Optimization-based forward musculoskeletal simulation of pathological gait – OMEGA

Each clinical situation consisted in two conditions: the initial condition and the altered condition. The gait analysis data was obtained with a motion capture system and ground reaction forces were measured. The effect of the scaling method on the gait conditions was analyzed, and inverse dynamics simulation for the knee brace and drop foot pathology was performed and several gait parameters regarding the kinematic and kinetic were calculated. Two forward simulators were developed. A feedback-oriented simulator was designed with specific components to follow a reference movement, to maintain balance, and to control the heading which parameters were partly found through an optimization process. A second simulator based on neural networks and optimization processes was also designed. A thorough analysis of the properties of this simulator was done, regarding its precision, robustness and sensitivity and was used to determine the most optimal settings for prediction. In order to predict gaits, our approach was to apply deformations to the reference motion to obtain novel and stable simulations without retraining the neural networks. Predictions of altered conditions for the knee-brace situation were compared to the clinical data.

The inverse dynamics analysis of the clinical data allowed to characterize the gait patterns and addressed the main differences between the initial and altered gait conditions for each targeted pathology. The most promising predictive forward simulator designed in this project is based on neural networks and is capable to replicate gaits for a patient in an initial condition and in an altered condition. The simulator is also capable to predict gait patterns through the application of deformations to the initial motion which are directed by an optimization process. These results will contribute to improve individualized treatment of patients especially in cases with very low incidence or high individuality, for which evidence-based medicine has natural limits of applicability.
The clinical data collection and analysis of the drop foot pathology will be presented at ISB in July 2021 (Santos et al., 2021). The sensitivity analysis of the predictive forward simulator based on neural networks has been introduced during a national workshop (Bonis et al., 2018) and fully presented later at ICNR (Bonis et al., 2020). More scientific productions are expected in 2021. As this project is categorized as basic research, economical exploitable results were not directly expected. Nevertheless, a functional simulator was developed and tested on one clinical condition.

The effect of different parameters and scaling strategies on the clinical situations studied during the project was investigated. The inverse dynamics approach allowed us to characterize the gait patterns and address the main differences between the initial and altered gait conditions.
Several forward simulators have been proposed. Our neural network-oriented FMS is capable to replicate gaits similar to the reference motion for a patient in an initial condition and in an altered condition. The FMS is also capable to predict gait patterns through the application of deformations to an initial reference motion.
Our different studies have shown the importance of using an environment adapted to a neural network training and to optimization processes, and the computation times they require. Our analysis of the FMS determined how to reduce the search space during prediction. It also allowed to control the effect of the prediction on the precision, accuracy and robustness of the FMS.

The clinical data collection and analysis of the drop foot pathology will be presented at ISB in July 2021 (Santos et al., 2021). The sensitivity analysis of the predictive forward simulator based on neural networks has been introduced during a national workshop (Bonis et al., 2018) and fully presented later at ICNR (Bonis et al., 2020). More scientific productions are expected in 2021. As this project is categorized as basic research, economical exploitable results were not directly expected. Nevertheless, a functional simulator was developed and tested on one clinical condition.

Knowledge of the motion and loading of the human body and its parts is essential in many fields of human medicine and in particular in orthopedic and trauma surgery. Modern numerical methods have advanced to the point that they are useful for estimating forces in the relevant structures based on a priori knowledge motion and external loading. However, their general usefulness is limited by their inability to predict motion and loading in response to changes in the body caused by disease or as a result of applying a given therapy to treat a disease. The aim of this project is to develop a new generation of forward simulator, based on the computation of most optimal motions, with the promise of predicting and thus improving therapeutic strategies for individual patients. Such simulator will help moving medicine away from a purely empirical approach which relies on the evolution of therapies based on evidence alone, to a point where guiding and predicting therapies is possible.

Powerful recently developed methods of forward i.e. predictive simulation, which have to date been developed primarily for the representation of autonomous human-like motion in the entertainment and gaming industry, will be applied in a medical context. An innovative forward musculoskeletal simulator (FMS) will be developed and implemented for this purpose. Although the simulator could in the future be applied to a wide range of musculoskeletal disorders it will first be tested and further evolved in three specially selected clinical situations: Knee and Ankle Bracing, Drop Foot Pathology, and Above Knee Amputation treated with a microprocessor-controlled knee-prosthesis. These three clinical situations were chosen because they are well characterized, treatments modalities can be altered non-invasively without undue risk to the patients, and sufficient numbers of patients are available. Although relatively simple, each of these pathologies represents a serious reduction in quality of life for the affected patients and general improvements in treatment approaches and in particular the adaptation of treatment modalities to the needs of individual patients will have a serious impact on quality of life but also on the economic costs associated with treatments. In all three cases a framework will be developed in the context of a clinical gait analysis laboratory, validated and further developed by using a combination of direct kinematic and kinetic measurements and inverse dynamic simulations as ground-truth data obtained from respective patient collectives.

One of the most obvious areas of future application of the predictive capabilities of FMS is the treatment of advanced OA of the hip and knee, i.e. total hip (THA) and total knee arthroplasty (TKA). The cost of these procedures represents a significant and increasing economic challenge. Thus, the OECD reports that the growing volume of hip and knee replacement will continue to contribute to health expenditure growth as these are expensive interventions. This is a further indication of the significance of the basic-research to be performed in this project and demonstrates a major medical and economic impact: improving the treatment of, and providing powerful new methods for predicting the outcome of treatments in many contexts related to musculoskeletal diseases.