Page 109 - Read Online

Chang et al. Soft Sci. 2026, 6, 29 Page 17 of 26

Hydration status of the body and electrolyte management
Sweat is an important medium for the excretion of metabolites and electrolytes during exercise. The dynamic
changes in its components (such as lactic acid, glucose, electrolytes, urea, etc.) , secretion rate and total
[40]
amount are associated with the metabolic load, hydration status and physiological stress level of athletes.
Wearable sweat sensing technology integrates multiple types of sensors to achieve real-time and non-invasive
monitoring of multiple biochemical indicators (like glucose, lactic acid, Na and K ). It accurately captures
+
+
the physiological responses under different hydration states, as well as the dynamic change patterns of
Na /K concentrations under hydrated and non-hydrated conditions. The technology enables accurate
+
+
assessment of the dehydration risk and electrolyte loss characteristics for professional athletes, and therefore
aids the formulation of targeted fluid replacement volume, timing and electrolyte supplementation plans to
effectively reduce the risk of heat-related diseases such as muscle spasms and heat exhaustion . The
[13]
battery-free, short-range wireless communication microfluidic patch further expands the monitoring modes
via integrating electrochemical detection (lactic acid, glucose), colorimetric analysis (pH, Cl ), and volumetric
-
measurement (sweat rate, total sweat volume) within a single device. Long-term tests reveal an explainable
time-lag correlation between its sweat-based detection results and corresponding marker levels in blood.
Employing the patch may offer objective clues for reasonably adjusting the interval time, training intensity
and recovery interval, by recording the lactate-power curve and tracking the metabolic stress distribution
and recovery process of interval training [143,144] [Figure 2C].

Regulation of the microenvironment for tissue repair
The quality of tissue repair after local muscle-fascia injury directly affects the recovery process and re-injury
risk of athletes. Traditional repair strategies mostly focus on mechanical rehabilitation training and lack
precise regulation of the local microenvironment of the injured tissue. The development of mechanically
active therapy provides a new route for precise tissue repair: by programming the local stress/strain state to
optimize the mechanical microenvironment of the injured region, it promotes the healing of injured
tissues [145,146] . It is anticipated to combine such active tissue mechano-modulation with wearable stiffness
sensing to establish monitoring-intervention closed-loop injury repair strategy. Specifically, the healing
status of injured muscle or tendon can be evaluated by wearable stiffness monitoring, and this in turn, based
on biomechanical and mechanobiological pro-healing mechanisms, guides the rational design of mechanical
loading parameters to accelerate healing and avoid scar formation [147,147] .

AI-enabled multimodal tracking
Unimodal fatigue monitoring technologies - particularly those based on sEMG - have long dominated the
field. However, increasing empirical evidence suggests that reliance on a single physiological signal is
insufficient for achieving stable and reliable muscle fatigue assessment. The limitations primarily include
high sensitivity to inter-individual variability, limited capability in identifying transitional states, and
susceptibility to motion artifacts during dynamic exercise.

Combined with artificial intelligence (AI) algorithms, unimodal fatigue evaluation based on wearable sEMG
sensing shows reasonable average recognition performance, however with substantial inter-subject
variability . Specifically, sEMG coupled with support vector machine (SVM) achieves peak accuracy over
[35]
90%, yet the lowest individual accuracy is merely < 65%. Compared to unimodal sEMG with temporal
convolutional network (TCN) processing, a dual-modal TCN model integrating ECG and sEMG exhibits
significantly improved fatigue classification performance, with an overall average accuracy of 88.5% and an
inter-personal variation range of ~ 10% (> 20% for unimodal model). In another study, a multimodal
framework integrating muscle thickness, joint angle, and sEMG is developed to reduce the false positive rate
down to 3%, thereby enhancing the functional robustness under dynamic conditions and improving
reliability in real-world training scenarios . Expanding multimodal frameworks to biochemical sensing, the
[148]