Journal of neural engineering最新文献

Objective: Visual prostheses are effective tools for restoring vision, yet real-world complexities pose ongoing challenges. The progress in AI has led to the emergence of the concept of intelligent visual prosthetics with auditory support, leveraging deep learning to create practical artificial vision perception beyond merely restoring natural sight for the blind.

Approach: This study introduces an object-based attention mechanism that simulates human gaze points when observing the external world to descriptions of physical regions. By transforming this mechanism into a ranking problem of salient entity regions, we introduce prior visual attention cues to build a new salient object ranking dataset, and propose a salient object ranking (SaOR) network aimed at providing depth perception for prosthetic vision. Furthermore, we propose a SaOR-guided image description method to align with human observation patterns, toward providing additional visual information by auditory feedback. Finally, the integration of the two aforementioned algorithms constitutes an audiovisual cognitive optimization strategy for prosthetic vision.

Main results: Through conducting psychophysical experiments based on scene description tasks under simulated prosthetic vision, we verify that the SaOR method improves the subjects' performance in terms of object identification and understanding the correlation among objects. Additionally, the cognitive optimization strategy incorporating image description further enhances their prosthetic visual cognition.

Significance: This offers valuable technical insights for designing next-generation intelligent visual prostheses and establishes a theoretical groundwork for developing their visual information processing strategies. Code will be made publicly available.

Objective: Neuroprostheses typically operate under supervised learning, in which a machine-learning algorithm is trained to correlate neural or myoelectric activity with an individual's motor intent. Due to the stochastic nature of neuromyoelectric signals, algorithm performance decays over time. This decay is accelerated when attempting to regress proportional control of multiple joints in parallel, compared with the more typical classification-based pattern recognition control. To overcome this degradation, neuroprostheses and commercial myoelectric prostheses are often recalibrated and retrained frequently so that only the most recent, up-to-date data influences the algorithm performance. Here, we introduce and validate an alternative training paradigm in which training data from past calibrations is aggregated and reused in future calibrations for regression control. Approach: Using a cohort of four transradial amputees implanted with intramuscular electromyographic recording leads, we demonstrate that aggregating prior datasets improves prosthetic regression-based control in offline analyses and an online human-in-the-loop task. In offline analyses, we compared the performance of a convolutional neural network (CNN) and a modified Kalman filter (MKF) to simultaneously regress the kinematics of an eight-degree-of-freedom prosthesis. Both algorithms were trained under the traditional paradigm using a single dataset, as well as under the new paradigm using aggregated datasets from the past five or ten trainings. Main Results: Dataset aggregation reduced the root-mean-squared error of algorithm estimates for both the CNN and MKF, although the CNN saw a greater reduction in error. Further offline analyses revealed that dataset aggregation improved CNN robustness when reusing the same algorithm on subsequent test days, as indicated by a smaller increase in RMSE per day. Finally, data from an online virtual-target-touching task with one amputee showed significantly better real-time prosthetic control when using aggregated training data from just two prior datasets. Significance: Altogether, these results demonstrate that training data from past calibrations should not be discarded but, rather, should be reused in an aggregated training dataset such that the increased amount and diversity of data improve algorithm performance. More broadly, this work supports a paradigm shift for the field of neuroprostheses away from daily data recalibration for linear classification models and towards daily data aggregation for non-linear regression models.

Objective Neuronal oscillatory patterns are believed to underpin multiple cognitive mechanisms. Accordingly, compromised oscillatory dynamics were shown to be associated with neuropsychiatric conditions. Therefore, the possibility of modulating, or controlling, oscillatory components of brain activity as a therapeutic approach has emerged. Typical non-invasive brain-computer interfaces (BCI) based on EEG have been used to decode volitional motor brain signals for interaction with external devices. Here we aimed at feedback through visual stimulation which returns directly back to the visual cortex. Approach Our architecture permits the implementation of feedback control-loops capable of controlling, or at least modulating, visual cortical activity. As this type of neurofeedback depends on early visual cortical activity, mainly driven by external stimulation it is called non-volitional or implicit neurofeedback. Because retino-cortical 40-100ms delays in the feedback loop severely degrade controller performance, we implemented a predictive control system, called a Smith-Predictor (SP) controller, which compensates for fixed delays in the control loop by building an internal model of the system to be controlled, in this case the EEG response to stimuli in the visual cortex. Main Results Response models were obtained by analyzing, EEG data (n=8) of experiments using periodically inverting stimuli causing prominent parieto-occipital oscillations, the Steady-State Visual Evoked Potentials (SSVEPs). Averaged subject-specific SSVEPs, and associated retina-cortical delays, were subsequently used to obtain the SP controler's Linear, Time-Invariant (LTI) models of individual responses. The SSVEP models were first successfully validated against the experimental data. When placed in closed loop with the designed SP controller configuration, the SSVEP amplitude level oscillated around several reference values, accounting for inter-individual variability. Significance In silico and in vivo data matched, suggesting model's robustness, paving the way for the experimental validation of this non-volitional neurofeedback system to control the amplitude of abnormal brain oscillations in autism and attention and hyperactivity deficits. .

Objective.Phrenic nerve stimulation reduces ventilator-induced-diaphragmatic-dysfunction, which is a potential complication of mechanical ventilation. Electromagnetic simulations provide valuable information about the effects of the stimulation and are used to determine appropriate stimulation parameters and evaluate possible co-activation.Approach.Using a multiscale approach, we built a novel detailed anatomical model of the neck and the phrenic nerve. The model consisted of a macroscale volume conduction model of the neck with 13 tissues, a mesoscale volume conduction model of the phrenic nerve with three tissues, and a microscale biophysiological model of axons with diameters ranging from 5 to 14 µm based on the McIntyre-Richardson-Grill-model for myelinated axons. This multiscale model was used to quantify activation thresholds of phrenic nerve fibers using different stimulation pulse parameters (pulse width, interphase delay, asymmetry of biphasic pulses, pulse polarity, and rise time) during non-invasive electrical stimulation. Electric field strength was used to evaluate co-activation of the other nerves in the neck.Main results.For monophasic pulses with a pulse width of 150 µs, the activation threshold depended on the fiber diameter and ranged from 20 to 156 mA, with highest activation threshold for the smallest fiber diameter. The relationship was approximated using a power fit functionx^-3. Biphasic (symmetric) pulses increased the activation threshold by 25 to 30 %. The use of asymmetric biphasic pulses or an interphase delay lowered the threshold close to the monophasic threshold. Possible co-activated nerves were the more superficial nerves and included the transverse cervical nerve, the supraclavicular nerve, the great auricular nerve, the cervical plexus, the brachial plexus, and the long thoracic nerve.Significance.Our multiscale model and electromagnetic simulations provided insight into phrenic nerve activation and possible co-activation by non-invasive electrical stimulation and provided guidance on the use of stimulation pulse types with minimal activation threshold.

Objective: Kilohertz (kHz) frequency stimulation has gained attention as a neuromodulation therapy in spinal cord and in peripheral nerve block applications, mainly for treating chronic pain. Yet, few studies have investigated the effects of high-frequency stimulation on the performance of the electrode materials. In this work, we assess the electrochemical characteristics and stability of sputtered iridium oxide film (SIROF) microelectrodes under kHz frequency pulsed electrical stimulation.

Approach: SIROF microelectrodes were subjected to 1.5-10 kHz pulsing at charge densities of 250-1000 µC cm^-2(25-100 nC phase^-1), under monopolar and bipolar configurations, in buffered saline solution. The electrochemical behavior as well as the long-term stability of the pulsed electrodes was evaluated by voltage transient, cyclic voltammetry, and electrochemical impedance spectroscopy measurements.

Main results: Electrode polarization was more pronounced at higher stimulation frequencies in both monopolar and bipolar configurations. Bipolar stimulation resulted in an overall higher level of polarization than monopolar stimulation with the same parameters. In all tested pulsing conditions, except one, the maximum cathodal and anodal potential excursions stayed within the water window of iridium oxide (-0.6 to 0.8 V vs Ag|AgCl). Additionally, these SIROF microelectrodes showed little or no changes in the electrochemical performance under continuous current pulsing at frequencies up to 10 kHz for more than 10⁹pulses.

Significance: Our results suggest that 10,000 μm²SIROF microelectrodes can deliver high-frequency neural stimulation up to 10 kHz in buffered saline at charge densities between 250 and 1000 µC cm^-2(25-100 nC phase^-1).

Objective: Low-intensity transcranial focused ultrasound (tFUS) has emerged as a powerful neuromodulation tool characterized by its deep penetration and precise spatial targeting to influence neural activity. Our study directed low-intensity tFUS stimulation onto a region of prefrontal cortex (the frontal eye field, or FEF) of a rhesus macaque to examine its impact on a remote site, the extrastriate visual cortex (area V4) through this top-down modulatory circuit that has been studied extensively with electrical microstimulation.

Approach: To measure the impact of tFUS stimulation, we recorded local field potentials (LFPs) and multi-unit spiking activities from a multi-electrode array implanted in the visual cortex. To deliver tFUS stimulation, we leveraged a customized 128-element random array ultrasound transducer with precise spatial targeting.

Main results: We observed that tFUS stimulation in FEF produced modulation of V4 neuronal activity, either through enhancement or suppression, dependent on the pulse repetition frequency of the tFUS stimulation. Electronically steering the transcranial ultrasound focus through the targeted FEF cortical region produced changes in the level of modulation, indicating that the tFUS stimulation was spatially targeted within FEF. Modulation of V4 activity was confined to specific frequency bands, and this modulation was dependent on the presence or absence of a visual stimulus during tFUS stimulation. A control study targeting the insula produced no effect, emphasizing the region-specific nature of tFUS neuromodulation.

Significance: Our findings shed light on the capacity of tFUS to modulate specific neural pathways and provide a comprehensive understanding of its potential applications for neuromodulation within brain networks.

Electroencephalogram (EEG) signals exhibit multi-domain features, and electrode distributions follow non-Euclidean topology. To fully resolve the EEG signals, this study proposes a Temporal-Frequency-Spatial multi-domain feature fusion Graph Attention Network (TFSGAT) for motor imagery (MI) intention recognition in spinal cord injury (SCI) patients. The proposed model uses phase-locked value (PLV) to extract spatial phase connectivity information between EEG channels and continuous wavelet transform to extract valid EEG information in the time-frequency domain. It then models a graph data structure containing multi-domain information. The gated recurrent unit and GAT learn EEG's dynamic temporal-spatial information. Finally, the fully connected layer outputs the MI intention recognition results. After 10 times 10-fold cross-validation, the proposed model can achieve an average accuracy of 95.82%. Furthermore, this study analyzes the Event-Related Desynchronization/Event-Related Synchronization and PLV brain network to explore the brain activity of SCI patients during MI. This study confirms the potential of the proposed model in terms of EEG decoding performance and provides a reference for the mechanism of neural activity in SCI patients.

Objective: Brain-computer interfaces (BCI) are promising for severe neurological conditions and there are ongoing efforts to develop state-of-the-art neural interfaces, hardware, and software tools. We tested the potential of novel reduced graphene oxide (rGO) electrodes implanted epidurally over the hind limb representation of the primary somatosensory (S1) cortex of rats and compared them to commercial platinum-iridium (Pt-Ir) 16-channel electrodes (active site diameter: 25 μm).

Approach: Motor and somatosensory information was decoded offline from microelectrocorticography (μECoG) signals recorded while unrestrained rats performed a simple behavioral task: pressing a lever and the subsequent vibrotactile stimulation of the glabrous skin at three displacement amplitude levels and at two sinusoidal frequencies. μECoG data were initially analyzed by standard time-frequency methods. Next, signal powers of oscillatory bands recorded from multiple electrode channels were used as features for sensorimotor classification by a machine learning algorithm.

Main results: Both electrode types performed quite well and similar to each other for predicting the motor interval and the presence of the vibrotactile stimulus. Average accuracies were relatively lower for predicting 3-class vibrotactile frequency and 4-class amplitude level by both electrode types.

Significance: Given some confounding factors during the free movement of rats, the results show that both sensory and motor information can be recorded reliably from the hind limb area of S1 cortex by using μECoG arrays. The chronic use of novel rGO electrodes was demonstrated successfully. The hind limb area may be convenient for the future evaluation of new tools in neurotechnology, especially those for bidirectional BCIs.

Objective.Finger dexterity, and finger individuation in particular, is crucial for human movement, and disruptions due to brain injury can significantly impact quality of life. Understanding the neurological mechanisms responsible for recovery is vital for effective neurorehabilitation. This study explores the role of two key pathways in finger individuation: the corticospinal (CS) tract from the primary motor cortex and premotor areas, and the subcortical reticulospinal (RS) tract from the brainstem. We aimed to investigate how the cortical-reticular network reorganizes to aid recovery of finger dexterity following lesions in these areas.Approach.To provide a potential biologically plausible answer to this question, we developed an artificial neural network (ANN) to model the interaction between a premotor planning layer, a cortical layer with excitatory and inhibitory CS outputs, and RS outputs controlling finger movements. The ANN was trained to simulate normal finger individuation and strength. A simulated stroke was then applied to the CS area, RS area, or both, and the recovery of finger dexterity was analyzed.Main results.In the intact model, the ANN demonstrated a near-linear relationship between the forces of instructed and uninstructed fingers, resembling human individuation patterns. Post-stroke simulations revealed that lesions in both CS and RS regions led to increased unintended force in uninstructed fingers, immediate weakening of instructed fingers, improved control during early recovery, and increased neural plasticity. Lesions in the CS region alone significantly impaired individuation, while RS lesions affected strength and to a lesser extent, individuation. The model also predicted the impact of stroke severity on finger individuation, highlighting the combined effects of CS and RS lesions.Significance.This model provides insights into the interactive role of cortical and subcortical regions in finger individuation. It suggests that recovery mechanisms involve reorganization of these networks, which may inform neurorehabilitation strategies.

Objective.Deep learning is increasingly permeating neuroscience, leading to a rise in signal-processing applications for extracellular recordings. These signals capture the activity of small neuronal populations, necessitating 'spike sorting' to assign action potentials (spikes) to their underlying neurons. With the rise in publications delving into new methodologies and techniques for deep learning-based spike sorting, it is crucial to synthesise these findings critically. This survey provides an in-depth evaluation of the approaches, methodologies and outcomes presented in recent articles, shedding light on the current state-of-the-art.Approach.Twenty-four articles published until December 2023 on deep learning-based spike sorting have been examined. The proposed methods are divided into three sub-problems of spike sorting: spike detection, feature extraction and classification. Moreover, integrated systems, i.e. models that detect spikes and extract features or do classification within a single network, are included.Main results.Although most algorithms have been developed for single-channel recordings, models utilising multi-channel data have already shown promising results, with efficient hardware implementations running quantised models on application-specific integrated circuits and field programmable gate arrays. Convolutional neural networks have been used extensively for spike detection and classification as the data can be processed spatiotemporally while maintaining low-parameter models and increasing generalisation and efficiency. Autoencoders have been mainly utilised for dimensionality reduction, enabling subsequent clustering with standard methods. Also, integrated systems have shown great potential in solving the spike sorting problem from end to end.Significance.This survey explores recent articles on deep learning-based spike sorting and highlights the capabilities of deep neural networks in overcoming associated challenges, but also highlights potential biases of certain models. Serving as a resource for both newcomers and seasoned researchers in the field, this work provides insights into the latest advancements and may inspire future model development.