Dexterous multi-finger robotic grasping faces two primary limitations: (1) traditional demonstration-based learning requires a separate trajectory to be trained for each object type, resulting in limited generalization across different objects; and (2) single-agent reinforcement learning struggles to coordinate the asynchronous motions of the thumb and fingers with heterogeneous joints, leading to inefficient exploration and slow convergence. To overcome these limitations, this study proposed two core innovations, including (1) a dynamic modulation kernel movement primitive (DM-KMP) algorithm, which combines dynamic movement primitives with kernel-based movement primitives to preserve trajectory topology and local probabilistic modulation characteristics; and (2) a novel potential-guided weight scheduling multi-agent deep deterministic policy gradient method (PWS-MADDPG), in which the thumb and fingers are modeled as independent agents. This enables the generation of grasping trajectories for diverse object categories using only a limited number of demonstrations. By leveraging a potential field model to optimize action selection, this method accelerates model convergence. Moreover, a multi-modal reward function with weight scheduling was designed to facilitate a seamless transition from individual optimization to coordinated team behavior. Experiment results demonstrated that the proposed PWS-MADDPG achieved a grasping success rate of 85.48% on a dexterous hand grasping across 11 object categories, outperforming the existing approaches—with a 40.31% improvement over the DDPG algorithm and a 20.05% improvement over the MADDPG algorithm. Furthermore, to verify the grasping stability and adaptability of the system in real-world scenarios, a grip-force validation experiment was designed using a custom cylindrical apparatus embedded with five six-axis sensors. These results demonstrate that the robotic hand using this novel method can maintain stable grasping and adaptively regulate grip and load forces in response to incremental external loads, based on a control policy that is trained in both simulation and real-world environments. These findings suggest that this method has strong Sim-to-Real transferability, biological plausibility in force adjustment, and robustness to environmental disturbances, enhancing both the training efficiency and generalization in simulation, and demonstrating high grasping robustness and successful rates in the real-world tasks. The proposed method holds strong potential for high-precision, safety-critical applications, including intelligent manufacturing, prosthetic control, and living assistance.

Patients with epilepsy experience significant neurological impairments due to abnormal electrical activities in their brains, impacting daily activities. Computer-aided diagnosis systems can assist neurologists in managing patients with this disease. Moreover, feature extraction has been proposed to be an integral part of the classification process. However, training machine learning (ML) models with multiple features from multichannel electroencephalogram (EEG) signals is computationally demanding. Feature selection (FS) minimizes the system’s computational cost by identifying deterministic features. However, individual FS techniques often show unstable performance across EEG datasets. Therefore, this study proposed a hybrid FS technique with a Random Forest Bayesian optimization classifier to dynamically select relevant features that improve ML performance across EEG datasets. Initially, the Bonn, CHB-MIT, and TUH EEG datasets were segmented into 1-s epochs with an overlap of 0.75. Relevant features are obtained through a hybrid of ANOVA, correlation, and graph-based FS techniques. Mean decrease impurity and meta-model were used to evaluate feature importance and ranking. RF-BO was used to predict the outcome of each FS technique for relevant cases of the Bonn, CHB-MIT, and TUH datasets after subject-level splitting with holdout. An average accuracy of 98.93% and 95.91% was obtained for the Bonn dataset’s binary and ternary classification cases, respectively. In addition, accuracies of 96.51% and 91.71%, respectively, obtained for the binary case of CHB-MIT and the multiclass case of TUH datasets with 60 features are better than those in previous studies, making the proposed model effective for seizure classifications.

Auditory brainstem response (ABR) is a weak evoked EEG signal that provides an objective measure for assessing auditory function. However, the traditional extraction method, namely, averaging is noise-sensitive and needs thousands of trials, which places high demands on the subjects and the experimental environment. Kalman weighted (KW) technique has the potential to extract ABR with high quality but relies heavily on expert experience for precise parameter tuning. In this article, an adaptive Kalman denoising technique, which can adaptively adjust the parameter, was developed. A comprehensive investigation was carried out on different noise types (pink noise/Gaussian white noise/uniform noise), proportions (20%/40%/60%/80%/100%), amplitudes (20/40/60/80μV), and integrated manners (early-noised/intermittent-noised/late-noised). Multiple metrics, such as Pearson correlation coefficient, root mean square error, latency and amplitude of characteristics wave, and the wave recognition rate were calculated for evaluation. The simulation results showed that the proposed method outperformed the averaging and KW techniques over these evaluation metrics. Also, these evaluation metrics of the proposed method were much more stable than those of averaging the KW. Finally, we verified the proposed method in the real scenario. It is believed that the proposed method opens a window for daily ABR-based auditory health condition screening, which can benefit the early detection and diagnosis of auditory diseases.

The control performance of myoelectric prostheses would not only depend on the feature extraction and classification algorithms but also on interactions of dynamic window-based hyperparameters (WBHP) used to construct input signals. However, the relationship between these hyperparameters and how they influence the performance of the convolutional neural networks (CNNs) during motor intent decoding has not been studied. Therefore, we investigated the impact of various combinations of WBHP (window length and overlap) employed for the construction of raw two-dimensional (2-D) surface electromyogram (sEMG) signals on the performance of CNNs when used for motion intent decoding. Moreover, we examined the relationship between the window length of the 2-D sEMG and three commonly used CNN kernel sizes. To ensure high confidence in the findings, we implemented three CNNs, which are variants of the existing models, and a newly proposed CNN model. Experimental analysis was conducted using three distinct benchmark databases, two from upper limb amputees and one from able-bodied subjects. The results demonstrate that the performance of the CNNs improved as the overlap between consecutively generated 2-D signals increased, with 75% overlap yielding the optimal improvement by 12.62% accuracy and 39.60% F1-score compared to no overlap. Moreover, the CNNs performance was better for kernel size of seven than three and five across the databases. For the first time, we have established with multiple evidence that WBHP would substantially impact the decoding outcome and computational complexity of deep neural networks, and we anticipate that this may spur positive advancement in myoelectric control and related fields.

Epilepsy is a neurological disorder characterized by abnormal neuronal discharges that manifest in life-threatening seizures. These are often monitored via EEG signals, a key aspect of biomedical signal processing (BSP). Accurate epileptic seizure (ES) detection significantly depends on the precise identification of key EEG features, which requires a deep understanding of the data's intrinsic domain. Therefore, this study presents an Advanced Multi-View Deep Feature Learning (AMV-DFL) framework based on machine learning (ML) technology to enhance the detection of relevant EEG signal features for ES. Our method initially applies a fast Fourier transform (FFT) on EEG data for traditional frequency domain feature (TFD-F) extraction and directly incorporates time domain (TD) features from the raw EEG signals, establishing a comprehensive traditional multi-view feature (TMV-F). Deep features are subsequently extracted autonomously from optimal layers of one-dimensional convolutional neural networks (1D CNN), resulting in multi-view deep features (MV-DF) integrating both time and frequency domains. A multi-view forest (MV-F) is an interpretable rule-based advanced ML classifier used to construct a robust, generalized classification. Tree-based SHAP explainable artificial intelligence (T-XAI) is incorporated for interpreting and explaining the underlying rules. Experimental results confirm our method's superiority, surpassing models using TMV-FL and single-view deep features (SV-DF) by 4% and outperforming other state-of-the-art methods by an average of 3% in classification accuracy. The AMV-DFL approach aids clinicians in identifying EEG features indicative of ES, potentially discovering novel biomarkers, and improving diagnostic capabilities in epilepsy management.

The efficient patient-independent and interpretable framework for electroencephalogram (EEG) epileptic seizure detection (ESD) has informative challenges due to the complex pattern of EEG nature. Automated detection of ES is crucial, while Explainable Artificial Intelligence (XAI) is urgently needed to justify the model detection of epileptic seizures in clinical applications. Therefore, this study implements an XAI-based computer-aided ES detection system (XAI-CAESDs), comprising three major modules, including of feature engineering module, a seizure detection module, and an explainable decision-making process module in a smart healthcare system. To ensure the privacy and security of biomedical EEG data, the blockchain is employed. Initially, the Butterworth filter eliminates various artifacts, and the Dual-Tree Complex Wavelet Transform (DTCWT) decomposes EEG signals, extracting real and imaginary eigenvalue features using frequency domain (FD), time domain (TD) linear feature, and Fractal Dimension (FD) of non-linear features. The best features are selected by using Correlation Coefficients (CC) and Distance Correlation (DC). The selected features are fed into the Stacking Ensemble Classifiers (SEC) for EEG ES detection. Further, the Shapley Additive Explanations (SHAP) method of XAI is implemented to facilitate the interpretation of predictions made by the proposed approach, enabling medical experts to make accurate and understandable decisions. The proposed Stacking Ensemble Classifiers (SEC) in XAI-CAESDs have demonstrated 2% best average accuracy, recall, specificity, and F1-score using the University of California, Irvine, Bonn University, and Boston Children's Hospital-MIT EEG data sets. The proposed framework enhances decision-making and the diagnosis process using biomedical EEG signals and ensures data security in smart healthcare systems.

OBJECTIVE: The decline in vascular elasticity with aging can be manifested in the shape of pulse wave. The study investigated the pulse wave features that are sensitive to age and the pattern of these features change with increasing age were examined. METHODS: Five features were proposed and extracted from the photoplethysmography (PPG)-based pulse wave or its first derivative wave. The correlation between these PPG features and ages was studied in 100 healthy subjects with a wide range of ages (20-71 years). Piecewise regression coefficients were calculated to examine the rates of change of the PPG features with age at different age stages. RESULTS: The proposed PPG features obtained from the finger showed a strong and significant correlation with age (with r = 0.76 - 0.77, p < 0.01), indicating higher sensitivity to age changes compared to the PPG features reported in previous studies (with r = 0.66 - 0.75). The correlation remained significant even after correcting for other clinical variables. The rate of change of the PPG feature values was found to be significantly faster in subjects aged ≥40 years compared to those aged < 40 years in the healthy population. This rate of change was similar to the age-related progression of arterial stiffness evaluated by pulse wave velocity (PWV), which is considered a gold standard for evaluating vascular stiffness. CONCLUSIONS: The proposed PPG features showed a high correlation with chronological age in healthy subjects and exhibited a similar age-related change trend as PWV. SIGNIFICANCE: With the convenience of PPG measures, the proposed age-related features have the potential to be used as biomarkers for vascular aging and estimating the risk of cardiovascular disease.

BACKGROUND: Monitoring spine kinematics is crucial for applications like disease evaluation and ergonomics analysis. However, the small scale of vertebrae and the number of degrees of freedom present significant challenges for noninvasive and convenient spine kinematics estimation. METHODS: This study developed a dynamic optimization framework for wearable spine motion tracking at the intervertebral joint level by integrating smartphone videos and Inertia Measurement Units (IMUs) with dynamic constraints from a thoracolumbar spine model. Validation involved motion data from 10 healthy males performing static standing, dynamic upright trunk rotations, and gait. This data included rotations of ten IMUs on vertebrae and virtual landmarks from three smartphone videos preprocessed by OpenCap, an application leveraging computer vision for pose estimation. The kinematic measures derived from the optimized solution were compared against simultaneously collected infrared optical marker-based measurements and in vivo literature data. Solutions only based on IMUs or videos were also compared for accuracy evaluation. RESULTS: The proposed optimization approach closely matched the reference data in the intervertebral or segmental rotation range, demonstrating minimal angular differences across all motions and the highest correlation in 3D rotations (maximal Pearson and intraclass correlation coefficients of 0.92 and 0.94, respectively). Time-series changes of joint angles also aligned well with the optical-marker reference. CONCLUSION: Dynamic optimization of the spine simulation that integrates IMUs and computer vision outperforms the single-modality method. SIGNIFICANCE: This markerless 3D spine motion capture method holds potential for spinal health assessment in large cohorts in real-world settings without dedicated laboratories.

Continuous estimation of finger joints based on surface electromyography (sEMG) has attracted much attention in the field of human-machine interface (HMI). A couple of deep learning models were proposed to estimate the finger joint angles for specific subject. When applied onto a new subject, however, the performance of the subject-specific model would degrade significantly due to the inter-subject differences. Therefore, a novel cross-subject generic (CSG) model was proposed in this study to estimate continuous kinematics of finger joints for new users. Firstly, a multi-subject model based on the LSTA-Conv network was built by using sEMG and finger joint angles data from multiple subjects. Then, the subjects adversarial knowledge (SAK) transfer learning strategy was adopted to calibrate the multi-subject model with the training data from a new user. With the updated model parameters and the testing data from the new user, multiple finger joint angles could be estimated afterwards. The overall performance of the CSG model for new users was validated on three public datasets from Ninapro. The results showed that the newly proposed CSG model significantly outperformed five subject-specific models and two transfer learning models in terms of Pearson correlation coefficient, root mean square error, and coefficient of determination. Comparison analysis showed that both the long short-term feature aggregation (LSTA) module and the SAK transfer learning strategy contributed to the CSG model. Moreover, increasing number of subjects in training set improved the generalization capability of the CSG model. The novel CSG model would facilitate the application of robotic hand control and other HMI settings.

Gaussian mixture model (GMM) is a powerful statistical tool in data modeling, especially for unsupervised learning tasks. Traditional learning methods for GMM such as expectation maximization (EM) require the covariance of the Gaussian components to be non-singular, a condition that is often not satisfied in real-world applications. This paper presents a new learning method called G$^2$M$^2$ (General Gaussian Mixture Model) by fitting an unnormalized Gaussian mixture function (UGMF) to a data distribution. At the core of G$^2$M$^2$ is the introduction of an integral cosine similarity (ICS) function for comparing the UGMF and the unknown data density distribution without having to explicitly estimate it. By maximizing the ICS through Monte Carlo sampling, the UGMF can be made to overlap with the unknown data density distribution such that the two only differ by a constant scalar, and the UGMF can be normalized to obtain the data density distribution. A Siamese convolutional neural network is also designed for optimizing the ICS function. Experimental results show that our method is more competitive in modeling data having correlations that may lead to singular covariance matrices in GMM, and it outperforms state-of-the-art methods in unsupervised anomaly detection.

Electromyogram pattern recognition (EMG-PR) is considered a potential method for upper limb prosthesis control. In principle, the feature extraction technique has been ranked the most influential factor that affect the EMG-PR method's performance. Despite the progress made thus far, there are inevitable interferences that could not be handled by the usual signal filtering approaches that are applied to enhance the extracted features. To address this issue, this study proposed a technique based on Wiener filtering for the preprocessing of EMG signals towards increasing the classification performance of EMG-PR systems. The performance of the proposed approach was investigated with recordings of high-density surface EMG which was obtained from four transhumeral amputees who performed five classes of limb movements. Then, features of five time-domain were analyzed in terms of their decoding accuracy, sensitive, and F1-score, with and without the application of the proposed for linear discriminant analysis and support vector machine classifiers techniques. Experimental results showed that by applying the proposed technique to the different feature sets, significant improvements in classification accuracy, sensitivity, and F1-score were observed across all subjects and classifiers. The proposed method improved the average classification accuracy by an increase of approximately 6.24% compared with the conventional method, while an increment as high as 16.77% was recorded for individual classes of movements. The outcomes of this study indicate that Wiener filtering may potentially boost the performance of EMG-PR systems in practical applications.

This paper presents a gear-driven prosthetic hand designed for toddlers with transradial amputation. The hand design considers three main issues: weight, cost, and operability. The prosthetic hand and the cosmetic silicon glove are made based on the dimensions of a real hand. The simple, stable, and reliable gear-driven transmission helps to reduce the weight and the cost. A small actuator is embedded in the palm. During the grasp, the four fingers and the thumb flexes and extends as a unit to provide a wide range of holding area. The kinematics and static analysis in grasping was performed and the simulation results were compared with measured data. The motion performance and practical operability of the proposed hand was verified experimentally by a test system and a transradial subject.

This paper presents a lightweight (<;250 g) and low-cost (<;350 USD) biomimetic prosthetic hand with two actuators embedded in the palm. One of them is employed for flexion/extension of the five digits, and the other one is used for the adduction/abduction of thumb. Thus, the hand can achieve major grasping tasks that account for about 85% of activities in daily life. The unique transmission provides various advantages such as a compact structure, weight saving, and short driven distance. Furthermore, by using 3D printing technology, most parts of the prosthetic hand were made to be much lighter and have a humanlike appearance, compared with conventionally manufactured artificial hand. Finally, the performance and practical applicability of the proposed design was verified experimentally through both of a motion verification and an intuitive control test by a healthy subject and a transradial amputee.

Three-dimensional (3D) models of tooth-alveolar bone complex are needed in treatment planning and simulation for computer-aided orthodontics. Tooth and alveolar bone segmentation from computed tomography (CT) images is a fundamental step in reconstructing their models. Due to less application of alveolar bone in conventional orthodontic treatment which may cause undesired side effects, the previous studies mainly focused on tooth segmentation and reconstruction, and did not consider the alveolar bone. In this study, we proposed a method to implement both tooth and alveolar bone segmentation from dental CT images for reconstructing their 3D models. First, the proposed method extracted the connected region of tooth and alveolar bone from CT images using a global convex level set model. Then, individual tooth and alveolar bone are separated from the connected region based on Radon transform and a local level set model. The experimental results showed that the proposed method could successfully complete both the tooth and alveolar bone segmentation from CT images, and outperformed the state of the art tooth segmentation methods in terms of accuracy. This suggests that the proposed method can be used in reconstructing the 3D models of tooth-alveolar bone complex for precise treatment.

In this paper, a sensory-motor fusion-based manipulation and grasping control strategy has been developed for a robotic hand-eye system. The proposed hierarchical control architecture has three modules: 1) vision servoing; 2) surface electromyography (sEMG)-based movement recognition; and 3) hybrid force and motion optimization for manipulation and grasping. A stereo camera is used to obtain the 3-D point cloud of a target object and provides the desired operational position. The AdaBoost-based motion recognition is employed to discriminate different movements based on sEMG of human upper limbs. The operational space motion planning for bionic arm and force planning for multifingered robotic hand can be both transformed as a convex optimization problem with various constraints. A neural dynamics optimization solution is proposed and implemented online. The proposed formulation can achieve a substantial reduction of computational load. The actual implementation includes a bionic arm with dextrous hand, high-speed active vision, and an EMG sensors. A series of manipulation tasks consisting of tracking/recogniting/grasping of an object are implemented, and experiment results exhibit the responsiveness and flexibility of the proposed sensory motion fusion approach.

Piezoelectrets are polymer-foam based space-charge electrets with strong piezoelectric effect. The piezoelectricity in piezoelectrets occurs due to the elastic heterogeneous cellular structure and the regularly arranged dipolar space charges stored therein. Some polymers have been experimented for piezoelectret preparation, where polypropylene (PP) is the mostly applied material at present. PP piezoelectrets have several promising features, such as large piezoelectric d33coefficient, small thickness, light weight, low cost, large area scale, as well as flexibility and even stretchability, which would enable them very suitable for applications in signal sensing and energy harvesting. In this work, the electromechanical properties of flexible and stretchable PP piezoelectrets are introduced and some of their possible applications as wearable physiological-signal sensors and micro-energy harvesters are demonstrated by experiments.

This paper presents fuzzy approximation-based adaptive backstepping control of an exoskeleton for human upper limbs to provide forearm movement assistance so that a human forearm can track any continuous desired trajectory (or constant setpoint) in the presence of parametric/functional uncertainties, unmodeled dynamics, actuator dynamics, and/or disturbances from environments. Given the desired trajectories of human forearm positions, in the developed control, adaptive fuzzy approximators are used to estimate the dynamical uncertainties of the human-robot system, and an iterative learning scheme is utilized to compensate for unknown time-varying periodic disturbances. With the synthesis of the backstepping, iterative learning, and Lyapunov function approaches, the developed controller does not require exact knowledge of the exoskeleton model, and the close-loop system can be proven to be semiglobally uniformly bounded. Three comparison experiments are conducted to illustrate the effectiveness of the proposed control scheme by tracking periodic/repeated trajectories.

A lack of myoelectric sources after limb amputation is a critical challenge in the control of multifunctional motorized prostheses. To reconstruct myoelectric sources physiologically related to lost limbs, a newly proposed neural-function construction method, targeted muscle reinnervation (TMR), appears promising. Recent advances in the TMR technique suggest that TMR could provide additional motor command information for the control of multifunctional myoelectric prostheses. However, little is known about the nature of the physiological functional recovery of the reinnervated muscles. More understanding of the underlying mechanism of TMR could help us fine tune the technique to maximize its capability to achieve a much higher performance in the control of multifunctional prostheses. In this study, rats were used as an animal model for TMR surgery involving transferring a median nerve into the pectoralis major, which served as the target muscle. Intramuscular myoelectric signals reconstructed following TMR were recorded by implanted wire electrodes and analyzed to explore the nature of the neural-function reconstruction achieved by reinnervation of targeted muscles. Our results showed that the active myoelectric signal reconstructed in the targeted muscle was acquired one week after TMR surgery, and its amplitude gradually became stronger over time. These preliminary results from rats may serve as a basis for exploring the mechanism of neural-function reconstruction by the TMR technique in human subjects.

Distortion-product otoacoustic emissions (DPOAEs) are sound energy generated by healthy inner ears when stimulated by two tones. Since DPOAEs are physiologically related with the functional status of the inner ear, they have been widely used as a clinical tool in hearing screening and diagnoses. Currently, almost all DPOAEs recording systems use pure tones as the stimuli and can test only one frequency at a time, resulting in low efficiency and insufficient resolution. In this study, conventional pure tones were replaced by swept tones with time-varying frequencies to overcome the limitation of current DPOAEs measurements. A tracking filter with dynamic center frequencies was proposed to extract the swept-tone DPOAEs from recorded signals with stimulus artifacts and background noises. The results of this study showed that the dynamic tracking filter had great performance in effectively extracting the swept-tone DPOAEs under different noise conditions for both the simulation and experimental data. The spectrogram of the extracted swept-tone DPOAEs could provide useful information to examine the functional status of the inner ear and to identify the detailed frequency regions of the hearing loss. These preliminary findings suggested that the swept-tone DPOAEs might be useful for developing a more efficient and accurate tool for hearing loss screening in the clinic.

Wearable systems based on continuously monitoring of vital physiological signals without interfering with user's daily life much are desired urgently in health care. Similarly, the limb amputees who need to wear their myoelectric prostheses for a long time daily expect a comfortable and reliable prosthetic system. It is inconvenient in clinical application of a myoelectric prosthesis to use the commonly used gel electrode for electromyography (EMG) recording over all day. Textile electrode with characteristics of ventilation, flexibility, and folding, may be an ideal selection of physiological signal monitoring in clinical applications. In this study, the textile electrodes made using screen printing technology were used for EMG recordings and the real-time performance of the textile-electrode EMG in myoelectric control of multifunctional prostheses was investigated in transradial amputees and able-bodied subjects for comparison purpose. The results over seven able-bodied subjects showed that the textile electrode could achieve similar performance as conventional metal electrodes for both the off-line classification accuracy and the real-time motion completion rate in operating a virtual hand. With the textile electrodes, the average off-line classification accuracy of 73.4% and the real-time motion completion rate of 81.9% within a 5 s time limit were achieved in three transradial amputees. These pilot results suggested that the textile electrodes might be feasible for EMG recordings in control of myoelectric prostheses.

The goal of this special session is to present original and relevant contributions in the area of information sensing and computing for control of or communication with wearable robots in medical applications.