Mid-Activity and At-Home Wearable Bioimpedance Elucidates an Interpretable Digital Biomarker of Muscle Fatigue

Abstract

Objective: Muscle health and decreased muscle performance (fatigue) quantification has proven to be an invaluable tool for both athletic performance assessment and injury prevention. However, existing methods estimating muscle fatigue are infeasible for everyday use. Wearable technologies are feasible for everyday use and can enable discovery of digital biomarkers of muscle fatigue. Unfortunately, the current state-of-the-art wearable systems for muscle fatigue tracking suffer from either low specificity or poor usability.

Methods: We propose using dual-frequency bioimpedance analysis (DFBIA) to non-invasively assess intramuscular fluid dynamics and thereby muscle fatigue. A wearable DFBIA system was developed to measure leg muscle fatigue of 11 individuals during a 13-day protocol consisting of exercise and unsupervised at-home portions.

Results: We derived a digital biomarker of muscle fatigue, fatigue score, from the DFBIA signals that was able to estimate the percent reduction in muscle force during exercise with repeated-measures Pearson's r = 0.90 and mean absolute error (MAE) of 3.6%. This fatigue score also estimated delayed onset muscle soreness with repeated-measures Pearson's r = 0.83 and MAE = 0.83. Using at-home data, DFBIA was strongly associated with absolute muscle force of participants (n = 198, p < 0.001).

Conclusion: These results demonstrate the utility of wearable DFBIA for non-invasively estimating muscle force and pain through the changes in intramuscular fluid dynamics.

Significance: The presented approach may inform development of future wearable systems for quantifying muscle health and provide a novel framework for athletic performance optimization and injury prevention.

Fig. 1.

The scientific hypotheses put forth in this study. During activities, muscle force of an individual may decrease because of muscle fatigue. This decrease in muscle force could be tracked by wearable DFBIA, and the DFBIA measures could be used to advise individuals when to stop before overexercising. This muscle force loss could induce local muscle damage by means of muscle fiber tears, for example. This damage would result in DOMS, swelling and absolute muscle force loss, which could prevent individuals return exercising or rehabilitation. Mid-activity changes in muscle DFBIA may estimate this DOMS and excessive DOMS may be avoided by having the human in the loop or utilizing an exoskleton assistance. Lastly, tracking the baseline muscle force is necessary. By having human-in-the-loop, individuals may be advised to return when they recover their muscle force for maximized human performance and injury prevention. At-home DFBIA may address this need of absolute muscle force tracking, which would help with human performance augmentation and injury prevention (* means that the drawn characteristics were reproduced using the results presented in [32, 48]).

Fig. 2.

Wearable DFBIA system and experimental protocol to test the hypotheses put forth in this study. A wearable multimodal DFBIA system was designed and to measure leg DFBIA and knee kinematics in dynamic settings. To test the hypotheses on leg DFBIA tracking muscle pain and recovery, a 13-day protocol was designed including two in-lab fatigue protocols. On nine days during this protocol, participants measured their leg DFBIA at home using the custom wearable DFBIA system, and their muscle force was measured using a dynamometer. They also rated their leg pain using visual analog scale. During fatigue protocol, participants performed 15 sets of 20 split leg squats with two-minute walking at 1.1 m/s in between squat sets. Leg DFBIA of participants were measured using two wearable DFBIA systems on both legs during this protocol.

Fig. 3.

Details of the fatigue protocol and representative raw data from a subject. The leg fatiguing protocol consists of 15 sets of 20 split leg squats where participants performed the exercise on their dominant legs. Between each set of split leg squats, participants walked on a treadmill at 1.1 m/s for two-minutes. The exercised leg resistances at both 5kHz and 100 kHz decreased gradually towards the end of the protocol, while these resistances in control leg did not change noticeably. In addition to this slow change in the average resistance value, mid-activity DFBIA followed a periodic waveform during gait with a distinct shape. The wearable DFBIA system captured high quality data both at the beginning and the end of the protocol. To quantify the slow change in 5kHz and 100kHz leg resistances, we calculated their ratio for each gait cycle, R_5kHz/R_100kHz.

Fig. 4.

Derivation of fatigue score and its utility in estimation of percent reduction in muscle force mid-activity. (A) Percent reduction in exercised leg muscle force was significantly larger than that of control leg for both weeks. Also, the percent reduction in exercised leg muscle force in the first week was significantly larger than that of second week. (B) Exercised leg fatigue score was significantly larger than that of control leg for both weeks. Also, exercised leg fatigue score in the first week was significantly larger than that of second week. (C) Fatigue score is defined as the reduction in R_5kHz/R_100kHz with respect to the beginning of the protocol. A LOPO-CV method was used to train and test a simple linear regression model to estimate percent reduction in muscle force. (D) Fatigue score estimated the percent reduction in muscle force with repeated-measures Pearson’s r of 0.90 and a RMSE of 3.6% at a population level. (* denotes p<0.05).

Fig. 5.

The effect of DOMS is observed both in absolute muscle force and at-home DFBIA. The R_5kHz/R_100kHz measured at home has a strong linear effect with muscle force. (A) Muscle force measured in the lab and (B) R_5kHz/R_100kHz measured at home decreased significantly two days after the fatigue protocol on both weeks for 11 participants. (C) R_5kHz/R_100kHz had a strong effect on absolute muscle force with effect size β=646 N, p<0.001 at a population level (198 data points). (* denotes p<0.05)

Fig. 6.

Fatigue score estimates delayed pain quantified bay VAS score better than human perception of exertion during exercise. (A) End of exercise Borg RPE estimates the delayed VAS score poorly with repeated-measures Pearson’s r = - 0.60, MAE = 1.71, and (B) Bland-Altman plot for this estimation shows a 95% adjusted LOA at 4.75. (C) Fatigue score estimates the delayed VAS score successfully with repeated-measures Pearson’s r = 0.83, MAE = 0.83, and (D) Bland-Altman plot for this estimation shows a 95% adjusted LOA at 2.01.