A systematic review of approaches to modelling lower limb muscle forces during gait: Applicability to clinical gait analyses

Abstract

Computational methods to estimate muscle forces during walking are becoming more common in biomechanical research but not yet in clinical gait analysis. This systematic review aims to identify the current state-of-the-art, examine the differences between approaches, and consider applicability of the current approaches in clinical gait analysis. A systematic database search identified studies including estimated muscle force profiles of the lower limb during healthy walking. These were rated for quality and the muscle force profiles digitised for comparison. From 13.449 identified studies, 22 were finally included which used four modelling approaches: static optimisation, enhanced static optimisation, forward dynamics and EMG-driven. These used a range of different musculoskeletal models, muscle-tendon characteristics and cost functions. There is visually broad agreement between and within approaches about when muscles are active throughout the gait cycle. There remain, considerable differences (CV 7%-151%, range of timing of peak forces in gait cycle 1%-31%) in patterns and magnitudes of force between and within modelling approaches. The main source of this variability is not clear. Different musculoskeletal models, experimental protocols, and modelling approaches will clearly have an effect as will the variability of joint kinetics between healthy individuals. Limited validation of modelling approaches, particularly at the level of individual participants, makes it difficult to conclude if any of the approaches give consistently better estimates than others. While muscle force modelling has clear potential to enhance clinical gait analyses future research is needed to improve validation, accuracy and feasibility of implementation in clinical practice.

Can a reduction approach predict reliable joint contact and musculo-tendon forces?

“…The comparison is more problematic for the quadriceps because more complex musculoskeletal models differentiate between vastii and rectus femoris. Still, when superimposing the two forces synthetized by Trinler et al (2018), the pattern and amplitude of the quadriceps musculo-tendon forces obtained with the reduction approach seem in agreement with the ones reported in the literature.…”

“…(Winby et al, 2009). Moreover, the patterns and amplitudes of the musculo-tendon forces, especially for the gastrocnemii and hamstrings groups, match the ones reported in the literature with generic musculoskeletal models (Trinler et al, 2018). The comparison is more problematic for the quadriceps because more complex musculoskeletal models differentiate between vastii and rectus femoris.…”

Estimating Biomechanical Time-Series with Wearable Sensors: A Systematic Review of Machine Learning Techniques

“…Faisal et al (2019) recently provided a high-level overview of sensing technologies, applications of wearables in monitoring joint health, and analysis techniques [38]. Several reviews are available concerning the use of Hill-type muscle models for sEMG informed muscle force estimation which can be used to estimate kinematics via forward-dynamics [26,27,32,39]. Dowling (1997) mentions the potential use of neural networks in this context but does not review any relevant literature.…”

Estimation of muscle activation during different walking speeds with two mathematical approaches compared to surface EMG

“…Knowing the activation and force profiles of individual muscles during various movements can help to gain further insight into the causes of impairments and inform targeted treatments. Yet, a range of muscle activation and force modelling approaches have been applied in research studies of sport performance or clinical interventions [3][4][5][6][7] but use of modelling has not yet become established as a part of routine clinical gait analysis [8].…”

“…One reason muscle force modelling has not yet been used to its full advantage in clinical practice is that there is a huge variety of approaches producing different results [8]. The influence of different modelling assumptions and approaches on model outputs are still not fully understood.…”