Clinical studies have shown that speech pauses can reflect the cognitive function differences between Alzheimer’s Disease (AD) and non-AD patients, while the value of pause information in AD detection has not been fully explored. Herein, we propose a speech pause feature extraction and encoding strategy for only acoustic-signal-based AD detection. First, a voice activity detection (VAD) method was constructed to detect pause/non-pause feature and encode it to binary pause sequences that are easier to calculate. Then, an ensemble machine-learning-based approach was proposed for the classification of AD from the participants’ spontaneous speech, based on the VAD Pause feature sequence and common acoustic feature sets (ComParE and eGeMAPS). The proposed pause feature sequence was verified in five machine-learning models. The validation data included two public challenge datasets (ADReSS and ADReSSo, English voice) and a local dataset (10 audio recordings containing five patients and five controls, Chinese voice). Results showed that the VAD Pause feature was more effective than common feature sets (ComParE: 6373 features and eGeMAPS: 88 features) for AD classification, and that the ensemble method improved the accuracy by more than 5% compared to several baseline methods (8% on the ADReSS dataset; 5.9% on the ADReSSo dataset). Moreover, the pause-sequence-based AD detection method could achieve 80% accuracy on the local dataset. Our study further demonstrated the potential of pause information in speech-based AD detection, and also contributed to a more accessible and general pause feature extraction and encoding method for AD detection.

Objectives: The goal of this study is to design a novel approach for extracting volitional electromyography (vEMG) contaminated by functional electrical stimulation (FES) with time variant amplitudes and frequencies. Methods: A selective interpolation (SI) is adopted to eliminate the initial spike. Then the interpolated signal is decomposed into intrinsic mode functions by using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Each IMF is window-filtered based on a logistic regression (LR) classifier to identify the IMFs contaminated by FES. Semi-simulated signals were generated using EMG and stimulation response and three metrics were adopted to validate the performance of the proposed algorithm, including the a) signal-to-noise ratio (SNR), b) normalized root mean squared error (NRMSE) and c) cross-correlation coefficient between the clean EMG and the extracted EMG. Real FEScontaminated EMG was collected from six able bodied volunteers and one stroke patient. The correlation coefficients between the extracted EMG and the wrist torque were analyzed. Results: The simulation results showed a higher SNR (2.12 to -2.13dB), higher correlation (0.8810.08) and lower NRMSE (0.78 to 1.29) than those of the comb filter and EMD-Notch algorithm. The EMGTorque correlation coefficients were 0.7510.07 for monopolar pulses and 0.7710.12 for bipolar pulses. For the stroke patient, the algorithm also successfully extracted underlying vEMG from time variant FES noises. Conclusions: All results showed that SICEEMDAN-LR is capable of extracting EMG during FES with time-variant parameters.

In this study, a wearable prototype system was developed for multiple-gesture rehabilitation using electrical stimulation controlled by a volitional surface electromyography (sEMG) scan of a healthy forearm. The purpose of the prototype system is to reconstruct multiple gestures of a paralysed limb and to simplify the positioning of sEMG detection sites on a healthy forearm. A self-designed eight-channel sEMG detection armband was used to detect the sEMG signal distributions of the muscle groups in healthy forearms. Linear discriminant analysis (LDA) was used to classify the sEMG signal distributions corresponding to different gestures, and then the classification results were mapped to corresponding stimulation channels. The sEMG signal with the maximum root mean square (RMS) was used as the source of stimulus coding for each gesture. Our proposed mean absolute value (MAV)/number of slope sign changes (NSS) dual-coding (MNDC) algorithm was used to encode the sEMG signal into an electrical stimulus with a dynamic pulse width and frequency. The constant-current stimulation armband electrically stimulated multiple muscles in the affected forearm by means of a circuit designed with a time-division multiplexed stimulation channel. An experiment involving 6 able-bodied volunteers showed that when the detection armband was located near the middle of the forearm, the gesture classification accuracy was greater than 90%, and each active sEMG signal was high. Gesture bridge experiments, including grasping, wrist flexion, wrist extension and finger extension, were carried out among six hemiplegic subjects and between one able-bodied volunteer acting as a controller and each of six stroke patients as the controllee. Both sets of results show that the proposed system can reconstruct these four gestures in the controlled subject with a delay of at most 360 ms and with a correlation coefficient of > 0.72.

Objective. Recent studies have shown that integrating inertial measurement unit (IMU) signals with surface electromyographic (sEMG) can greatly improve hand gesture recognition (HGR) performance in applications such as prosthetic control and rehabilitation training. However, current deep learning models for multimodal HGR encounter difficulties in invasive modal fusion, complex feature extraction from heterogeneous signals, and limited inter-subject model generalization. To address these challenges, this study aims to develop an end-to-end and inter-subject transferable model that utilizes non-invasively fused sEMG and acceleration (ACC) data. Approach. The proposed non-invasive modal fusion-transformer (NIMFT) model utilizes 1D-convolutional neural networks-based patch embedding for local information extraction and employs a multi-head cross-attention (MCA) mechanism to non-invasively integrate sEMG and ACC signals, stabilizing the variability induced by sEMG. The proposed architecture undergoes detailed ablation studies after hyperparameter tuning. Transfer learning is employed by fine-tuning a pre-trained model on new subject and a comparative analysis is performed between the fine-tuning and subject-specific model. Additionally, the performance of NIMFT is compared to state-of-the-art fusion models. Main results. The NIMFT model achieved recognition accuracies of 93.91%, 91.02%, and 95.56% on the three action sets in the Ninapro DB2 dataset. The proposed embedding method and MCA outperformed the traditional invasive modal fusion transformer by 2.01% (embedding) and 1.23% (fusion), respectively. In comparison to subject-specific models, the fine-tuning model exhibited the highest average accuracy improvement of 2.26%, achieving a final accuracy of 96.13%. Moreover, the NIMFT model demonstrated superiority in terms of accuracy, recall, precision, and F1-score compared to the latest modal fusion models with similar model scale. Significance. The NIMFT is a novel end-to-end HGR model, utilizes a non-invasive MCA mechanism to integrate long-range intermodal information effectively. Compared to recent modal fusion models, it demonstrates superior performance in inter-subject experiments and offers higher training efficiency and accuracy levels through transfer learning than subject-specific approaches.

Muscle synergies offer valuable insights into the movement strategies employed by the central nervous system and present a promising avenue for clinical applications. However, the field lacks a complete understanding of how surface electromyography processing parameters affect muscle synergy analysis, which in turn has hindered cross-study comparisons and the translation of experimental results to clinical contexts. To address the gap, this study presents a systematic evaluation of interactive effects of three key parameters on muscle synergy analysis, including nine cut-off frequencies of low-pass filters, five normalization methods, and five synergy extraction criteria, covering 225 unique combinations. Using a comprehensive running dataset of 135 subjects, this study examined variance accounted for (VAF) and correlation coefficient (R2) metrics, the number of synergies, and synergy structure consistency under different parameter settings. Synergy similarity was used as a quantitative measure of synergy stability across different parameter settings. The results demonstrated that cut-off frequencies, normalization methods, and criteria choices interactively influenced the outcomes. Notably, VAF consistently yielded higher values than R2, highlighting differences in how these metrics capture explained variance. Error VAF (EVAF) emerged as the most robust criterion for determining the number of synergies, especially when combined with normalization methods by maximum value (MAX), average value (AVE), or unit variance (UVA) and moderately high cut-off frequencies, which led to more stable synergy structures across conditions. Furthermore, the predefined threshold associated with each criterion markedly affected the estimated number of synergies. These findings provide structured guidelines for muscle synergy analysis, helping to standardize preprocessing and extraction parameters, improve reproducibility across studies, and enhance the clinical applicability of synergy-based assessments.