Towards the application of multi-DOF EMG-driven neuromusculoskeletal modeling in clinical practice: methodological aspects

New methods able to assess the individual ability of patients to generate motion and adaptation strategies are increasingly required for clinical applications aiming at recovering motor functions. Indeed, more effective rehabilitation treatments are designed to be personalized on the subject capabilities. In this context, neuromusculoskeletal (NMS) models represent a valuable tool, as they can provide important information about the unique anatomical, neurological, and functional characteristics of different subjects, through the computation of human internal variables, such as muscle activations, muscle forces, joint contact forces and moments. A first possible approach is to estimate these values using optimization-based NMS models. However, these models require to make assumptions on how the muscles contribute to the observed movement. More promising are instead NMS models driven by electromyographic signals (EMG), which use experimentally recorded signals that can be considered a direct representation of the subject motor intentions. This allows to account for the actual differences in an individual neuromuscular control system, without making any preliminary assumptions. Therefore these models have the potentialities to provide the level of personalization that is essential for applications in the clinical field. Although EMG-driven NMS models have been investigated in the literature, even for clinical purposes, they are mostly limited to one degree of freedom (DOF), and consider only the muscles spanning that DOF. Additionally, despite the promising results, they are still not introduced in the clinical practice; the main reason possibly being their complexity, that makes them not usable in clinical context, where standard and reliable procedures are required. The importance of EMG-driven NMS modeling for clinical applications would be even higher with the availability of multi-DOF models, as impairments usually compromise multiple joints. Nevertheless, even if a first multi-DOF EMG-driven NMS model for the lower limbs has been recently introduced in literature, its even greater complexity makes more difficult an analysis of its applicability in the clinical field. This work represents a first effort towards a critical analysis of multi-DOF EMG-driven NMS models to evaluate their possible use in clinical practice. To achieve this objective, several issues and limitations have been addressed. In the specific, the attention has been focused on two aspects: (i) making the methodology usable, to foster its adoption by multiple laboratories and research groups, and to facilitate sensitivity analyses required to assess its accuracy; (ii) highlighting the effects of some methodological aspects related to data acquisition and processing, and evaluating their impact on the accuracy of estimated parameters and muscle forces. This analysis is even more important for multi-DOF EMG-driven NMS model as it is still not present in the literature. To accomplish the first goal, a software tool (MOtoNMS) has been developed and it is freely available for the research community. It is a complete, flexible, and user-friendly tool that allows to automatically process experimental motion data from different laboratories in a transparent and repeatable way, for their subsequent use with neuromusculoskeletal modeling software. MOtoNMS generalizes data processing methods across laboratories, and simplifies and speeds up the demanding data elaboration workflow. This simplification represents an indispensable step towards an actual translation of NMS methods in clinical practice. The second part of the work has been, instead, dedicated to analyze the impact on model parameters and muscle forces prediction of different techniques for EMG data collection and processing that are feasible for clinical settings, in particular concentrating on EMGs normalization. Indeed, moving EMG-driven NMS modeling towards clinical applications that deal with multiple DOFs requires to carefully consider subject's motor limitations due to his/her mobility impairments. This results in a rethinking about the methodologies for data acquisition and processing. Therefore, the impact of using only data from walking trials on both calibration of model parameters and computing the maximum EMG values needed for the normalization step, has been assessed with two case studies. Moreover, a protocol for the collection of maximum voluntary contractions has been proposed. This protocol is suitable for multiple DOFs applications involving patients with reduced motor ability and it requires only low-cost and easy to acquire tools to make it applicable in any laboratory. The research proposed in this thesis provides tools to simplify the use of multi-DOF EMG-driven neuromusculoskeletal models and proposes analyses and procedures to evaluate the accuracy and reliability of the obtained results with the aim of pursuing clinical applications.