搜索结果：Journal of neural engineering

Extensive research over the past two decades has focused on identifying a preictal period in scalp and intracranial encephalography (iEEG). This effort has led to numerous seizure prediction and forecasting algorithms, with moderate success on datasets consisting of curated and pre-segmented EEG. When evaluated pseudo-prospectively on continuous EEG recordings, existing algorithms often exhibit low sensitivity, high time in warning, or both. In this study, we investigate whether predictive modeling of temporal dynamics of iEEG features and seizure risk can improve pseudo-prospective forecasting performance.

Approach: Using iEEG data from n = 5 patients undergoing presurgical evaluation at the Hospital of the University of Pennsylvania and six state-of-the-art baseline models, we shift the focus from designing new features and classifiers to modeling temporal evolution of iEEG features (classifier inputs) and seizure risk (classifier outputs). We develop autoregressive models to predict iEEG features and seizure risk over timescales of several minutes and incorporate these predictions into existing forecasting pipelines.

Main results: We first demonstrate that a wide range of iEEG features are predictable over time, with over 99% and 35% of features achieving R2> 0 for 10-second and 10-minute-ahead predictions (mean R2> of 0.85 and 0.2), respectively. We observe a strong correlation between feature predictability and classification-based feature importance. Accordingly, we show that incorporating an autoregressive model that predicts iEEG features approximately 12 ± 4 minutes into the future improves pseudo-prospective performance, with a mean increase of 28% in the area under the sensitivity versus time-in-warning curve (PP-AUC). The addition of a second autoregressive model at the level of seizure risk yields further gains, resulting in a total mean improvement of 51% in PP-AUC.

Significance: These results provide evidence for long-term predictability of seizure-relevant iEEG features and demonstrate the value of time-series predictive modeling for improving seizure forecasting from continuous intracranial EEG.

To address the gap in quantitatively modeling dynamic failure mechanisms for Gasifier lock bucket valve system reliability, this study proposes an innovative method: using backpropagation (BP) neural network to optimize the prior data of dynamic Bayesian network (DBN). Firstly, based on the empirical formula for the number of hidden layer neurons, the original DBN model of the system is adapted to a structurally adaptive BP neural network to calibrate its prior parameters，and the correspondence between the prior distribution of DBN and the input-output functions of the BP network is established. Subsequently, utilizing the core characteristics of BP network, iterative optimization of DBN prior data is achieved through continuous learning of the operating performance of the lock bucket valve system. Next, the optimized DBN model is subjected to dynamic system reliability evaluation using bidirectional inference analysis. The results show that in the positive prediction, the reliability of the system after 300 hours of operation without considering maintenance is only 0.047, which can be improved to 0.302 after incorporating maintenance factors. The reliability of the optimized system is lower than before optimization, and the gap gradually widens over time. Reverse reasoning clearly identifies the weak links in the system as high-pressure coal powder flushing, adhesion between ball seats, internal deformation and wear. Targeted preventive measures can improve the reliability of the system and extend its service life.

Chronic kidney disease (CKD) risk assessment is shifting from centralized instrument-heavy testing to personalized point-of-care evaluation enabled by portable platforms. However, integration of multibiomarker analysis and the deep learning algorithm to improve detection accuracy in CKD is still a persistent goal. Herein, this work has developed an artificial intelligence (AI)-empowered and bio-/nanoenzyme-hybrid multisensors array (AI-BMA) for multidimensional precise diagnosis of early stage CKD. The laser-induced electrochemical sensors array is functionalized by creatinine deaminase (CDI), urate oxidase (UOx), and polyaniline for the differential detection of urinary creatinine (Cr), uric acid (UA), and pH. The integrated analysis of multimodal/multimarker electrochemical characteristic spectra and one-dimensional convolutional neural network (1D-CNN) along with multilayer perceptron (MLP) establishes an end-to-end workflow from electrochemical signal acquisition to individualized CKD risk assessment. The proposed bio-/nanoenzyme-hybrid multiplexed sensors strategy demonstrates well-defined analytical performance, covering detection ranges of 3-15 mM creatinine with limit of detection of 300.50 μM, 0.1-1.0 mM uric acid with LOD value of 19.17 μM, and 3.0-9.0 pH. By employing the self-developed 1D-CNN and MLP model for multimarker joint prediction, the average prediction accuracy of CKD biomarkers reaches 98.67%. This overcomes limitations of high-dimensional electrochemical signal feature extraction and multi-index joint prediction. The robust AI-BMA platform can automatically convert complex electrochemical detection data of urinary metabolites into understandable risk stratification results. This provides an alternative solution for the early screening of kidney injury, which is expected to assist patients/clinicians in identifying CKD's risk assessment in the home-care scenario and limited resources.

Deep neural network (DNN)-based noise reduction has emerged as a promising advancement in hearing aid signal processing as a means to improve speech intelligibility in noisy environments for hearing aid wearers. The aim of this study was to investigate the impact of the DNN on speech intelligibility in speech-shaped noise (SSN) and multitalker babble (MTB) when the target speech was coming from the front and from the side. Subjective ratings of clarity, total impression, listening effort, and background noise awareness were also collected. Twenty adult participants with mild to moderately severe sensorineural hearing loss were fitted with hearing aids from a single manufacturer, programmed with four different settings that varied across combinations of microphone directionality (omnidirectional and directional beamforming) and noise reduction (off, traditional, DNN). Results showed that DNN, when combined with beamforming, consistently outperformed the other programs across all metrics. Outcomes were influenced by both noise type and spatial configuration. DNN was more effective in SSN than MTB. Beamforming was especially beneficial when the target speech came from the front. Listening in programs that included both DNN and beamforming together resulted in additional benefits shown in the outcome measures, most likely due to the beamforming improving the signal-to-noise ratio and providing a cleaner signal for the DNN to work with.

Interferential stimulation uses multiple independent groups of electrodes to apply high-frequency currents with a small-frequency offset. At a specific region(s) in space, superposition of the high-frequency currents creates low-frequency amplitude modulation that can drive neural activation. Therefore, with interferential spinal cord stimulation (IF-SCS), it may be possible to focus stimulation on target areas while avoiding stimulation of non-target areas that could produce unwanted side effects. In this study, we used a comprehensive computational modeling approach to evaluate the potential efficacy of IF-SCS to improve targeting within the spinal cord. 

Approach. We constructed a finite element method model of the human lower thoracic spinal cord and surrounding anatomy with two eight-contact percutaneous electrode arrays in the epidural tissue and calculated the extracellular potentials generated during IF-SCS. We applied these potential fields to multi-compartment axon models distributed throughout the spinal cord to simulate the neural response to IF-SCS. We examined how various factors, such as stimulation configuration, carrier and beat frequencies, electrode spacing, and dorsal cerebrospinal fluid (CSF) thickness, affected the neural response to IF-SCS. 

Main Results. IF-SCS produced different types of axonal responses, such as phasic, tonic, and quiescent. Phasic activation thresholds increased with increasing carrier and beat frequencies, electrode spacing, and dorsal CSF thickness. As we increased the stimulation amplitude, we observed that deeper regions of the dorsal columns exhibited phasic responses. Finally, a comparison of frequency-dependent and frequency-independent tissue properties revealed only minor differences in activation thresholds.

Significance. Our results demonstrate that several factors affect the spatial selectivity and neural response to IF-SCS. This computational modeling study highlights the potential for IF-SCS to improve targeting within the spinal cord and supports its development into a clinically effective therapy that provides advantages over conventional spinal cord stimulation therapies.

Empathy development in early childhood involves complex neurocognitive processes, but the neural mechanisms underlying this development have yet to be fully explored. This study combined EEG microstate and spectral analyses to investigate empathy-related neural processes in 71 typically developing preschoolers (aged 3-6 years) during naturalistic viewing of socioemotional animations. Spectral analysis revealed increased occipital theta, centrotemporal alpha, and temporal beta activities during empathy-eliciting clips compared to baseline. Theta activity in response to Theory of Mind (ToM) clips negatively correlated with children's negative emotion understanding, while alpha activity during Emotional Contagion (EC) clips positively correlated with positive emotion resonance. Age-related decreases in occipital theta power and increases in centrotemporal alpha power were observed, paralleling behavioral improvements in prosocial helping behavior. Microstate analysis identified hierarchical age-related shifts: reduced coverage of microstate MS5 and increased transitions from MS5 to MS2 during ToM processing, implying that as children mature, perspective-taking efficiency increases, redirecting neural resources to response preparation. These findings suggest an age-related neural cascade framework where perceptual efficiency gains may free cognitive resources for mentalizing and emotion regulation, enabling more advanced prosocial actions. The present work suggests that EEG spectral and microstate metrics serve as sensitive neurophysiological indices for assessing early childhood age-related changes in empathy within naturalistic contexts.

Computational point-of-care (POC) sensors enable rapid, low-cost, and accessible diagnostics in emergency, remote, and resource-limited areas that lack access to centralized medical facilities. These systems can use neural network-based algorithms to accurately infer diagnoses from signals generated by rapid diagnostic tests or sensors. However, neural network-based diagnostic models are subject to hallucinations and can produce erroneous predictions, posing a risk of misdiagnosis and inaccurate clinical decisions. To address this challenge, here we present an autonomous uncertainty quantification technique developed for POC diagnostics. As our test bed, we used a paper-based, computational vertical flow assay (xVFA) platform developed for rapid POC diagnosis of Lyme disease, the most prevalent tick-borne disease globally. The xVFA platform integrates a disposable paper-based assay, a hand-held optical reader, and a neural network-based inference algorithm, providing rapid and cost-effective Lyme disease diagnostics in under 20 min using only 20 μL of patient serum. By incorporating a Monte Carlo dropout (MCDO)-based uncertainty quantification approach into the diagnostics pipeline with minimal computational and memory overhead, we identified and excluded erroneous predictions with high uncertainty, significantly improving the sensitivity and reliability of the xVFA in an autonomous manner, without access to the ground truth diagnostic information on patients. Blinded testing using new patient samples demonstrated an increase in diagnostic sensitivity from 88.2% to 95.7%, indicating the effectiveness of MCDO-based uncertainty quantification in enhancing the robustness of neural network-driven computational POC sensing systems.

Poststroke cognitive impairment (PSCI) is a prevalent complication of stroke, characterised by deficits in one or more cognitive domains (eg, memory, attention, executive function). Beyond increasing mortality and disability risks, PSCI frequently co-occurs with motor dysfunction, which impairs activities of daily living and reduces quality of life. Due to the complexity of neural networks involved in PSCI, clinical practice currently lacks targeted therapeutic strategies; existing interventions (eg, pharmacotherapy, traditional cognitive training) are limited in scope and variable in efficacy. Here, we developed an innovative dynamic cognitive training system integrated with virtual reality (VR) technology, based on principles of neuroplasticity and multisensory integration. This study aimed to explore the intervention effects of this system on cognitive function in patients with PSCI while incorporating exploratory neuroimaging assessments to provide descriptive and hypothesis-generating information regarding brain functional changes associated with the intervention. This single-centre, randomised controlled, evaluator-blinded clinical trial will assess the rehabilitative efficacy of VR-based cognitive training in patients with PSCI. A total of 60 patients who had a stroke will be enrolled and randomised to either a conventional rehabilitation group or a VR intervention group. The intervention will last 2 weeks, with five sessions of 60 min each training session per week. During the 60-minute training session, both groups will receive 30 min of conventional rehabilitation training. For the remaining 30 min, the control group will undergo traditional cognitive rehabilitation while the experimental group will be subjected to VR-based cognitive rehabilitation training. The primary outcome measure is the Montreal Cognitive Assessment; secondary outcomes include the Mini-Mental State Examination, Trail Making Test and Stroop Test. Assessments will be conducted at three time points: baseline (T0), immediately postintervention (T1) and 4 weeks after completing the intervention (T2). This study aims to evaluate the preliminary effectiveness of a VR-based intervention in improving multidimensional cognitive function, while incorporating exploratory neuroimaging outcomes to generate hypothesis-forming insights into potential neural correlates. The trial was approved by the Ethics Committee of Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine (2025-1933-273-02).The results will be submitted to a peer review journal or at a conference. ChiCTR2600116040.

Objective.Accurate simulations of electric fields (E-fields) in neural stimulation depend on tissue conductivity representations that link underlying microscopic tissue structure with macroscopic assumptions. Mesoscale conductivity variations can produce meaningful changes in E-fields and neural activation thresholds but remain largely absent from standard macroscopic models. Conductivity variations within the cortex are expected given the differences in cell density and volume fraction across layers. We sought to estimate layer-specific conductivity within the cortex using computational models.Approach.We review recent efforts modeling microscopic and mesoscopic E-fields and outline approaches that bridge micro- and macroscales to derive consistent mesoscale conductivity distributions. Using simplified microscopic models, effective tissue conductivity was estimated as a function of volume fraction of extracellular space, and the conductivities of different cortical layers were interpolated based on experimental volume fraction.Main results.The effective tissue conductivities were monotonically decreasing convex functions of the cell volume fraction. They followed the theoretical power formulas only for small volume fractions (<20%) and deviated to higher values for larger volume fractions. With decreasing cell volume fraction, the conductivity of cortical layers increased with depth from layer 2 to 6. Due to the high relative sensitivity, the conductivity difference between layers was considerably larger than their variation in volume fraction, e.g. with layers 3 and 6 being 5% and 24% more conductive than layer 2, respectively, despite their volume fraction being only 1.3% and 5% lower than that of the latter.Significance.The review and analysis provide a foundation for accurate multiscale models of E-fields and neural stimulation. Overcoming technical limitations that microscale and macroscale methods face will allow validation to arrive at consistent conductivity distributions. Using layer-specific conductivity values within the cortex could improve the accuracy of estimations of thresholds and distributions of neural activation in E-field models of brain stimulation.

Speech imagination is one of the most important research directions in the field of brain-computer interfaces. However, there is insufficient research on silent brain-computer interfaces based on the Chinese stimulating materials. An experimental paradigm for Chinese speech imagery, which features a distinctive&#xa0;initial-and-final&#xa0;structure, is designed in this paper. According to the characteristics of vocalization and structure, the collected EEG data can be organized into a multi-level tree structure. Compared to conventional multi-label classification, our paper aims to study how to effectively utilize hierarchical structural information in the multi-granularity hierarchical classification tasks. We propose a hierarchical capsule network based on bidirectional knowledge transfer by using multi-band feature matrix, which is tailor-made for the phonological structure of Mandarin Chinese. The adoption of capsule network as the primary architecture is mainly due to the dynamic routing mechanism that can naturally model hierarchical relationships in the syllable hierarchy. In addition, we introduce the bidirectional knowledge transfer strategy to further improve the classical dynamic routing. Specifically, features from coarse-grained levels are added to fine-grained levels to fully utilize the dependency information between levels. In order to mitigate error propagation in the forward learning process, we also employ reverse knowledge transfer constrained via soft labels. The hierarchical classification results and ablation experiments both demonstrate the effectiveness of our proposed algorithm. The highest recognition rates for each layer reach 90.86%, 73.69%, and 69.45%, respectively. This article offers a novel perspective for decoding&#xa0;hierarchical Chinese silent BCI paradigms. Our study not only reveals the potential of linguistic domain knowledge in guiding neural network architectures for task-specific applications, but also provides a robust foundation for future individual phoneme classification.

Objective.Conventional neural prostheses use brief charge-balanced pulses to minimize the risk of electrode polarization and irreversible electrochemistry at the electrode-electrolyte interface, constraining the waveforms that can be delivered long-term. Direct current (DC) stimulation has shown advantages, such as direct excitation and inhibition of tissue, but priorin vivoreports have been largely limited to acute durations. This study explores the physiological and histological effects of prolonged DC stimulation at amplitudes effective for neuromodulation while isolating tissue from the electrode-electrolyte interface.Approach. We developed a separated interface nerve electrode device for long-term ionic DC (iDC) stimulation in free-roaming rodents. Using our device, we applied 14 d of continuous iDC targeting the vestibular periphery in nine chinchillas. We measured vestibulo-ocular reflex responses to iDC before and after stimulation. Postmortem, we performed a histological comparison of stimulated ears and implanted but unstimulated controls.Main results.All chinchillas retained robust vestibular reflex function after 14 d of continuous iDC at up to 30&#xb5;A. Histological exam found no differences associated with iDC stimulation between the stimulated ears and the implanted but unstimulated ears.Significance.Decoupling the metal-electrolyte interface from the target tissue enables prolonged delivery of iDC at amplitudes effective for neuromodulation without causing a loss of physiological responses or histological signs of structural damage.

Kilohertz frequency waveforms have received increasing attention in the field of neuromodulation in recent years. These waveforms are frequently used to develop transcutaneous stimulation therapies, although they have also been used with implanted electrodes. While the goal is non- or minimally-invasive stimulation, an underappreciated aspect of modulated waveforms with kilohertz frequencies is the ability to remove the stimulation artifact with simple linear filters. These modulated sinusoids can be made using kilohertz frequency signals that do not overlap with the neural signal spectrum. In this work we deliver amplitude modulated kilohertz frequency waveforms directly into the brain via implanted electrodes. We refer to these waveforms as premodulated TI as they are created directly by the stimulator hardware. We performed both benchtop and in vivo experiments to study premodulated TI. Benchtop measurements investigated the role of intermodulation, a phenomenon that can generate phantom artifacts. In vivo experiments with parameter sweeps compared the evoked responses between premodulated TI and square pulses. Simulations in the NEURON environment were performed on an MRG axon model. We demonstrate that hardware-based intermodulation can be significantly reduced by using parallel stimulators and separate electrodes. We further identified another source of intermodulation, the neural amplifier. We showed that premodulated TI evokes similar neural response as conventional pulses in the single tested neural pathway. Simulation results mirrored the threshold results found in vivo. Together we present a method to stimulate neural tissue with significantly reduced hardware-based artifacts as compared to conventional pulse waveforms. This technique could open possibilities for studying direct neural responses during electrical stimulation in closed loop applications.

Positive aging, a concept found in positive psychology, serves as the theoretical foundation for this study. To age positively, one must manage hidden or unrecognized challenges, show flexibility in behavior and thought, adopt a positive outlook on problems involving regression, and make decisions that promote one's well-being. This study examined the role of wisdom and life purpose in the mental well-being of middle-aged and older adults. More specifically, we tested 4 hypotheses: wisdom would exhibit a positive correlation with mental well-being, quality of life would exhibit a positive correlation with mental well-being, meaning and purpose would exhibit a positive correlation with mental well-being, and freedom would exhibit a positive correlation with mental well-being. The research used a multianalytical methodology combining covariance-based structural equation modeling and artificial neural network techniques to analyze data from 377 individuals aged 50 to 102 years. Results from the covariance-based structural equation modeling indicate that meaning and purpose, wisdom, and quality of life were significantly associated with the mental well-being, accounting for 71% of the explained variance. Additionally, the artificial network analysis yielded exact forecasts of mental well-being. The artificial network model achieved an accuracy of 82.1% and 73% on the training and test sets, respectively, for predicting mental well-being. Sensitivity analysis revealed that meaning and purpose were the most critical factors in explaining participants' mental well-being. These findings have prominent theoretical implications for social psychology researchers and practical consequences for authorities involved in the care of older adults, who can use the results to develop strategic plans and take necessary actions.

High accuracy in medical classification tasks does not ensure that neural networks reason in ways consistent with clinical or neurobiological understanding. This study examines whether a Vision Transformer (ViT) trained on resting-state EEG infers cognitive impairment through physiologically meaningful mechanisms. A lightweight ViT was trained on multi-center resting-state EEG to detect mild cognitive impairment. The model's probabilistic outputs were interpreted as continuous cognitive risk scores. Knowledge distillation and spatial perturbation analyses were performed to identify the electrophysiological features and cortical regions underlying the model's predictions. The model achieved an average accuracy of 75.4% in five-fold cross-validation, and generalized to Alzheimer's disease cohorts and an external clinical center. The derived risk scores correlated with MoCA subdomains, particularly memory, language and orientation. Key drivers included increased autocorrelation, reduced Lempel-Ziv complexity and changes in power spectral density. Perturbation analyses highlighted strong contributions from the insular cortex and the transverse temporal regions. The model's decision process reflects physiologically and anatomically interpretable patterns consistent with clinical reasoning, supporting EEG-based modeling as an objective tool for quantifying cognitive function.

Modern neuroelectronic interfaces have shown great potential to diagnose conditions, address neurological dysfunction, and advance neuroscientific knowledge. However, neural interface systems today require tethered connections that restrict mobility, prevent testing across ecological contexts, and inhibit clinical translation to at-home use. Fully implantable commercial systems have previously been developed, but exhibit significant constraints, including limited modularity, low bandwidth, or unidirectional communication. We aimed to close this gap by developing a neuroelectronic interface that can be deployed flexibly with a variety of third-party neural probes. We have developed the Modular Bionic Interface (MBI), a system composed of a fully implantable device and a worn unit for high-bandwidth, bidirectional interfacing with the nervous system. The MBI can record high fidelity electrophysiological signals and deliver spatiotemporally modulated electrical stimulation for clinical and research purposes through flexible interaction with third party implantable devices. We performed benchtop evaluation to validate the recording and stimulation capabilities of the MBI across a diverse range of inputs and outputs. We then evaluated the MBI system in vivo through chronic implantation within a sheep, where results were stable for the length of evaluation, over six months. While connected to an actively powered, third-party high-resolution spinal cord stimulation electrode array, the MBI system was able to deliver stimulation to evoke lower extremity motor responses and record spinal compound action potentials evoked by peripheral nerve and spinal stimulation. We demonstrate a fully implantable system with a small footprint capable of high-resolution, bi-directional communication with the nervous system via modular connections to third-party devices. We expect that modular devices will further our ability to treat complex neurological disease and injury.

Objective.The human brain consists of multiple interacting neuronal networks that interleave at fine spatial and temporal scales. This complexity presents a challenge for scalp EEG and other noninvasive mapping techniques to accurately identify seizures and abnormal current patterns from background activity. To address this unmet need, this study investigates transcranial acoustoelectric brain imaging (tABI) with neuronavigation as a new method for mapping EEG-derived currents through the skull.Approach.In tABI, ultrasound (US) is focused and steered in the brain as surface electrodes record an acoustoelectric (AE) interaction signal. Space and time varying current maps are then generated at a resolution determined by the US focus. To test the efficacy of this method, a human skull was filled with conductive agarose gel, and a clinical depth electrode array was implanted 43 mm below the skull surface to generate artificial current waveform segments taken from normal and seizure activity. A 0.6 MHz 2D array was used to electronically focus and steer US through the skull while gold cup electrodes recorded high frequency AE signals and low frequency surface potentials. A 2D Wiener filter (WF) was introduced during preprocessing to enhance SNR followed by singular value decomposition (SVD) to selectively identify pixels correlated with different temporal patterns.Main results.Whereas the WF enhanced SNR up to 16.9 dB at 6.4 mA of current, SVD enabled color-coding of tABI to highlight activation patterns correlated with different current waveforms with a spatial resolution of 5 mm. Finally, the current detection limit depended on the duration and bandwidth of the selected currents with the ictal waveform yielding the lowest detection (28 &#xb5;A and 78 &#xb5;A cm-2*MPa, p < 0.05).Significance.These results support the development of tABI for noninvasive mapping of neuronal currents in epilepsy patients for surgical planning and other applications.

Complexity-based metrics have been applied to surface electromyography (sEMG) to characterize fatigue-related changes in the temporal structure of myoelectric signals beyond amplitude and spectral features. Optically pumped magnetometers (OPM) are sensors that enable non-invasive recordings of magnetomyographic (MMG) signals from skeletal muscle and are increasingly used to complement surface electromyography; however, it remains unclear whether complexity measures derived from magnetic recordings are comparable to those obtained from sEMG. Here, we directly compared fatigue-related dynamics of conventional and complexity-based signal features of sEMG and OPM-MMG measured from the biceps brachii during sustained elbow flexion. Healthy participants performed isometric contractions at 20% maximal voluntary contraction (MVC; 20 min) or 60% MVC (3 min). sEMG and OPM-based MMG were recorded simultaneously, and signal median frequency, root mean square (RMS), and Lempel-Ziv (LZ) complexity were calculated over time. Across contraction intensities, sEMG and MMG showed consistent fatigue-related changes, characterized by increasing RMS, decreasing median frequency, and a progressive decline in LZ over time. In addition, multiple regression analyses indicated that the decrease in LZ was not fully accounted for by concurrent amplitude or spectral changes, suggesting that complexity captures aspects of signal organization that are not fully explained by established features. Finally, while sEMG showed higher LZ complexity and median frequency at 60% compared to 20% MVC, corresponding intensitydependent effects were not observed in OPM-based MMG. These findings suggest that complexity-based metrics capture fatigue-related changes in neuromuscular signal organization beyond conventional measures, and that sEMG and OPM-based MMG provide similar, though modalityspecific, information. Together, the results support the use of complexity metrics in multimodal electrophysiological and biomagnetic assessments of neuromuscular fatigue.

Objective.PRIMA subretinal implants provide prosthetic vision to patients blinded by age-related macular degeneration, with acuity closely matching the sampling limit of the pixel pitch: a single 100&#x2009;&#x3bc;m pixel per line of a letter corresponds to 20/420 acuity. Decreasing the pixel size in the same flat geometry is difficult due to the constrained electric field, especially considering a 40&#x2009;&#x3bc;m thick debris layer separating the implant from the target neurons. Here we optimize the electrode design to help overcome such limitations.Approach.An end-to-end modeling pipeline combines the retinal photovoltaic implant simulator based on the Xyce circuit simulator with an interface to COMSOL Multiphysics for electric field modelling. It was used to generate and characterize implants in an open-loop sampling-based optimization. Implant performance was evaluated with respect to voltage drop across bipolar cells (BC) (representing the stimulation strength), pattern contrast, and neural selectivity.Main results.The highest selectivity in stimulation of BCs was achieved with arrays having active electrodes on pillars and return electrodes connected in a mesh surrounding the photovoltaic pixels in the array. Such a design, even with pixels down to 20&#x2009;&#x3bc;m, provides stimulation strength exceeding, and contrast similar to that of flat 100&#x2009;&#x3bc;m PRIMA pixels.Significance.Using a novel 3D electrode design, the pitch of the photovoltaic array can be decreased to 20&#x2009;&#x3bc;m, while providing performance that exceeds the flat 100&#x2009;&#x3bc;m PRIMA pixels. In humans, 20&#x2009;&#x3bc;m resolution on the retina corresponds to a visual acuity of 20/80 - a five times improvement compared to the current clinical device.

Early and accurate detection of oral cancer plays a pivotal role in improving patient prognosis and survival rates. Deep learning (DL) models have shown promise in automating medical image classification; however, performance optimization remains a challenge due to complex network configurations and hyperparameter dependencies. This study introduces an enhanced diagnostic framework combining the InceptionV3 convolutional neural network with the Aquila Optimizer (AO), a nature-inspired metaheuristic algorithm, to achieve superior classification accuracy in identifying oral cancer lesions. A standardized dataset of labeled oral lesion images, including both benign and malignant cases captured via mobile and intraoral cameras, was used for training. The InceptionV3 model, initially pre-trained, was fine-tuned for binary classification tasks. AO was employed to optimize the hyperparameters by defining a search space and iteratively improving model performance through accuracy maximization and loss minimization strategies. The optimized model was compared against leading architectures such as AlexNet, MobileNet, Xception, ResNet-50, and the original InceptionV3, using comprehensive performance indicators like accuracy, precision, recall, F1-score, AUC-ROC, specificity, log loss, and Matthews Correlation Coefficient (MCC). The proposed AO-InceptionV3 model consistently outperformed the other DL architectures across all metrics. It achieved a classification accuracy of 97.80%, precision of 97.81%, recall of 97.79%, and an MCC of 0.956, while maintaining a low log loss of 0.0735 and an AUC-ROC of 99.81%. Visual analyses, including ROC curves and 3D plots, reinforced the robustness and reliability of the model in distinguishing between benign and malignant lesions with minimal inference time. The integration of the Aquila Optimizer into the InceptionV3 architecture significantly improves the diagnostic performance of DL models for oral cancer detection. The proposed framework demonstrates excellent potential for real-time clinical deployment, offering high accuracy, efficiency, and reliability, and sets a benchmark for future AI-driven cancer diagnostic systems.