A central goal of system neuroscience is to understand not only how a set of neurons encode information, but also how the activity of neurons is decoded to ultimately result in behaviors. It requires neural technologies to condition neural activity and interrogate neural circuits, where a simultaneous stimulation and low-noise artifact-free recording represents the key technology. Another pull motive is the need to treat diseases and target conditions that are currently intractable. Many studies have suggested that closed-loop stimulation can improve therapeutic outcomes and reduce side effects. However, due to recorder saturation and excessive noise caused by stimulation, it is not readily possible to observe direct neural responses and perform closed-loop neuromodulation with current technologies in the literature.
Our contribution is a MIST technology that for the first time allows continuous, simultaneous neural recording and stimulation. To the best of the knowledge, our invention is the only technology that can record neural signals under electrical stimulation without saturation, where this funciton has been validated in in-vivo experiments. We are now studying the feasibility to perform closed-loop neuromodulation based on direct neuronal feedback under electrical stimulation. We propose a data-driven approach to train deep neural network that predicts direct neural responses based on multi-channel data history and stimulation parameters and reveals the actual impact of electrical stimulation on neural circuits and activities.
This research is in collaboration with Plexon. MIST is funded by DARPA ElectRx.
A fundamental premise of Bioelectronic Medicine is that effectiveness of neuromodulation therapy can be monitored and optimized in each patient by applying controlled amounts of charge into the nerve while monitoring the neural response from the treated organ or system. While existing neural amplifier technology has sufficient input-referred noise characteristics for resolving action potentials in brain recordings, it is inadequate for resolving small neural signals (typically less than ten microvolts) from noise sources in cuff electrode interfaces on autonomic nerves. Penetrating electrodes can improve the signal quality at early time points. However, the signal will decay over time due to foreign body response and micro-motion of the tethered electrodes relative to the soft tissue. In addition, electrical stimulation produced by adjacent electrodes on the same nerve generates large noise and recording artifacts that can be several orders of magnitude larger than the spontaneous nerve activity, making it extremely challenging to perform simultaneous stimulation and recording on autonomic nerves, especially when the stimulating and recording electrodes need to be positioned close to each other.
Functional magnetic resonance imaging (fMRI) based on the BOLD contrast has become a powerful neuroimaging modality for mapping brain activity change, functional connectivity, neural network and circuitry under normal and diseased conditions. Nevertheless, due to a variety of technical hurdles and safety concern in dealing with traditional metal electrodes, it is extremely challenging to perform simultaneous electrophysiological recording and brain stimulation with fMRI without scarifying the imaging quality, especially at a high and ultrahigh field. It has significantly limited the research ability aiming to understand the fundamental basis of fMRI, BOLD correlation with neuronal activity, and underlying mechanism of neuromodulation.
Sensing is the process of deriving signals from the environment that allows artificial systems to interact with the physical world. The Shannon theorem specifies the maximum rate at which information can be acquired. However, this upper bound is hard to achieve in many man-made systems. The biological visual systems, on the other hand, have highly efficient signal representation and processing mechanisms that allow precise sensing. In this work, we argue that redundancy is one of the critical characteristics for such superior performance. We show architectural advantages by utilizing redundant sensing, including correction of mismatch error and significant precision enhancement. For a proof-of-concept demonstration, we have designed a heuristic-based analog-to-digital converter - a zero-dimensional quantizer. Through Monte Carlo simulation with the error probabilistic distribution as a priori, the performance approaching the Shannon limit is feasible. In actual measurements without knowing the error distribution, we observe at least 2-bit extra precision. The results may also help explain biological processes including the dominance of binocular vision, the functional roles of the fixational eye movements, and the structural mechanisms allowing hyperacuity.
Spike sorting is a fundamental preprocessing step for many neuroscience studies which rely on the analysis of spike trains. Most of the feature extraction and dimensionality reduction techniques that have been used for spike sorting give a projection subspace which is not necessarily the most discriminative one. Therefore, the clusters which appear inherently separable in some discriminative subspace may overlap if projected using conventional feature extraction approaches leading to a poor sorting accuracy especially when the noise level is high. We propose a noise robust and unsupervised spike sorting algorithm based on learning discriminative spike features for clustering. The proposed algorithm uses discriminative subspace learning to extract low dimensional and most discriminative features from the spike waveforms, and perform clustering with automatic detection of the number of the clusters. The core part of the algorithm involves iterative subspace selection using linear discriminant analysis and clustering using Gaussian mixture model with outlier detection. A statistical test in the discriminative subspace is proposed to automatically detect the number of the clusters. Comparative results on publicly available simulated and real in-vivo datasets demonstrate that our algorithm achieves substantially improved cluster distinction leading to higher sorting accuracy and more reliable detection of clusters which are highly overlapping and not detectable using conventional feature extraction techniques such PCA or wavelets. By providing more accurate information about the activity of more number of individual neurons with high robustness to neural noise and outliers, the proposed unsupervised spike sorting algorithm facilitates more detailed and accurate analysis of single- and multi-unit activities in neuroscience and brain machine interface studies.
Neural recording system miniaturization and integration with low-power wireless technologies require compressing neural data before transmission. Feature extraction is a procedure to represent data in a low dimensional space, its integration into a recording chip can be an efficient approach to compress neural data. In this paper, we propose a streaming principal components analysis (PCA) algorithm and its microchip implementation to compress multichannel LFP and spike data. The circuits have been designed in a 65nm CMOS technology and occupy a silicon area of 0.06mm^2. Throughout the experiments, the chip compresses LFPs by 10times at the expense of as low as 1% reconstruction errors and 144nW/channel power consumption; for spikes, the achieved compression ratio is 25times with 8% reconstruction errors and 3.05uW/channel power consumption. In addition, the algorithm and its hardware architecture can swiftly adapt to nonstationary spiking activities, which enables efficient hardware sharing among multiple channels to support a high-channel count recorder.
There is a growing demand for chronic, wireless neurosensor interface with on-the-fly processing capabilities. Such neurosensor interface is designed with low-power, low-noise operation, thus meeting the urgent clinical needs of providing long-term, neurological health monitoring for patients suffering from conditions such as epilepsy, Alzheimer's disease, and sleep apnea. In collaboration with Boston research groups, we have a wireless recorder designed and validated through bench-top experiments, animal experiments, and scalp EEG recordings.
What we have achieved is an implantable chip capable of recording both regular EEG and high-frequency oscillations, with the following specifications:
Neurotechnologies that allow precise interactions with large-scale neural networks or peripheral nerves are ongoing pursuits of the neural engineering/neuroscience community and industry companies. High density recording is pushed by the BRAIN Initiative, which is to 10,000 times increase the number of neurons simultaneously recorded in the next 15 years. This is towards a better understanding of the brain. Minimally invasive neural interfaces that can selectively address individual axons and fascicles are pushed by healthcare companies and funding agencies, as an alternative approach to treat both mental and physical diseases.
What we have achieved:
Wearable electronics has a 30+ years' history, which shares the vision of interweaving technology into the everyday life. Due to technology advancement and constant human needs for better quality of life, the market of wearable electronics is booming especially in healthcare and life style applications. In 2014, there are 90 million wearables reaching customers including smart watches, wearable 3D motion trackers, sport/activity trackers, smart glasses, smart clothing, and wearable cameras. Along this direction, our group has developed multiple prototypes in collaboration with medical doctors and computer scientists, building tools for disease diagnosis and treatment. Some of our prototypes are in pilot clinical trials with positive feedbacks from both clinicians and patients.
What we are working on: