Recent studies have highlighted the possibility of using surface electromyographic (EMG) signals to develop human–computer interfaces that are also able to recognize complex motor tasks involving the hand as the handwriting of digits. However, the automatic recognition of words from EMG information has not yet been studied. The aim of this study is to investigate the feasibility of using combined forearm and wrist EMG probes for solving the handwriting recognition problem of 30 words with consolidated machine-learning techniques and aggregating state-of-the-art features extracted in the time and frequency domains. Six healthy subjects, three females and three males aged between 25 and 40 years, were recruited for the study. Two tests in pattern recognition were conducted to assess the possibility of classifying fine hand movements through EMG signals. The first test was designed to assess the feasibility of using consolidated myoelectric control technology with shallow machine-learning methods in the field of handwriting detection. The second test was implemented to assess if specific feature extraction schemes can guarantee high performances with limited complexity of the processing pipeline. Among support vector machine, linear discriminant analysis, and K-nearest neighbours (KNN), the last one showed the best classification performances in the 30-word classification problem, with a mean accuracy of 95% and 85% when using all the features and a specific feature set known as TDAR, respectively. The obtained results confirmed the validity of using combined wrist and forearm EMG data for intelligent handwriting recognition through pattern recognition approaches in real scenarios.

Objective: Despite hand gesture recognition is a widely investigated field, the design of myoelectric architectures for detecting finer motor task, like the handwriting, is less studied. However, writing tasks involving cognitive loads represent an important aspect toward the generalization of myoelectric-based human-machine interfaces (HMI), and also for many rehabilitative tasks. In this study, the handwriting recognition of the ten digits was faced under the myoelectric control perspective, considering the probes setup and the feature extraction step. Methods: Time and frequency domain features were extracted from surface electromyography (sEMG) signals of 11 subjects who wrote the ten digits following a standardized template and 8 sEMG probes were equally distributed between forearm and wrist. Feature class separability was investigated and an aggregated feature set was built to train pattern recognition architectures, i.e. linear discriminant analysis (LDA) and quadratic support vector machine (QSVM). Also, four reduced probes setups were investigated. Results: LDA and QSVM showed mean accuracy of about 97%, with all the forearm and wrist sEMG information. A significant reduction of performances was observed considering the wrist or the forearm only (≤92%) and when LDA and QSVM were trained with two electrodes information (≤90%). Conclusions: For the reliable classification performances in a motor task involving high cognitive demands, like the handwriting, it is required the use of probes fully covering forearm and wrist. Outcomes support the methodological transfer from myoelectric hand gesture to the handwriting recognition, which represents a key aspect in the development of new HMI for rehabilitation tasks.

The volitional control of powered assistive devices is commonly performed by mapping the electromyographic (EMG) activity of the lower limb to joints’ angular kinematics, which are then used as the input for regulation. However, during walking, the ground reaction force (GRF) plays a central role in the modulation of the gait, providing dynamic stability and propulsion during the stance phase. Including this information within the control loop of prosthetic devices can improve the quality of the final output, providing more physiological walking dynamics that enhances the usability and patient comfort. In this work, we explored the feasibility of the estimation of the ground reaction force vertical component (VGRF) by using only the EMG activities of the thigh and shank muscles. We compared two deep learning models in three experiments that involved different muscular configurations. Overall, the outcomes show that the EMG signals could be leveraged to obtain a reliable estimation of the VGRF during walking, and the shank muscles alone represent a viable solution if a reduced recording setup is needed. On the other hand, the thigh muscles failed in providing performance enhancements, either when used alone or together with the shank muscles. The results outline the feasibility of including GRF information within an EMG-driven control scheme for prosthetic and assistive devices.

The use of inertial measurement units is gaining attention to estimate human joint kinematics. However, to obtain clinically meaningful results, sensor frame needs to be aligned with the underlying anatomical one. Although during the years different approaches have been proposed, a common consensus has not been reached. Further, inertial sensor positioning on human segments can affect frame definition and kinematics estimation. Thus, the aim of the present work is to define an anatomical calibration procedure for lower limb joints kinematics, robust with respect to sensor misalignment, and based on a limited set of movements, with static and functional assumptions. To this purpose, straight walking and turning motor tasks in six healthy subjects were considered, and results were compared with those provided by an optoelectronic system. Three sensor placements have been also evaluated to test the procedure with respect to sensor positioning. After offset removal, an average RMSE ≤2.5 deg in gait, and ≤2 deg in turning for all the configurations were obtained, outperforming results from previous approaches. Average offset values resulted about 6 deg for hip and ankle, whereas negligible for the knee. Outcomes of this study enable a simple and accurate measurement of clinically meaningful joints kinematics, also without a strict sensor placement.

The early diagnosis of diabetic neuropathy (DN) is fundamental in order to enact timely therapeutic strategies for limiting disease progression. In this work, we explored the suitability of standing balance task for identifying the presence of DN. Further, we proposed two diagnosis pathways in order to succeed in distinguishing between different stages of the disease. We considered a cohort of non-neuropathic (NN), asymptomatic neuropathic (AN), and symptomatic neuropathic (SN) diabetic patients. From the center of pressure (COP), a series of features belonging to different description domains were extracted. In order to exploit the whole information retrievable from COP, a majority voting ensemble was applied to the output of classifiers trained separately on different COP components. The ensemble of kNN classifiers provided over 86% accuracy for the first diagnosis pathway, made by a 3-class classification task for distinguishing between NN, AN, and SN patients. The second pathway offered higher performances, with over 97% accuracy in identifying patients with symptomatic and asymptomatic neuropathy. Notably, in the last case, no asymptomatic patient went undetected. This work showed that properly leveraging all the information that can be mined from COP trajectory recorded during standing balance is effective for achieving reliable DN identification. This work is a step toward a clinical tool for neuropathy diagnosis, also in the early stages of the disease.

This study proposes a feature extraction scheme that fuses accelerometric (ACC) and electromyographic (EMG) data to improve shoulder movement identification in individuals with transhumeral amputation, in whom the clinical need for intuitive control strategies enabling reliable activation of full-arm prostheses is underinvestigated. A novel spatiotemporal warping feature extraction architecture was employed to realize EMG and ACC information fusion at the feature level. EMG and ACC data were collected from six participants with intact limbs and four participants with transhumeral amputation using an NI USB-6009 device at 1000 Hz to support the proposed feature extraction scheme. For each participant, a leave-one-trial-out (LOTO) training and testing approach was used for developing pattern recognition models for both the intact-limb (IL) and amputee (AMP) groups. The analysis revealed that the introduction of ACC information has a positive impact when using windows of length (WLs) lower than 150 ms. A linear discriminant analysis (LDA) classifier was able to exceed the accuracy of 90% in each WL condition and for each group. Similar results were observed for an extreme learning machine (ELM), whereas k-nearest neighbors (kNN) and an autonomous learning multi-model classifier showed a mean accuracy of less than 87% for both IL and AMP groups at different WLs, guaranteeing applicability over a large set of shallow pattern-recognition models that can be used in real scenarios. The present work lays the groundwork for future studies involving real-time validation of the proposed methodology on a larger population, acknowledging the current limitation of offline analysis.