Wavelet-derived features as indicators of physiological changes induced by bed rest

Objective: Bed rest studies are employed to simulate microgravity situations as encountered in spaceflight. Current methods of assessing muscle function impairment due to microgravity exposure include techniques such as maximum voluntary contraction assessments using force measurements. Such techniques involve impractical long-feedback loops for applications involving rehabilitation or otherwise detecting physiological changes. Recent studies have made use of the discrete wavelet transform in combination with machine learning methods to classify hand gestures and detect pathologies. In this paper, we demonstrate models capable of discriminating between the before and after bed rest states by extracting features from surface electromyography measurements. Methods: A previously conducted and studied bed rest experiment is examined by discrete wavelet transform for tractable feature sets for the purpose of k-nearest neighbor and Support Vector Machine classification. Forward feature selection is used with k-nearest neighbor or Support Vector Machine selection criteria. Classifiers are evaluated on non-wavelet-derived features for sake of comparison. Results: Wavelet-derived features perform well for both classifiers with classification accuracies as high as 95%. Models without wavelet-derived features do not perform as well overall. Conclusion: These high-accuracy results are promising for future efforts in neuromuscular monitoring and further investigations with larger sample sizes. Significance: Classification algorithms utilizing features derived by wavelet transforms provide a method toward development of short-feedback loop measurements of the physiological effects of prolonged disuse.