Journal of neural engineering最新文献

{"title":"E-Sort: empowering end-to-end neural network for multi-channel spike sorting with transfer learning and fast post-processing.","authors":"Yuntao Han, Shiwei Wang","doi":"10.1088/1741-2552/ae0d33","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0d33","url":null,"abstract":"<p><strong>Objective: </strong>Spike sorting, which involves detecting and attributing spikes to their putative neurons from extracellular recordings, is a common process in electrophysiology and brain-computer interface systems. Recent advances in large-scale neural recording technologies are challenging the conventional algorithms because of the intensive computational workloads required and the accuracy degradation suffered from time-variant spike patterns and significant levels of noise. Neural networks have demonstrated promising performance in processing these large-scale neural recordings. However, their applications are constrained by the labor-intensive data labeling and the lack of fully vectorised frameworks with end-to-end neural networks.</p><p><strong>Approach: </strong>We propose E-Sort, an end-to-end neural network-based spike sorter with transfer learning and parallelizable post-processing to address both obstacles.</p><p><strong>Main results: </strong>We examined our framework in both synthetic and real datasets. The results of the processing of the synthetic datasets show that our approach can reduce the number of annotated spikes required for training by 44% compared to training from scratch, achieving up to 25.7% higher accuracy. We evaluated E-Sort on various probe geometries, noise levels, and drift patterns, which demonstrates that our design can achieve an accuracy that is comparable with Kilosort4 while sorting 50 seconds of data in only 1.32 seconds. To test with real datasets, we first sorted the spikes using Kilosort4 and used the sorted spikes at the initial period to pre-train the neural network; then we compared and measured the agreement between the results from the trained model and those from Kilosort4. On average the pre-training process improved the result agreement by 30% approximately.</p><p><strong>Significance: </strong>E-Sort offers a scalable, efficient, and accurate neural network-based framework for large-scale spike sorting, significantly reducing manual labelling effort and processing time.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145194363","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A methodological framework for the efficient characterization of peripheral nerve stimulation parameters.","authors":"Rachel S Jakes, Benjamin J Alexander, Vlad I Marcu, A Bolu Ajiboye, Dustin J Tyler","doi":"10.1088/1741-2552/ae0d31","DOIUrl":"10.1088/1741-2552/ae0d31","url":null,"abstract":"<p><strong>Objective: </strong>Restoring movement and somatosensation with peripheral nerve stimulation (PNS) requires precise neural activation. Because pulse amplitude (PA) and pulse width (PW) recruit axons differently, intentionally modulating both could enable more complex PNS. However, mapping the PA-PW space is currently prohibitively time-intensive. This paper proposes and clinically validates an efficient method to characterize multiple intensities in the PA-PW space for motor and perceptual sensory applications using minimal data collection.</p><p><strong>Approach: </strong>We used cuff electrodes implanted in one participant with a spinal cord injury to generate iso-EMG activation contours and two participants with upper limb loss to generate somatosensory perceptual iso-intensity contours in the PA-PW space. Strength-duration (SD) curves were mapped to the contours using varying sample point subsets and assessed for fit quality. Finite element modeling of a human nerve and activation simulations evaluated differences in recruited axon populations across the PA-PW space.</p><p><strong>Main results: </strong>SD curves accurately fit all levels of motor activation and perceptual intensity (median R^2 = 0.996 and 0.984, respectively). Reliable estimates of SD curves at any intensity require only two sufficiently-spaced points (motor R2 = 0.991, sensory R2 = 0.977). Using this data, we present and validate a novel method for efficiently characterizing the PA-PW space using SD curves, including a metric that quantifies mapping accuracy based on two sampled points. In silico, intensity-matched high-PW and high-PA stimulation recruited overlapping, but not equivalent, axon sets, with high-PA stimuli preferentially recruiting large-diameter fibers and axons farther from the contact.</p><p><strong>Significance: </strong>This method enables rapid, accurate mapping of the stimulation parameter space for clinical motor and sensory PNS. The efficiency of the proposed characterization approach enhances the clinical feasibility of multiparameter modulation, establishing a framework for further exploration of two-parameter modulation for increased selectivity and resolution, reduced fatigue, and unique percept generation. (ClinicalTrials.gov ID NCT03898804).</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145194343","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From zero- to few-shot: deep temporal learning of wrist EMG enables scalable cross-user gesture recognition.","authors":"Fady S Botros, Heather E Williams, Angkoon Phinyomark, Erik J Scheme","doi":"10.1088/1741-2552/ae08eb","DOIUrl":"10.1088/1741-2552/ae08eb","url":null,"abstract":"<p><p><i>Objective.</i>Wrist electromyography (EMG) is emerging as an enticing wearable input modality for human-machine interaction. Traditionally recorded from the forearm for use in transradial prostheses, wrist-based EMG sensors are now being integrated into devices such as watches and wristbands for hand gesture recognition (HGR). Consumer familiarity with wrist-worn devices makes wrist EMG a compelling option, but the need for individualized user calibration remains a challenge.<i>Approach.</i>This study therefore evaluated various cross-user models to reduce the calibration burden and compared wrist- and forearm-based models. Eight different machine learning architectures were evaluated across 33 users, using varying amounts of data from the end user.<i>Main results.</i>A temporal convolutional network-bidirectional long short-term memory architecture, applied for the first time to EMG classification, was found to significantly (<i>p</i> < 0.05) outperform other tested machine learning architectures. An inter-day feature set combined with<i>Z</i>-score normalization achieved the best performance when classifying five gestures (plus a rest class) using either wrist or forearm EMG. Consistent with other recent results, wrist EMG consistently outperformed forearm EMG in all analyses, including within- and across-user comparisons (<i>p</i> < 0.05). In cross-user models, wrist EMG demonstrated a zero-shot performance of 78.2%, compared to 71.6% for forearm EMG (<i>p</i> < 0.05). Introducing one calibration repetition from the end user increased one-shot performance of wrist EMG to 91.6%, compared to 86.9% for forearm EMG (<i>p</i> < 0.05). Adding further training repetitions boosted wrist EMG performance to 98.3%, compared to 97.4% for forearm EMG.<i>Significance.</i>These findings provide new evidence supporting the viability of wrist EMG for cross-user HGR models that generalize to new users with minimal calibration, suggesting promising potential for its broader adoption in wearable devices.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145088512","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CS-Net: convolutional spider neural network for surface-EMG-based hybrid gesture recognition.","authors":"Xi Zhang, Jiannan Chen, Lei Liu, Fuchun Sun","doi":"10.1088/1741-2552/ae0c38","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0c38","url":null,"abstract":"<p><p>In this paper, we propose a novel neural network architecture, the Convolutional Spider Neural Network (CS-Net), combined with a transfer learning strategy, to classify hybrid gestures that integrate wrist postures and hand movements. The CS-Net framework incorporates diverse surface electromyography (sEMG) features, including raw signals and FFT representations, through a multi-stream information fusion mechanism to enhance classification performance. The proposed transfer learning strategy involves pre-training the model on specific wrist postures and fine-tuning it on the full set of hybrid gestures, leveraging the intrinsic relationships between composite gestures and their constituent movements to improve accuracy.&#xD;&#xD;The framework is evaluated through extensive offline experiments using a dataset of 12 hybrid gestures (combining three wrist postures and four hand movements) collected from six subjects, comparing its performance against three deep learning algorithms in sEMG recognition filed. The average experimental result for the proposed CS-Net with transfer learning (TL) reached 90.6%. Additionally, its generalization ability is validated with the Ninapro public databases, which are DB1, DB4, and DB5. The 30 action classification accuracy of CS-Net on the Ninapro datasets was 68.7%, 61.5%, and 66.3%, respectively. To demonstrate practical applicability, real-time online experiments involving object grasping tasks is conducted, achieving a success rate of 90%. The results show that CS-Net significantly improves sEMG classification accuracy, while the transfer learning strategy further enhances performance. Moreover, the algorithm achieved a high success rate in online experiments, confirming its robustness and practical utility for real-world applications. Our hybrid gesture dataset and source codes are available on Github(https://github.com/Xi-Ravenclaw/CS-Net).</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145180018","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development of novel signal and spike velocity analysis tools in compact peripheral nerve recording designs.","authors":"Jonas Klus, Alexander J Boys, Ruben Ruiz-Mateos Serrano, George G Malliaras, Alejandro Carnicer-Lombarte","doi":"10.1088/1741-2552/ae0c3b","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0c3b","url":null,"abstract":"<p><strong>Objective: </strong>Analysis tools for peripheral nerve recordings remain underdeveloped compared to those for brain signals, limiting the advancement of nerve neurotechnologies for clinical treatments such as closed-loop systems. This study introduces and explores the performance of two novel nerve signal analysis techniques - cross-correlation analysis and spike delay velocity analysis - which rely on a defining feature of peripheral nerve signals: the reliable conduction velocity of signals transmitted by axons in nerves. &#xD;Approach. We test the capabilities of the introduced cross-correlation and spike delay velocity analysis techniques both in silico on synthetic nerve signals and on in vivo nerve signals acquired from freely-moving rats. &#xD;Main results. Our findings show that both techniques can be successfully employed to extract transmission direction and velocity information from compact two-electrode site peripheral nerve recording designs. Notably, cross-correlation analysis can be employed to detect neural signals of very low signal-to-noise ratio, otherwise undetectable by typical spike detection approaches. &#xD;Significance. Our findings provide new techniques to both enhance detection and extract new information in the form of velocity data from nerve recordings using a compact two-electrode site recording setup. Unlike traditional methods, this design eliminates the need for long electrode arrays, making it particularly well-suited for use in freely-moving animal models and translational applications. As axon signal conduction direction and velocity is tightly linked to neural function, these techniques can support new research into peripheral nervous system function and new therapeutic approaches driven by neural interfaces.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145180023","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bayesian decoding and its application in reading out spatial memory from neural ensembles.","authors":"Ning Wang, Xinyi Deng, Nan Zhu, Xueling Wang, Yimeng Wang, Biao Sun, Chenguang Zheng","doi":"10.1088/1741-2552/ae0c3c","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0c3c","url":null,"abstract":"<p><p>Spatial memory serves as a foundation to establish cognitive map, supporting navigation and decision-making processes across species. Essential brain regions such as the hippocampus and entorhinal cortex enable these functions through spatially tuned neurons, particularly place cells, which encode an animal's precise location. The continuous spatial trajectories are then able to be represented by temporally sequential firing of these cells at neural ensemble level. Bayesian frameworks are powerful tools for reconstructing such \"mind travel\". In this article, we focus on the principles and advances of Bayesian decoding methods for extracting spatial memory information from neural ensembles. First, we review non-recursive approaches and recursive point process filters, paying special attention to clusterless decoding strategies. We also discuss emerging approaches such as neural manifolds within Bayesian estimation. Next, we discuss the advanced application of Bayesian decoding in understanding the neuronal coding mechanisms of memory consolidation and planning, and in supporting computational model establishment and closed-loop manipulation. Finally, we discuss the limitations and challenges of recent approaches, highlighting the promising strategies that could raise the decoding efficiency and adapt the growing scale of neural data. We believe that the developing of Bayesian decoding approach would significantly benefit for techniques and applications of memory-related brain machine interface.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145180026","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neuralace: manufacture, parylene-C coating, and mechanical properties.","authors":"Juan Pablo Botero, Spencer M Roberts, Piotr Mackowiak, Nicholas Witham, Lukas Selzer, Balaji Srikanthan, Kai Zoschke, Sandeep Negi, Florian Solzbacher","doi":"10.1088/1741-2552/ae0c39","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0c39","url":null,"abstract":"<p><strong>Objective: </strong>This study investigates the mechanical properties of the Neuralace, a novel ultra-thin, high-channel-count mesh-type subdural electrode array, to characterize its mechanical compatibility with neural tissue (i.e., the forces exerted onto the brain upon conformation) for chronic brain-computer interface (BCI) applications.</p><p><strong>Approach: </strong>A full-factorial design of experiments was used to assess the effects of geometrical variations, orientation, and polymeric encapsulation on the stiffness of silicon-based Neuralace structures. A custom low-force four-point bending setup was developed to measure flexural stiffness in a physiologically relevant displacement range.</p><p><strong>Main results: </strong>The stiffness values of Neuralace structures ranged from 2.99 N/m to 7.21 N/m, depending on the cell-wall thickness (CWT) of the lace, orientation, and parylene-C (PPXC) encapsulation. Orientation and CWT had the largest impact on the stiffness of the structures, while the effects of PPXC encapsulation were statistically significant but more subtle. The stiffest Neuralace configuration is expected to exert forces approximately 10 to 100 times lower than commercially available subdural implants would when conforming to the brain's topology (considering a gyrus of 60 mm radius).</p><p><strong>Significance: </strong>Subdural electrode arrays have traditionally been used for epilepsy monitoring and surgical planning. These arrays are now transitioning from short-term implantation in epilepsy monitoring to long-term use in BCIs, which requires consideration of the foreign body response to ensure long-term durability and functionality. Biocompatibility challenges, such as fibrotic encapsulation and reactive astrogliosis, highlight the need for conformal subdural implant designs that minimize mechanical stress on neural tissue. This study establishes a rigorous and reproducible framework for mechanical characterization of conformable neural implants and demonstrates the feasibility of tuning design parameters to reduce implant-induced mechanical stress on cortical tissue. The results support future development of chronic BCI-compatible subdural electrodes with improved biocompatibility through mechanical design.&#xD.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145180053","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"BGTransform: a neurophysiologically informed EEG data augmentation framework.","authors":"Jin Yue, Xiaolin Xiao, Hao Zhang, Minpeng Xu, Dong Ming","doi":"10.1088/1741-2552/ae0c3a","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0c3a","url":null,"abstract":"<p><strong>Objective: </strong>Deep learning has emerged as a powerful approach for decoding electroencephalography (EEG)-based brain-computer interface (BCI) signals. However, its effectiveness is often limited by the scarcity and variability of available training data. Existing data augmentation methods often introduce signal distortions or lack physiological validity. This study proposes a novel augmentation strategy designed to improve generalization while preserving the underlying neurophysiological structure of EEG signals.&#xD;Approach. We propose Background EEG Transform (BGTransform), a principled data augmentation framework that leverages the neurophysiological dissociation between task-related activity and ongoing background EEG. In contrast to existing methods, BGTransform generates new trials by selectively perturbing the background EEG component while preserving the task-related signal, thus enabling controlled variability without compromising class-discriminative features. We applied BGTransform to three publicly available EEG-BCI datasets spanning steady-state visual evoked potential (SSVEP) and P300 paradigms. The effectiveness of BGTransform is evaluated using several widely adopted neural decoding models under three training regimes: (1) without augmentation (baseline model), (2) with conventional augmentation methods, and (3) with BGTransform. &#xD;Main Results. Across all datasets and model architectures, BGTransform consistently outperformed both baseline models and conventional augmentation techniques. Compared to models trained without BGTransform, it achieved average classification accuracy improvements of 2.45%-15.52%, 4.36-17.15%, and 7.55-10.47% across the three datasets, respectively. In addition, BGTransform demonstrated greater robustness across subjects and tasks, maintaining stable performance under varying recording conditions. &#xD;Significance. BGTransform provides a principled and effective approach to augmenting EEG data, informed by neurophysiological insight. By preserving task-related components and introducing controlled variability, the method addresses the challenge of data sparsity in EEG-BCI training. These findings support the utility of BGTransform for improving the accuracy, robustness, and generalizability of deep learning models in neural engineering applications.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145180062","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Amorphous silicon resistors enable smaller pixels in photovoltaic retinal prosthesis.","authors":"Andrew Shin, Nathan Jensen, Emma Butt, Jeonghyun An, Davis Pham-Howard, Ludwig Galambos, Keith Mathieson, Theodore Kamins, Daniel Palanker","doi":"10.1088/1741-2552/ae0522","DOIUrl":"10.1088/1741-2552/ae0522","url":null,"abstract":"<p><p><i>Objective.</i>Clinical trials of the photovoltaic subretinal prosthesis PRIMA demonstrated feasibility of prosthetic central vision with resolution matching its 100<i>µ</i>m pixel width. To improve prosthetic acuity further, pixel size should be decreased. However, there are multiple challenges, one of which is related to accommodating a compact shunt resistor within each pixel that discharges the electrodes between stimulation pulses and helps increase the contrast of the electric field pattern. Unfortunately, standard materials used in integrated circuit resistors do not match the resistivity required for small photovoltaic pixels. Therefore, we used a novel material-doped amorphous silicon (a-Si) and integrated it into photovoltaic arrays with pixel sizes down to 20<i>µ</i>m.<i>Approach.</i>To fit within a few<i>µ</i>m²area of the pixels and provide resistance in the MΩ range, the material should have sheet resistance of a few 100 kΩ sq^-1, which translates to resistivity of a few Ω * cm. The a-Si layer was deposited by low-pressure chemical vapor deposition and its resistivity was adjusted by PH₃doping before encapsulating the resistors between SiO₂and SiC for stability<i>in-vivo. Main results.</i>High-resolution retinal implants with integrated shunt resistors were fabricated with values ranging from 0.75 to 4 MΩ on top of the photovoltaic pixels of 55, 40, 30 and 20<i>µ</i>m in size. Photoresponsivity with all pixel sizes was approximately 0.53 A W^-1, as high as in the arrays with no shunt resistor. The shunts shortened electrodes discharge time, with the average electric potential in electrolyte decreasing by only 21%-31 % when repetition rate increased from 2 to 30 Hz, as opposed to a 54%-55 % decrease without a shunt. Similarly, contrast of a Landolt C pattern increased from 16%-22 % with no shunt to 22%-34 % with a shunt. Further improvement in contrast is expected with pillar electrodes and local returns within each pixel.<i>Significance.</i>Miniature shunt resistors in a MΩ range can be fabricated from doped a-Si in a process compatible with manufacturing of photovoltaic arrays. The shunt resistors improved current injection and spatial contrast at video frame rates, without compromising the photoresponsivity. These advances are critical for scaling pixel sizes below 100 <i>µ</i>m to improve visual acuity of prosthetic vision.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12462637/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145031507","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Using economic value signals from primate prefrontal cortex in neuro-engineering applications.","authors":"Tevin Rouse, Shira M Lupkin, Vincent B McGinty","doi":"10.1088/1741-2552/ae0bf6","DOIUrl":"https://doi.org/10.1088/1741-2552/ae0bf6","url":null,"abstract":"<p><strong>Objective: </strong>Brain-machine interface research has shown the efficacy of using motor and sensory-related neural signals to assist physically impaired patients. Despite the comparable ability to extract more abstract cognitive signals from the brain, little effort has been devoted to leveraging these signals in neuroengineering applications. In this study, we explore the use of neural signals related to economic value, a key cognitive construct, in a BMI context.</p><p><strong>Approach: </strong>Using multivariate time series data collected from the orbitofrontal cortex in non-human primates, we develop deep learning-based neural decoders to predict the monkey's choice in a value-based decision-making task. We implement a reinforcement learning-based training approach to develop adaptive decoders that can be extended to handle multi-step decisions, which frequently arise in real-world settings.</p><p><strong>Main results: </strong>We develop neural decoders leveraging subjective value signals to predict the monkey's choice with < 70% accuracy on average, with above-chance accuracy even when choice options are objectively equal. We show that this same decoder architecture can be trained to execute choice-related actions and execute action sequences aligned with the user's goal. Finally, we explore a decoder architecture that uses a neural forecasting model equipped with task-related information, and show that it makes high accuracy predictions ∼ 300 ms sooner than would otherwise be possible.</p><p><strong>Significance: </strong>These findings support the feasibility of user preference-informed neuroengineering devices that leverage abstract cognitive signals to aid users in goal-directed behavior. It demonstrates that using abstract cognitive signals in real-world settings may be more accurate when combined with information from multiple sources, such as motor and sensory regions. This research also highlights the potential need for systems to measure their confidence in their actions when user input is minimal.</p>","PeriodicalId":94096,"journal":{"name":"Journal of neural engineering","volume":" ","pages":""},"PeriodicalIF":3.8,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145152290","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}