Fundamental underpinnings of electromyography-based biofeedback for transtibial amputee gait rehabilitation

Electromyography (EMG) feedback can deliver targeted information about the generation of movement patterns, with potential for daily living applications. Our understanding of how to implement EMG-based feedback within lower limb amputees is limited. This thesis investigated three fundamental features of EMG that require consideration prior to being incorporated into a biofeedback system. Gait temporal parameters, commonly derived from ground reaction forces and kinematic algorithms, identify key gait phases and permit analysis of variables in reference to the gait cycle. Identification of these parameters from EMG without the necessity for additional biomechanical measures is key for the application of an EMG-based biofeedback system outside of a clinical setting during daily living. Gait phases (stance and swing) were classified from three able-bodied lower limb muscles using a binary support vector machine (97.58 – 97.62% accuracy). The absolute error in heel strike and ground contact time were smaller than commonly implemented kinematic algorithms (absolute error in toe off was comparable). An able-bodied trained model (different muscles) was implemented within a lower limb amputee population, obtaining an accuracy of 92.39%. Prior warning of impending heel strikes could potentially help augment the diminished internal feedback from the prosthetic limb, however, the real time application was limited by the feature window length. To evaluate whether a feedback intervention has elicited a meaningful muscle activation adaptation beyond normal variation, knowledge of the signal’s nature is required. Variability quantified how much motor patterns varied in magnitude at time points, and Lyapunov exponents quantified how much the motor pattern varied across a time series, providing information regarding control strategies (stability), with analysis requiring many consecutive gait cycles. The present thesis proposed a method to concatenate short kinematic time series (overground trials) to provide accurate estimates of short-term exponents (0 – 1 stride). Gait rehabilitation typically occurs within two environments: overground or treadmill. Treadmill walking displayed a tendency to decrease variability and increase stability of muscle activations and kinematics compared to overground walking (mean ± standard deviation differences were not observed for many variables), which potentially limits motor learning and the translation of learnt motor patterns to daily living. Population-level differences in muscle activation (except for stability) and kinematic, variability and stability were observed. Kinematic differences were masked during treadmill walking (with the exception of the prosthetic limb ankle), highlighting the importance of feedback-augmented rehabilitation within a habitual overground environment. A recent shift towards individualised rehabilitation was supported by the present thesis. Able-bodied and lower limb amputee (intact and prosthetic limb) participants were identified to an accuracy of 98.1% and 100.0% respectively, using a multi-class support vector machine. Demonstrating the presence of large inter-participant variability and the necessity for personalised feedback. The observed unique motor patterns differed with walking modality. Walking modality was identified to an accuracy of 92.2% using a binary support vector machine, intra-participant approach. Previous lower limb amputee feedback devices have focused on delivering information about spatiotemporal parameters, kinematics, and lower limb loading. Electromyography has the capacity to provide information about how gross and compensatory movement patterns are generated. Accurate classification of gait events from EMG permits identification of key gait phases outside of the clinical setting, without the necessity of additional biomechanical variables. Feedback-augmented overground-based rehabilitation uses EMGs wireless and wearable features and is favourable to treadmill-based rehabilitation. Unique muscle activation profiles and motor control strategies were not consistent across both walking modalities potentially limiting motor learning and transferability of learnt motor patterns to daily living. Future research should focus on overground-based feedback, which limits the information that can be delivered to time-discrete variables, using audio or haptic feedback. The identification of unique muscle activation profiles highlights the necessity for future EMG-based biofeedback systems to adopt an individualised approach, as a comparison to an expected ‘norm’ is not appropriate.