The measurement and analysis of the electrical activity of muscle provide information that aids in the control of assistive devices. The investigation of these signals under varied physiological conditions, such as fatigue, enables reliable control. Muscle fatigue is a muscular condition associated with loss of muscle function. The early detection of muscle fatigue using surface Electromyography (sEMG)-based electrical measurements is challenging due to the nonlinear variations of the signal. In this work, an attempt has been made to understand the effect of dynamic nonlinear variations in the characteristics of the signal to develop a reliable fatigue index. The methodology involves the acquisition of myoelectric signals from the biceps brachii muscle of 52 healthy participants during dynamic contractions. The acquired signals are preprocessed and are analyzed with symbolic transition networks. Features such as symbolic entropy, network entropy, uniformity, and, minimum and maximum effective degrees (EDs) are extracted for further analysis. Appropriate decision boundaries are established for each feature using receiver operator characteristics (ROCs) and machine learning algorithms. The results indicate a decrease in signal complexity with fatigue. All the extracted features show a statistically significant difference (p < 0.05) between both conditions. Symbolic entropy achieves an accuracy of 89%, and the maximum ED yields an accuracy of 90% based on thresholds estimated with ROC. Furthermore, only a marginal improvement is observed with the combination of these features and the Naive Bayes classifier. It appears that the proposed maximum ED could be used as a reliable fatigue index in real-time applications for the improvement of rehabilitation efficacy.

Exercise-induced muscle damage is a condition which results in the loss of muscle function due to overexertion. Muscle fatigue is a precursor of this phenomenon. The characterization of muscle fatigue plays a crucial role in preventing muscle damage. In this work, an attempt is made to develop signal processing methods to understand the dynamics of the muscle’s electrical properties. Surface electromyography signals are recorded from 50 healthy adult volunteers under dynamic curl exercise. The signals are preprocessed, and the first difference signal is computed. Furthermore, ascending and descending slopes are used to generate a binary sequence. The binary sequence of various motif lengths is analyzed using features such as the average symbolic occurrence, modified Shannon entropy, chi-square value, time irreversibility, maximum probability of pattern and forbidden pattern ratio. The progression of muscle fatigue is assessed using trend analysis techniques. The motif length is optimized to maximize the rho value of features. In addition, the first and the last zones of the signal are compared with standard statistical tests. The results indicate that the recorded signals differ in both frequency and amplitude in both inter- and intra-subjects along the period of the experiment. The binary sequence generated has information related to the complexity of the signal. The presence of more repetitive patterns across the motif lengths in the case of fatigue indicates that the signal has lower complexity. In most cases, larger motif length resulted in better rho values. In a comparison of the first and the last zones, most of the extracted features are statistically significant with p < 0.05. It is observed that at the motif length of 13 all the extracted features are significant. This analysis method can be extended to diagnose other neuromuscular conditions.

In drug-resistant epilepsy, a visual inspection of intracranial electroencephalography (iEEG) signals is often needed to localize the epileptogenic zone (EZ) and guide neurosurgery. The visual assessment of iEEG time-frequency (TF) images is an alternative to signal inspection, but subtle variations may escape the human eye. Here, we propose a deep learning-based metric of visual complexity to interpret TF images extracted from iEEG data and aim to assess its ability to identify the EZ in the brain. We analyzed interictal iEEG data from 1928 contacts recorded from 20 children with drug-resistant epilepsy who became seizure-free after neurosurgery. We localized each iEEG contact in the MRI, created TF images (1–70 Hz) for each contact, and used a pre-trained VGG16 network to measure their visual complexity by extracting unsupervised activation energy (UAE) from 13 convolutional layers. We identified points of interest in the brain using the UAE values via patient- and layer-specific thresholds (based on extreme value distribution) and using a support vector machine classifier. Results show that contacts inside the seizure onset zone exhibit lower UAE than outside, with larger differences in deep layers (L10, L12, and L13: p < 0.001). Furthermore, the points of interest identified using the support vector machine, localized the EZ with 7 mm accuracy. In conclusion, we presented a pre-surgical computerized tool that facilitates the EZ localization in the patient’s MRI without requiring long-term iEEG inspection.