Electromyographic Signal-Based Approaches: From Risk Prevention to Diagnostic Support and Functional Restoration

In recent years, the progressive miniaturization of sensors has fostered the development of advanced wearable systems capable of continuously and non-invasively acquiring physiological and biomechanical signals, enabling the monitoring of parameters such as muscle activity, joint kinematics, and cardiorespiratory variables. In this context, surface electromyography (sEMG) has emerged as one of the most versatile technologies for the analysis of neuromuscular function, owing to its ability to investigate the muscular origins of movement rather than merely the resulting motion itself. As a result, sEMG has found applications spanning occupational injury prevention, clinical diagnostics, rehabilitation support, and the control of advanced prosthetic devices. In the specific field of injury prevention, this technology has shown strong potential for identifying hazardous conditions associated with muscular fatigue even before the onset of the compensatory movements that typically emerge as fatigue progresses. In this context, the timely detection and quantification of worker fatigue are essential for implementing effective prevention strategies and reducing the risk of work-related musculoskeletal disorders. However, although the risk of injury can be mitigated, it can never be completely eliminated. Among occupational disorders, non-specific low back pain (LBP) stands out due to its high incidence and continuously increasing prevalence, highlighting the need for increasingly objective tools to support diagnosis and guide personalized rehabilitation programs. In this domain, sEMG can identify abnormal muscle activation patterns and compensatory motor strategies, providing quantitative information that may support clinicians in tailoring rehabilitation interventions and targeting specific muscular deficits. Finally, there are conditions in which the functional recovery achievable through rehabilitation is inherently limited. In the field of upper-limb amputation, the decoding of motor intention through surface electromyography signals represents one of the primary strategies for the intuitive control of myoelectric prostheses and for the restoration of lost functional capabilities. Nevertheless, the widespread adoption of these systems remains constrained by several challenges, including suboptimal electrode placement resulting from the highly variable muscle topology of amputees, as well as the reduced stability of gesture classification during the early phases of muscle contraction. This PhD thesis, entitled “Electromyographic Signal-Based Approaches: From Risk Prevention to Diagnostic Support and Functional Restoration”, proposes a unified framework based on the use of sEMG as a cross-domain technology for biomechanical risk prevention, diagnostic support, and functional substitution through advanced prosthetic systems. The overarching aim is to emphasize the largely untapped potential of this technology and to demonstrate its broad applicability in injury prevention, personalized rehabilitation support, and functional restoration through advanced prosthetic control following limb loss. In the field of injury prevention, the first study of this thesis investigated the use of surface and high-density electromyography (HD-sEMG) for estimating physical workload and muscle fatigue during a simulated industrial overhead task. Ten healthy participants performed a repetitive overhead screwing/unscrewing activity under two conditions: with and without the support of a passive exoskeleton. Bipolar and HD-sEMG signals, together with heart rate and respiratory rate, were acquired. Features extracted from the time, frequency, and spatial domains were then correlated with linear and exponential fatigue models to identify reliable indicators of muscular fatigue and biomechanical overload and subsequently ranked according to their correlation strength. Bilateral differences in muscle activation and fatigue-related behavior were then evaluated across seven monitored muscles. In the field of diagnostic support, a second study proposed a multimodal assessment platform combining sEMG and inertial measurement units for the evaluation of LBP. Patients with low back pain and healthy controls performed five functional tasks derived from the clinically validated Back Performance Scale. The objective was not only to identify altered muscle activation patterns, compensatory kinematic strategies, and quantitative biomarkers capable of discriminating between healthy and pathological conditions, but also to highlight potential therapeutic targets that could support and guide personalized rehabilitation programs. The third application area focused on functional substitution through prosthetic control. Initially, the optimization of the decoding process was addressed through the development of a data-driven algorithm for optimal electrode placement based on HD-sEMG spatial mapping of the residual limb, with the aim of identifying the most informative recording sites for reliable myoelectric control. Building upon these findings, a further optimization in the temporal domain was performed through the development of a hierarchical gesture- and force-classification strategy designed to improve system responsiveness during the early phases of muscle contraction by treating transient and steady-state contraction phases differently. In the context of biomechanical risk prevention, the results showed that several temporal- and frequency-domain features, particularly Root Mean Square and Instantaneous Median Frequency, exhibited strong correlations with the onset and progression of muscle fatigue, displaying predictable trends over time that enabled an accurate description of fatigue evolution during the task. Moreover, these features also demonstrated sensitivity to instantaneous physical effort, effectively discriminating between the exoskeleton-assisted and free-body conditions. Taken together, these results highlight their potential for the objective assessment of biomechanical workload and for evaluating the effectiveness of wearable support systems. Regarding LBP, the multimodal platform highlighted significant differences between patients and healthy controls in terms of muscle activation, range of motion, and trunk co-activation. Specific compensatory strategies, including trunk stiffness and reduced mobility, were identified, demonstrating the potential of combining inertial sensors and sEMG for the objective assessment of the pathology and for the characterization of the adaptive motor behaviors adopted by patients in response to pain. In the field of prosthetic control, the proposed spatial data-driven HD-sEMG channel selection method outperformed both equally spaced and Random Forest-based configurations in healthy subjects as well as in amputees. From a temporal optimization perspective, the proposed hierarchical classifier achieved high accuracy in the simultaneous recognition of gestures and force levels in both populations. Furthermore, the transient/steady-state strategy significantly improved system responsiveness, reducing reaction time from approximately 500 ms to 300 ms compared with conventional steady-state approaches. The obtained results demonstrate how sEMG can represent a cross-domain technology capable of supporting the entire continuum of neuromusculoskeletal care, from risk prevention to diagnosis and rehabilitation, up to functional substitution. In occupational settings, the identification of fatigue-sensitive features paves the way for intelligent systems capable of continuously monitoring biomechanical risk in real-world environments, thereby enabling the early detection of hazardous fatigue-related conditions. In the clinical domain, the integration of sEMG and inertial sensors overcomes the limitations of traditional subjective assessment scales by providing quantitative metrics on specific body districts, identifying compensatory behaviors adopted by patients in response to pain, and ultimately offering valuable insights for the development of personalized rehabilitation programs. Finally, in the field of myoelectric prostheses, the combination of spatial and temporal optimization achieved through the proposed electrode placement and hierarchical classification algorithms demonstrated strong potential for improving the robustness, intuitiveness, and responsiveness of prosthetic control, all of which are essential for promoting the effective daily use of prosthetic devices by amputees. This thesis proposes and validates an integrated framework based on sEMG for applications in biomechanical risk prevention, diagnostic support, and advanced prosthetic control. The results demonstrate that the integration of wearable sensors, biomechanical analysis, and data-driven algorithms enables the development of effective, non-invasive, and adaptable systems for different application scenarios. Overall, this work highlights the central role of sEMG as an enabling technology for industrial ergonomics, personalized medicine, and assistive technologies, contributing to the development of safer workplaces, more effective rehabilitation strategies, and more intuitive and high-performing prosthetic systems.