Poster Q6 - From Generic to Personalized: Examining How Models with Experimentally-Derived Muscle Parameters Influence Force Predictions in the Thumb

Introduction: Musculoskeletal models are physics-based representations of the musculoskeletal system that aid in understanding how muscles, bones, and joints work together to produce movement. For musculoskeletal models to be useful for analyzing biomechanical behavior, they need to be valid and accurate. Yet, generic models are often used in biomechanical simulations, particularly in the upper limb. These generic models represent an average adult using generalized parameters derived from cadaveric or small in vivo studies. This is problematic because the average does not represent the individual. Humans differ in size, mass, strength, and activity level. As a result, muscle parameters vary. For example, a physically active individual may have larger muscles and, hence, a larger physiological cross-sectional area (PCSA) than a less active individual, as PCSA is positively correlated with maximum isometric force.
Despite recent work advocating for personalized models, the cost of volumetric imaging and the analysis time required for measuring muscle parameters remain barriers to widespread usage. Extended field of view ultrasound (EFOV-US) imaging provides an inexpensive and reliable technique for non-invasively measuring muscle parameters. Specifically, this technique enables recording of panoramic images that capture a full muscle belly via one continuous scan.
The aim of this study is to understand to what extent incorporating personalized muscle parameters of the thumb affects biomechanical analysis. To achieve this aim, EFOV-US images of thumb muscles were acquired, muscle parameters derived from these images were used to create personalized models, and differences between biomechanical simulations using generic versus personalized models were assessed.

Materials and

Methods: This study is a preliminary sub-analysis of an ongoing IRB-approved study. Five healthy adults (3 males and 2 females, 22.8 ± 0.4 years, 81.64 ± 12.49 kg, 177.1 ± 6.71 cm) were randomly selected for analysis. In each participant, six thumb muscles [abductor pollicis brevis (APB) and longus (APL), extensor pollicis brevis (EPB) and longus (EPL), flexor pollicis brevis (FPB) and longus (FPL)] were imaged using EFOV-US. Five images were acquired for each muscle. Images were analyzed in ImageJ by manually measuring three parameters: fascicle length, pennation angle, and PCSA. Briefly, fascicle lengths were measured by tracing the longest distance from proximal to distal aponeurosis. Pennation angle was measured as the angle between the fascicle and aponeurosis. PCSA was approximated by the largest height of the muscle. Muscle parameters were averaged across all images for each participant.
To evaluate the influence of personalized muscle parameters, a thumb abduction-adduction motion was simulated in OpenSim (v4.4, SimTk) using both the generic and five personalized models. The generic model was the ARMS Wrist and Hand model. For each participant, a personalized model was defined by anthropometrically scaling the generic model and then modifying the muscle parameters to match those measured. For each simulation, the input motion was identical, only the underlying model changed. The simulation output was estimated muscle force calculated using computed muscle control. Individual and mean muscle forces across the five personalized models were compared to those calculated from the generic model.

Results, Conclusions, and Discussions: The personalized models consistently generated higher forces than the generic model for all muscles (Fig. 1). Notably, with the exception of APL, all examined muscles generated at least twice the force in the personalized models compared to the generic model. FPB and EPB had the most pronounced difference. Given that all inputs, except muscle parameters, were held constant across simulations, the differences in muscle forces are due to differences in the muscle parameters in the generic versus personalized models. Many of the muscle parameters in the generic model are derived from cadaveric specimens, which are not in an active state and therefore likely have different structural properties than the relaxed, but active in vivo muscles measured in this study.
Interestingly, when looking across all participants and muscles, the personalized models showed considerable variability in muscle forces. This is reflected in the large standard deviations. Additionally, across muscles within the same participant, muscle parameter differences between the personalized and generic models were greater for some muscles than others, as in the cases of FPB and EPB (data not shown). This highlights that muscles within an individual may adapt differently due to their unique anatomical and physiological properties; such differences are not fully captured by anthropometrically scaling a generic model, thereby emphasizing the importance of personalizing musculoskeletal models.
Overall, our findings demonstrate that given the same simulation input, personalized models predict different muscle forces than a generic model. These differences in muscle forces are important to consider when using simulations to evaluate healthy or pathological movement. The study also confirms the usefulness of EFOV-US for acquiring muscle parameters for creating personalized models. While the results are intriguing, the small sample size limits our ability to draw conclusions on the significance and generalizability of the differences observed. Thus, ongoing work is extending this analysis to a larger sample size. Understanding how these differences in predicted muscle force affect other simulation outputs, like joint torques and external forces, is also important because this knowledge could help us incorporate personalized models in the development of tailored treatment plans for musculoskeletal diseases, such as osteoarthritis.

Acknowledgements (Optional): Funding from NIH NIAMS (R01 AR078817) is gratefully acknowledged.