Modelling intra-muscular contraction dynamics using in silico to in vivo domain translation

Abstract

Background Advances in sports medicine, rehabilitation applications and diagnostics of neuromuscular disorders are based on the analysis of skeletal muscle contractions. Recently, medical imaging techniques have transformed the study of muscle contractions, by allowing identification of individual motor units’ activity, within the whole studied muscle. However, appropriate image-based simulation models, which would assist the continued development of these new imaging methods are missing. This is mainly due to a lack of models that describe the complex interaction between tissues within a muscle and its surroundings, e.g., muscle fibres, fascia, vasculature, bone, skin, and subcutaneous fat. Herein, we propose a new approach to overcome this limitation. Methods In this work, we propose to use deep learning to model the authentic intra-muscular skeletal muscle contraction pattern using domain-to-domain translation between in silico (simulated) and in vivo (experimental) image sequences of skeletal muscle contraction dynamics. For this purpose, the 3D cycle generative adversarial network (cycleGAN) models were evaluated on several hyperparameter settings and modifications. The results show that there were large differences between the spatial features of in silico and in vivo data, and that a model could be trained to generate authentic spatio-temporal features similar to those obtained from in vivo experimental data. In addition, we used difference maps between input and output of the trained model generator to study the translated characteristics of in vivo data. Results This work provides a model to generate authentic intra-muscular skeletal muscle contraction dynamics that could be used to gain further and much needed physiological and pathological insights and assess and overcome limitations within the newly developed research field of neuromuscular imaging.

Identification of motor unit discharges from ultrasound images: Analysis of in silico and in vivo experiments

“…For example, instead of simulating the velocity field directly, one may simulate the US field using Field II [22] and then use axial phase-shift correlation methods to obtain the velocity images. Deep learning-based approaches are more appropriate to make a more authentic simulation model including non-MU activity, e.g., using unsupervised domain-to-domain translation [23].…”

“…Physiologically, the displacement field during muscle contractions can be considered a non- linear system, especially due to the connective tissue and the push-and-pull on nearby units and tissue [6,23]. However, this does not imply that we cannot detect all spikes from all motoneurons because we only need to detect the twitch response to a neural discharge, which should be one-to-one.…”

Non-Linearity in Motor Unit Velocity Twitch Dynamics: Implications for Ultrafast Ultrasound Source Separation

“…Future efforts should test whether linear models of convolutive mixtures, not currently analyzed, would provide a sufficient approximation of the underlying generation model so that linear convolutive source separation approaches would substantially surpass linear instantaneous methods in term of accuracy in discharge times identification. Once these models are fully exploited, the current study indicates that future research should be further directed towards methods which do not impose the linearity assumption in the underlying model, such as non-linear ICA [52], [53], using deep learning to decouple MU activity from that of other units and the extracellular matrix [54], [55], or taking inspiration from methods for dynamic EMG decomposition in which the MU AP shape changes over time [56], [57]. The ability to directly infer neural information from an US image series will enable a transition from US-based muscle machine interfaces [58] to more natural neural interfaces which, unlike their EMG counterpart, are not limited to superficial muscle only.…”

Estimating the neural spike train from an unfused tetanic signal of low-threshold motor units using convolutive blind source separation

“…However, one could overcome this challenge by including spatial information, which has a high resolution (< 1 mm) as the MU relates to a physical component (muscle unit) in the spatial domain. A potential solution may be expanding current sEMG decomposition algorithms [ 2 , 3 ] to include spatial dependence or sparsity in addition to the temporal deconvolution and validating it using an authentic simulation model [ 29 ]. Yet, all these approaches need to deal with the motion of non-MU-related structures that hides a large part of the movement caused by a MU in ultrasound images [ 11 ].…”