Continuous wavelet transform-based analysis of shoulder muscle fatigue during dynamic abduction: a surface...

Quantifying neuromuscular fatigue from surface electromyography (sEMG) signals during dynamic movements remains a methodological challenge due to the signals' non-stationarity. While wavelet-based approaches have demonstrated potential for analyzing muscle fatigue during dynamic tasks, substantial heterogeneity exists regarding movement protocols, signal-processing strategies, and fatigue metrics. Given the shoulders' central role in daily activities, this study evaluates a predefined Continuous Wavelet Transform (CWT)-based analysis framework for assessing fatigue-related spectral changes during dynamic shoulder abduction.

Eight healthy subjects performed eleven standardized metronome-guided shoulder abductions under two conditions: without (W₀) and with (W₁) 2-kg handheld weights on three consecutive days. Bilateral sEMG recordings from the upper trapezius (UT) and lower trapezius (LT) were analyzed using CWT-based median frequency analysis. For comparison, the complete analytical pipeline was additionally applied using FFT-based median frequency estimation. Signal preprocessing included Butterworth- and Bandnotch filtering. After signal division into 50%-overlapping 1-s time windows, fatigue-related spectral changes were quantified by linear regression of median frequency over time. Statistical analysis was performed using repeated-measures ANOVA.

Repeated-measures ANOVA revealed significant load-related reductions in median-frequency regression slopes for the right UT (F (1,7) = 10.09, p = 0.016, ηp² = 0.59), left UT (F (1,7) = 15.71, p = 0.006, ηp² = 0.69), and right LT (F (1,7) = 7.54, p = 0.029, ηp² = 0.52). Similar trends were observed in the left LT but did not reach statistical significance. Analysis of muscular differences in response to load revealed significantly greater responses in the UT compared with the LT (F (1,15) = 6.35, p = 0.024, ηp² = 0.3). Exploratory hand dominance analysis demonstrated larger slope reductions on the non-dominant side, reaching statistical significance only for the LT in W₁ (p = 0.037). Additional FFT analysis showed comparable fatigue-related trends but consistently weaker statistical evidence than the CWT-based approach.

These findings demonstrate the feasibility of the applied CWT-based framework for detecting fatigue-related spectral changes in dynamic shoulder movements. By addressing limitations of conventional frequency-domain approaches, the approach represents a promising framework for functional sEMG analysis and may support a basis for future applications in biomechanics, rehabilitation, and performance monitoring.