Page 98 - Read Online

Page 8 of 26 Chang et al. Soft Sci. 2026, 6, 29

Table 1. Multimodal physiological characteristics and temporal features of muscle fatigue, injury and recovery
Muscle Primary biological nature Main changing indicators Temporal scale Reversibility
health state
MF ↓, RMS ↑, lactic acid ↑,
Functional decline (without Short-term (minutes to
Fatigue muscle stiffness ↑, muscle Rapid recovery (after rest)
structural disruption) hours)
strength ↓
Structural disruption (myofibers CK ↑, IL-6 ↑, muscle stiffness Intermediate to long-term Limited or delayed recovery
Injury
and extracellular matrix) ↑, muscle strength ↓ (days to weeks) (without intervention)
Structural remodeling and Intermediate to long-term Progressive restoration
Recovery Toward baseline
functional restoration (days to weeks) toward baseline
MF: Median frequency; RMS: root mean square; CK: creatine kinase; IL-6: Interleukin-6.

recognize that individual physiological markers rarely provide definitive diagnostic information in isolation.
Instead, wearable sensing technologies should be applied within a multimodal framework that integrates
neuromuscular activity, biomechanical performance, metabolic stress, and recovery trajectories over time. In
this context, wearable systems are particularly valuable for identifying functional deviations from
individualized baselines, monitoring dynamic transitions across fatigue, injury risk, and recovery phases, and
supporting longitudinal assessment in real-world athletic and daily exercise environments.

Wearable bioelectrical sensors
sEMG enables recording of the electrical activity of skeletal muscles driven under neural drive. Owing to its
non-invasive nature and wide applicability, it has emerged as a key technology for assessing exercise-related
muscle fatigue [34,90] . Implementation of wearable technology: Typically, fabric electrodes or array electrodes
are adopted, with skin-electrode contact optimized via conductive fibers and gels. Combined with low-noise
amplification, anti-power frequency and motion pseudo-aberration circuits, the signal front-end
conditioning is achieved. Feature extraction and pattern recognition in the time-frequency domain are
enabled through embedded or edge computing [84,91] . High-density sleeve arrays further break through the
limitations of complex muscle group monitoring, enabling motion recognition and continuous joint angle
estimation in areas such as the hand, demonstrating feasibility and low latency advantages in
human-computer interaction and rehabilitation scenarios [92,93] .Compared to traditional single-electrode
arrangements, arrayed wearables mitigate performance fluctuations caused by electrode placement
dependence and drift . However, sweat interference, conductive media drying, and fabric slippage during
[94]
dynamic movements still affect signal-to-noise ratio and reproducibility, necessitating resolution through
integrated structural design and domain adaptive algorithms. Currently, sEMG has been widely applied in
training and rehabilitation fields, including muscle group recruitment pattern recognition,
posture/movement phase annotation, and intent decoding interfaces coordinated with external
biomechanical quantities, providing technical support for individualized training prompts and rehabilitation
assistive device control [95-97] .

ECG is primarily used to evaluate cardiac rhythm and conduction function [98-100] . By analyzing the adjacent
R-R intervals, indicators of heart rate (HR) and heart rate variability (HRV) can be derived; both serve as
quantitative parameters for characterizing overall cardiopulmonary load and indirectly reflecting the impact
of exercise on the body’s overall physiological state [101] . During exercise, ECG signals are prone to
interference from exercise artifacts, leading to reduced accuracy of the indicators [102,103] . Meanwhile, overall
physiological indicators like HR and HRV cannot accurately reflect the local fatigue or repair status of
specific muscle groups. Therefore, they are more suitable for joint interpretation with local physiological
signals such as sEMG. Through multi-dimensional data fusion of “overall + local”, the accuracy of muscle
function assessment can be improved .
[104]