Journal of neural engineering最新文献

Objective.When a person listens to continuous speech, a corresponding response is elicited in the brain and can be recorded using electroencephalography (EEG). Linear models are presently used to relate the EEG recording to the corresponding speech signal. The ability of linear models to find a mapping between these two signals is used as a measure of neural tracking of speech. Such models are limited as they assume linearity in the EEG-speech relationship, which omits the nonlinear dynamics of the brain. As an alternative, deep learning models have recently been used to relate EEG to continuous speech.Approach.This paper reviews and comments on deep-learning-based studies that relate EEG to continuous speech in single- or multiple-speakers paradigms. We point out recurrent methodological pitfalls and the need for a standard benchmark of model analysis.Main results.We gathered 29 studies. The main methodological issues we found are biased cross-validations, data leakage leading to over-fitted models, or disproportionate data size compared to the model's complexity. In addition, we address requirements for a standard benchmark model analysis, such as public datasets, common evaluation metrics, and good practices for the match-mismatch task.Significance.We present a review paper summarizing the main deep-learning-based studies that relate EEG to speech while addressing methodological pitfalls and important considerations for this newly expanding field. Our study is particularly relevant given the growing application of deep learning in EEG-speech decoding.

Objective.Microfluidic devices interfaced with microelectrode arrays have in recent years emerged as powerful platforms for studying and manipulatingin vitroneuronal networks at the micro- and mesoscale. By segregating neuronal populations using microchannels only permissible to axons, neuronal networks can be designed to mimic the highly organized, modular topology of neuronal assemblies in the brain. However, little is known about how the underlying topological features of such engineered neuronal networks contribute to their functional profile. To start addressing this question, a key parameter is control of afferent or efferent connectivity within the network.Approach.In this study, we show that a microfluidic device featuring axon guiding channels with geometrical constraints inspired by a Tesla valve effectively promotes unidirectional axonal outgrowth between neuronal nodes, thereby enabling us to control afferent connectivity.Main results.Our results moreover indicate that these networks exhibit a more efficient network organization with higher modularity compared to single nodal controls. We verified this by applying designer viral tools to fluorescently label the neurons to visualize the structure of the networks, combined with extracellular electrophysiological recordings using embedded nanoporous microelectrodes to study the functional dynamics of these networks during maturation. We furthermore show that electrical stimulations of the networks induce signals selectively transmitted in a feedforward fashion between the neuronal populations.Significance.A key advantage with our microdevice is the ability to longitudinally study and manipulate both the structure and function of neuronal networks with high accuracy. This model system has the potential to provide novel insights into the development, topological organization, and neuroplasticity mechanisms of neuronal assemblies at the micro- and mesoscale in healthy and perturbed conditions.

Objective. The compromise of the hippocampal loop is a hallmark of mesial temporal lobe epilepsy (MTLE), the most frequent epileptic syndrome in the adult population and the most often refractory to medical therapy. Hippocampal sclerosis is found in >50% of drug-refractory MTLE patients and primarily involves the CA1, consequently disrupting the hippocampal output to the entorhinal cortex (EC). Closed-loop deep brain stimulation is the latest frontier to improve drug-refractory MTLE; however, current approaches do not restore the functional connectivity of the hippocampal loop, they are designed by trial-and-error and heavily rely on seizure detection or prediction algorithms. The objective of this study is to evaluate the anti-ictogenic efficacy and robustness of an artificial bridge restoring the dialog between hippocampus and EC.Approach. In mouse hippocampus-EC slices treated with 4-aminopyridine and in which the Schaffer Collaterals are severed, we established an artificial bridge between hippocampus and EC wherein interictal discharges originating in the CA3 triggered stimulation of the subiculum so to entrain EC networks. Combining quantification of ictal activity with tools from information theory, we addressed the efficacy of the bridge in controlling ictogenesis and in restoring the functional connectivity of the hippocampal loop.Main results. The bridge significantly decreased or even prevented ictal activity and proved robust to failure; when operating at 100% of its efficiency (i.e., delivering a pulse upon each interictal event), it recovered the functional connectivity of the hippocampal loop to a degree similar to what measured in the intact circuitry. The efficacy and robustness of the bridge stem in mirroring the adaptive properties of the CA3, which acts as biological neuromodulator.Significance. This work is the first stepping stone toward a paradigm shift in the conceptual design of stimulation devices for epilepsy treatment, from function control to functional restoration of the salient brain circuits.

Objective.Studying motor units is essential for understanding motor control, the detection of neuromuscular disorders and the control of human-machine interfaces. Individual motor unit firings are currently identifiedin vivoby decomposing electromyographic (EMG) signals. Due to our body's properties and anatomy, individual motor units can only be separated to a limited extent with surface EMG. Unlike electrical signals, magnetic fields do not interact with human tissues. This physical property and the emerging technology of quantum sensors make magnetomyography (MMG) a highly promising methodology. However, the full potential of MMG to study neuromuscular physiology has not yet been explored.Approach.In this work, we performin silicotrials that combine a biophysical model of EMG and MMG with state-of-the-art algorithms for the decomposition of motor units. This allows the prediction of an upper-bound for the motor unit decomposition accuracy.Main results.It is shown that non-invasive high-density MMG data is superior over comparable high-density surface EMG data for the robust identification of the discharge patterns of individual motor units. Decomposing MMG instead of EMG increased the number of identifiable motor units by 76%. Notably, MMG exhibits a less pronounced bias to detect superficial motor units.Significance.The presented simulations provide insights into methods to study the neuromuscular system non-invasively andin vivothat would not be easily feasible by other means. Hence, this study provides guidance for the development of novel biomedical technologies.

Objective.Neural decoding, an important area of neural engineering, helps to link neural activity to behavior. Deep neural networks (DNNs), which are becoming increasingly popular in many application fields of machine learning, show promising performance in neural decoding compared to traditional neural decoding methods. Various neural decoding applications, such as brain computer interface applications, require both high decoding accuracy and real-time decoding speed. Pruning methods are used to produce compact DNN models for faster computational speed. Greedy inter-layer order with Random Selection (GRS) is a recently-designed structured pruning method that derives compact DNN models for calcium-imaging-based neural decoding. Although GRS has advantages in terms of detailed structure analysis and consideration of both learned information and model structure during the pruning process, the method is very computationally intensive, and is not feasible when large-scale DNN models need to be pruned within typical constraints on time and computational resources. Large-scale DNN models arise in neural decoding when large numbers of neurons are involved. In this paper, we build on GRS to develop a new structured pruning algorithm called jump GRS (JGRS) that is designed to efficiently compress large-scale DNN models.Approach.On top of GRS, JGRS implements a 'jump mechanism', which bypasses retraining intermediate models when model accuracy is relatively less sensitive to pruning operations. Design of the jump mechanism is motivated by identifying different phases of the structured pruning process, where retraining can be done infrequently in earlier phases without sacrificing accuracy. The jump mechanism helps to significantly speed up execution of the pruning process and greatly enhance its scalability. We compare the pruning performance and speed of JGRS and GRS with extensive experiments in the context of neural decoding.Main results.Our results demonstrate that JGRS provides significantly faster pruning speed compared to GRS, and at the same time, JGRS provides pruned models that are similarly compact as those generated by GRS.Significance.In our experiments, we demonstrate that JGRS achieves on average 9%-20% more compressed models compared to GRS with 2-8 times faster speed (less time required for pruning) across four different initial models on a relevant dataset for neural data analysis.

Objective.In studying the spatial coding mechanism of visual evoked potentials, it is significant to construct a model that shows the relationship between steady-state visual evoked potential (SSVEP) responses to the local and global visual field stimulation. In order to investigate whether SSVEPs produced by sub-region stimulation can predict that produced by joint region stimulation, a sub-region combination scheme for spatial coding in a high-frequency SSVEP-based brain-computer interface (BCI) is developed innovatively.Approach.An annular visual field is divided equally into eight sub-regions. The 60 Hz visual stimuli in different sub-regions and joint regions are presented separately to participants. The SSVEP produced by the sub-region stimulation is superimposed to simulate the SSVEP produced by the joint region stimulation with different spatial combinations. A four-class spatially-coded BCI paradigm is used to evaluate the simulated classification performance, and the performance ranking of all simulated SSVEPs is obtained. Six representative stimulus patterns from two performance levels and three stimulus areas are applied to the online BCI system for each participant.Main results.The experimental result shows that the proposed scheme can implement a spatially-coded visual BCI system and realize satisfactory performance with imperceptible flicker. Offline analysis indicates that the classification accuracy and information transfer rate (ITR) are 89.69 ± 8.75% and 24.35 ± 7.09 bits min^-1with 3 s data length under the 3/8 stimulus area. The online BCI system reaches an average classification accuracy of 87.50 ± 9.13% with 3 s data length, resulting in an ITR of 22.48 ± 6.71 bits min^-1under the 3/8 stimulus area.Significance.This study proves the feasibility of using the sub-region's response to predict the joint region's response. It has the potential to extend to other frequency bands and lays a foundation for future research on more complex spatial coding methods.

Objective. Brain-computer interfaces can restore various forms of communication in paralyzed patients who have lost their ability to articulate intelligible speech. This study aimed to demonstrate the feasibility of closed-loop synthesis of artificial speech sounds from human cortical surface recordings during silent speech production.Approach. Ten participants with intractable epilepsy were temporarily implanted with intracranial electrode arrays over cortical surfaces. A decoding model that predicted audible outputs directly from patient-specific neural feature inputs was trained during overt word reading and immediately tested with overt, mimed and imagined word reading. Predicted outputs were later assessed objectively against corresponding voice recordings and subjectively through human perceptual judgments.Main results. Artificial speech sounds were successfully synthesized during overt and mimed utterances by two participants with some coverage of the precentral gyrus. About a third of these sounds were correctly identified by naïve listeners in two-alternative forced-choice tasks. A similar outcome could not be achieved during imagined utterances by any of the participants. However, neural feature contribution analyses suggested the presence of exploitable activation patterns during imagined speech in the postcentral gyrus and the superior temporal gyrus. In future work, a more comprehensive coverage of cortical surfaces, including posterior parts of the middle frontal gyrus and the inferior frontal gyrus, could improve synthesis performance during imagined speech.Significance.As the field of speech neuroprostheses is rapidly moving toward clinical trials, this study addressed important considerations about task instructions and brain coverage when conducting research on silent speech with non-target participants.

Objective.Ultrafast ultrasound (UUS) imaging has been used to detect intramuscular mechanical dynamics associated with single motor units (MUs). Detecting MUs from ultrasound sequences requires decomposing a velocity field into components, each consisting of an image and a signal. These components can be associated with putative MU activity or spurious movements (noise). The differentiation between putative MUs and noise has been accomplished by comparing the signals with MU firings obtained from needle electromyography (EMG). Here, we examined whether the repeatability of the images over brief time intervals can serve as a criterion for distinguishing putative MUs from noise in low-force isometric contractions.Approach.UUS images and high-density surface EMG (HDsEMG) were recorded simultaneously from 99 MUs in the biceps brachii of five healthy subjects. The MUs identified through HDsEMG decomposition were used as a reference to assess the outcomes of the ultrasound-based components. For each contraction, velocity sequences from the same eight-second ultrasound recording were separated into consecutive two-second epochs and decomposed. To evaluate the repeatability of components' images across epochs, we calculated the Jaccard similarity coefficient (JSC). JSC compares the similarity between two images providing values between 0 and 1. Finally, the association between the components and the MUs from HDsEMG was assessed.Main results.All the MU-matched components had JSC > 0.38, indicating they were repeatable and accounted for about one-third of the HDsEMG-detected MUs (1.8 ± 1.6 matches over 4.9 ± 1.8 MUs). The repeatable components (JSC > 0.38) represented 14% of the total components (6.5 ± 3.3 components). These findings align with our hypothesis that intra-sequence repeatability can differentiate putative MUs from noise and can be used for data reduction.Significance.This study provides the foundation for developing stand-alone methods to identify MU in UUS sequences and towards real-time imaging of MUs. These methods are relevant for studying muscle neuromechanics and designing novel neural interfaces.

Micro magnetic stimulation of the brain via implantable micro-coils is a promising novel technology for neuromodulation. Careful consideration of the thermodynamic profile of such devices is necessary for effective and safe designs.Objective.We seek to quantify the thermal profile of bent wire micro-coils in order to understand and mitigate thermal impacts of micro-coil stimulation.Approach. In this study, we use fine wire thermocouples and COMSOL finite element modeling to examine the profile of the thermal gradients generated near bent wire micro-coils submerged in a water bath during stimulation. We tested a range of stimulation parameters previously reported in the literature such as voltage amplitude, stimulus frequency, stimulus repetition rate and coil wire materials.Main results. We found temperature increases ranging from <1 °C to 8.4 °C depending upon the stimulation parameters tested and coil wire materials used. Numerical modeling of the thermodynamics identified hot spots of the highest temperatures along the micro-coil contributing to the thermal gradients and demonstrated that these thermal gradients can be mitigated by the choice of wire conductor material and construction geometry.Significance. ISO standard 14708-1 designates a thermal safety limit of 2 °C temperature increase for active implantable medical devices. By switching the coil wire material from platinum/iridium to gold, our study achieved a 5-6-fold decrease in the thermal impact of coil stimulation. The thermal gradients generated from the gold wire coil were measured below the 2 °C safety limit for all stimulation parameters tested.