Page 110 - Read Online

Page 18 of 26 Chang et al. Soft Sci. 2026, 6, 29

coordinated analysis of metabolism-related biochemical markers (i.e., sweat glucose and lactate) and sEMG
also aids partial differentiation of fatigue levels [149] . These findings highlight that the reliability of wearable
fatigue detection can be solidified by integrated evaluations of neuromuscular activation, mechanical
behavior, cardiovascular regulation, and metabolite accumulation. In this context, multimodal data fusion
combined with AI analysis extends fatigue assessment beyond state recognition toward decision support.

The increasingly reliable fatigue evaluation would also benefit quantitative training load regulation and
recovery management. Since muscle fatigue is regarded as the cumulative physiological stress responding to
external training stimuli, lower estimation error of training load can be realized with refined evaluation of
fatigue level, leading to more physiologically grounded adjustments of training intensity and volume [148-150] .
Furthermore, AI-based analysis of multimodal data, including biomechanical (muscle stiffness and strength)
and biochemical (lactic acid, urea and IL-6) indexes, can support objective identification of post-exercise
recovery windows. It can also assist in formulating personalized nutritional strategies, such as optimizing the
timing and quantity of carbohydrate and protein intake to facilitate energy replenishment and metabolic
repair [Figure 2D].
[151]
It is worth emphasizing that such performance improvements rely on the capability of AI algorithms to
model the complex relationships embedded in multivariate time-series data. In current studies on wearable
multimodal sensing, commonly used algorithms include SVM, TCN, bidirectional long short-term memory
(Bi-LSTM), and decision-level fusion methods based on Dempster-Shafer (D-S) evidence theory. SVM
demonstrates strong generalization performance in small-sample classification scenarios [152-154] . TCN can
effectively capture long-term temporal dependencies through dilated convolutions and have outperformed
conventional SVM in multimodal fusion models. Bi-LSTM is well suited for handling temporal dependency
features and signal drift. Meanwhile, decision-level fusion based on D-S evidence theory further integrates
probabilistic information from multiple sources, thereby reducing the influence of inter-individual
variability .
[104]

CONCLUSION AND OUTLOOK
This article systematically elaborates the underlying physiological mechanisms and corresponding
bioelectrical, biochemical and biomechanical indicators of muscle fatigue, injury and healing. Based on these,
the design and working principle of wearable sensors, as well as their applications in real-world settings for
competitive sports and public fitness, including load monitoring, fatigue evaluation and personalized
training and nutrition management, are comprehensively introduced.

In spite of the progress made so far, the development of wearable sports monitoring still faces many technical
bottlenecks and application challenges. Future research is anticipated to focus on the following aspects
[Figure 3]. (1) Technology optimization. Technical breakthroughs overcoming existing bottlenecks such as
the stable acquisition of bioelectrical signals (sEMG/HD-EMG) in dynamic environments , the long-term
[155]
monitoring of traced biomarkers (such as CK, IL-6) , and the precise estimation of biomechanical
[17]
parameters (such as ground reaction force) based on IMU and elastography, are demanded. For instance,
muscle healing represents a longer-term physiological process that may extend from days to weeks, whereas
many current wearable systems are predominantly applied in acute or session-based monitoring scenarios.
Continuous or longitudinal tracking of recovery trajectories presents additional challenges, including
long-term signal stability, sensor adhesion and biocompatibility, power management, data consistency, and
individualized baseline recalibration. Future development of energy-efficient flexible electronics, cloud-based
analytics, and personalized modeling strategies will be essential for enabling reliable long-term wearable
monitoring of muscle healing. (2) Data fusion. So far, no universally accepted thresholds currently exist that
allow wearable signals alone to distinguish muscle fatigue, injury and recovery with diagnostic certainty.