Joint and multiplex recurrence network analysis of muscle coordination during fatigue progression using sEMG

Muscle fatigue is recognized as a multi-muscle, coordination-driven phenomenon rather than an isolated single-muscle response. This study investigates whether joint and multiplex recurrence network (JMRN) features derived from surface electromyography (sEMG) signals capture fatigue-related changes in inter-muscle coordination during dynamic contractions. Signals are recorded from gastrocnemius medialis (GM), gastrocnemius lateralis (GL), and soleus (SOL) along with toe-force measurements of 40 male subjects (N = 40) during dynamic calf raise test (CRT) to exhaustion. Un-truncated sEMG bursts across all repetitions (mean 32.65 ± 8.18 bursts; range: 18-53 bursts: 1306 bursts total) are mapped into phase space (τ = 10 samples = 5.0 ms, m = 4, downsampled to 100 nodes per burst) to construct fuzzy recurrence layers (kNN = 10). Network features including multiplex edge overlap, pairwise cosine coupling, joint strength, and density are extracted. A coordination fatigue index (CFI) is derived using leave-one-subject-out cross-validated (LOSO-CV) principal component analysis (PCA) to track fatigue progression, with the first principal component (PC1) explaining 84.06% of total variance. The proposed CFI showed statistically significant monotonic association with subjective fatigue onset (N = 40: Spearman ρ = 0.393, p = 0.0122, mean absolute error (MAE) = 8.60 bursts, accuracy ± 5 = 67.50%, Bland-Altman bias: -0.11 bursts) and force-based fatigue onset (n = 14: ρ = 0.425, p = 0.130, MAE = 3.64 bursts). Coordination features showed significant post-fatigue increases (p_FDR < 0.005, Benjamini-Hochberg corrected), reflecting elevated spatial synchronization. The JMRN framework provides a non-invasive approach for studying multi-muscle interactions during dynamic fatigue progression.