several muscle recruitment strategies against public GHJ-RF measurements. Public kinematics, electromyography, and GHJ-RF data from a selected male participant executing a 2.4 kg weight shoulder abduction task up to 92° GHJ elevation were obtained. The Delft Shoulder and Elbow Model was scaled to the participant. Muscle recruitment was solved by 1) minimising muscle activations squared (SO), 2) accounting for dynamic muscle properties (CMC) and 3) constraining muscle excitations to corresponding surface electromyography measurements (CEINMS). Moreover, the spectrum of admissible GHJ-RF in the model was determined via Markov-chain Monte Carlo stochastic sampling. The experimental GHJ-RF was compared to the resultant GHJ-RF of the different muscle recruitment strategies as well as the admissible stochastic range. From 21 to 40 degrees of humeral elevation, the experimental measurement of the GHJ-RF was outside the admissible range of the model (21 to 659% of body weight (%BW)). Joint force RMSE was between 21 (SO) and 24%BW (CEINMS). At high elevation angles, CMC (11%BW) and CEINMS (14%BW) performed better than SO (25%BW). A guide has been proposed to best select muscle recruitment strategies. At high elevation angles, CMC and CEINMS were the two most accurate methods in terms of predicted GHJ-RF. SO performed best at low elevation angles. In addition, stochastic muscle sampling highlighted the lack of consistency between the model and experimental data at low elevation angles.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the verification and benchmarking in the modeling of shoulder muscle force estimation. Specifically, the research aims to evaluate the accuracy of several state - of - the - art muscle recruitment strategies (such as static optimization (SO), computational muscle control (CMC), and electromyography - assisted (CEINMS) methods) in predicting joint reaction forces (GHJ - RF) in shoulder modeling by comparing them. The study used publicly available shoulder kinematics, electromyography (EMG), and glenohumeral joint contact force data, which were measured from a male subject with an instrumented shoulder prosthesis during the performance of a shoulder abduction task.
### Research Background
Shoulder musculoskeletal (MSK) modeling and inverse dynamics (ID) are often used to estimate muscle and joint contact forces during movement, because these forces are usually difficult to measure directly in vivo. Estimating muscle forces through modeling can provide crucial knowledge of internal dynamics, which is helpful for understanding the physiological load in daily living activities, optimizing sports performance, and designing surgical and rehabilitation techniques. However, due to the fact that the number of muscles in most MSK systems (including the shoulder) is greater than the degrees of freedom, the muscle redundancy problem makes the prediction of individual muscle forces complicated.
### Research Methods
1. **Experimental Data**: The study used publicly available kinematics, surface electromyography (EMG), and glenohumeral joint contact force data of a 64 - year - old male. These data were synchronously collected when the subject performed shoulder abduction (up to 95° glenohumeral joint elevation) and shoulder flexion (up to 110° glenohumeral joint elevation) tasks, and also included a shoulder abduction task with a 2.4 - kg weight held in the hand (up to 92° glenohumeral joint elevation).
2. **Shoulder Modeling**: An adjusted Delft Shoulder Elbow Model (DSEM) was used, which contains 7 segments (sternum, clavicle, scapula, humerus, ulna, radius, hand), 11 degrees of freedom, and 18 muscle - tendon units.
3. **Muscle Recruitment Modeling Strategies**:
- **Static Optimization (SO)**: Minimize the sum of squared muscle activations.
- **Computational Muscle Control (CMC)**: Consider the time - dependent properties of muscles.
- **Electromyography - Assisted (CEINMS)**: Use electromyography signals to calibrate model parameters and minimize the difference between muscle excitation and experimental electromyography recordings through a forward - dynamics feedback loop.
4. **Random Sampling**: Use the Markov Chain Monte Carlo (MCMC) method to randomly sample muscle forces to determine the solution space of joint reaction forces consistent with movement.
### Research Results
- **Total Glenohumeral Joint Reaction Force (GHJ - RF)**: The randomly sampled solution space ranges from 21% to 659% of body weight (BW). The experimentally measured GHJ - RF is only within the acceptable range of the model when the shoulder is elevated above 40°.
- **Performance of Different Methods**:
- **Low Elevation Angles (21° - 40°)**: The experimentally measured GHJ - RF is not within the acceptable range of the model.
- **High Elevation Angles (above 75°)**: CMC and CEINMS (single - calibration and double - calibration) perform better than SO.
- **Overall Performance**: SO performs best at low elevation angles, while CMC and CEINMS perform better at high elevation angles.
- **Error Analysis**: The absolute error and root - mean - square error (RMSE) between the GHJ - RF predicted by different methods and the experimentally measured values were calculated. The results show that CMC and CEINMS have lower errors at high elevation angles.
### Discussion
The study applied a random modeling method in shoulder modeling for the first time, which can comprehensively explore the muscle synergy patterns in the model. The range of the random solution space provides information on the ability of the model to match experimental data, thus supplementing the verification process. The study found that the experimentally measured GHJ - RF is inconsistent with the model at low elevation angles (20° - 40°), which may be due to the relatively loose joint contact under experimental conditions or the fact that the rigid - body characteristics assumed by the model are not applicable in some cases. In addition, the noise and incompleteness of the data also increase the difficulty of processing.
### Conclusion
This study provides benchmarks and verification for shoulder modeling by comparing different muscle recruitment strategies. The research results show that CMC and CEINMS perform better at high elevation angles, while SO performs better at low elevation angles. The random modeling method provides a new perspective for understanding shoulder muscle synergy patterns, and future research can further explore the application of this method in shoulder modeling.