Advanced Algorithms for Neural Prosthetic Systems

Lead Research Organisation: University of Cambridge
Department Name: Engineering

Abstract

Our seemingly effortless ability to make coordinated movements belies the sophisticated computational machinery at work in our nervous system. In recent years, the field of neuroscience has been dramatically expanding the complexity of its data acquisition technologies and experiments. This technological development has created a preponderance of valuable experimental data, but the analytical methods required to deeply interrogate this data have not yet been developed. Simultaneously, the last decade has seen major advances in the fields of computational statistics, data analysis techniques, and machine learning. Research in these areas has enabled investigation into and understanding of previously uninterpretable data.This proposal seeks to bring together key research from these two fields to significantly advance the scientifically and medically important application of neural prosthetic systems, which seeks to improve greatly the quality of life of hundreds of thousands of severely disabled human patients worldwide. Debilitating diseases like Amyotrophic Lateral Sclerosis can leave a human without voluntary motor control. However, in most cases, the brain itself remains intact and has normal function. The same is true with spinal cord injuries that result in severe paralysis. In fact, tetrapalegic patients list ``regaining arm/hand control'' as the top priority for improving their quality of life, as regaining this function would allow significant patient independence. To address this priority, neural prosthetic systems seek to access the information in the brain and use that information to control a prosthetic device such as a robotic arm or a computer cursor. There are many medical, scientific, and engineering challenges in developing such a system, but all neural prosthetic systems share in common a decoding algorithm. Decoding algorithms map neural activity into physical commands such as parameters for controlling a robotic arm. Current decoding approaches have shown exciting proofs of concept, but there are a number of shortcomings that must be addressed before the field produces a clinically viable prosthetic device with speed and accuracy comparable to a healthy human arm. Our research programme will use advanced statistical and machine learning technologies to create algorithms that can decode neural activity with higher precision that previously seen. We have identified several opportunities for meaningful improvement, from incorporating the statistics of natural reaching to validating these algorithms in a realistic online setting. Taken together, these algorithmic developments should help create a much higher quality neural prosthetic device.

Planned Impact

We believe this research will provide important benefit to society, both in terms of medical technology and economic opportunity, as well as to the academic community. We outline our plan to make impact in several broad categories (as identified by the Research Council) below. Societal Impact - Health and Quality of Life We believe the most important and widespread benefit of this research programme will be for the hundreds of thousands of people living with severe motor impairments. Debilitating diseases like Amyotrophic Lateral Sclerosis (ALS, often called Lou Gehrig's disease) can leave a human without voluntary motor control. However, in most cases, the brain itself remains in tact and has normal function. The same is true with spinal cord injuries that result in severe paralysis. In fact, tetrapalegic patients list ``regaining arm/hand control'' as the top priority for improving their quality of life, as regaining this function would allow significant patient independence. To address this priority, neural prosthetic systems seek to access the information in the brain and use that information to control a prosthetic device such as a robotic arm or a computer cursor. Such systems, if successful, would clearly have massive quality of life and health impact for the hundreds of thousands of people living with these conditions. While there are many challenges in delivering a clinically viable neural prosthetic, one critical and unsolved challenge is the algorithms that decode neural activity into movement. Our research programme will develop algorithms to drastically improve our ability to decode neural activity. Thus, we believe there is significant societal impact for decode algorithms that can translate proof-of-concept systems to systems that are clinically viable for this important medical application. Further, we believe that we are well positioned to make this impact. When these algorithms are demonstrated to provide superior decode performance (in offline data analysis, and online data analysis as contemplated with our collaborators at Cambridge and at Stanford), and when these algorithms become available to the broader community through academic publication, this impact will be quickly realised as a significant step towards the goal of a medically useful neural prosthetic system. Societal Impact - Economic and Industrial The benefits of these advances to patients will be accompanied by economic and industrial benefits for companies who make relevant medical device technologies. Currently only early human clinical trials are underway as academic proofs-of-concept. Nonetheless, companies have emerged around this young technology. Thus, it is reasonable to anticipate that drastically improving our ability to decode neural activity will remove a key obstacle to industrial and economic growth in neural prosthetic systems. We believe this impact will naturally be driven by device companies. However, we believe that our algorithmic work will act as a key enabler for these engineers, medical professionals, and business people to make their important economic impact. Academic Impact - Contributions to Knowledge In addition to the large societal impact opportunities, we believe there are valuable impact opportunities for the academic community. We detail those benefits in the Academic Beneficiaries section.

Publications

10 25 50

publication icon
Churchland MM (2012) Neural population dynamics during reaching. in Nature

publication icon
Cunningham, J. (2015) Linear Dimensionality Reduction: Survey, Insights, and Generalizations in Journal of Machine Learning Research

publication icon
Gilboa E (2015) Scaling Multidimensional Inference for Structured Gaussian Processes. in IEEE transactions on pattern analysis and machine intelligence

publication icon
John Cunningham (Author) (2012) Gaussian Processes for time-marked time-series data

publication icon
Knowles D (2015) Pitman Yor Diffusion Trees for Bayesian Hierarchical Clustering in IEEE Transactions on Pattern Analysis and Machine Intelligence

 
Description In this project we developed new algorithmic and experimental techniques for understanding how the brain controls movement. These techniques are applicable to developing neural prosthetic devices---brain computer interfaces that could help for example paralysed patients. A number of new algorithms were developed with use both in the analysis of neural data and in the larger context of Big Data problems for time series, networks, and other kinds of data. Our methods have been published in high-impact journals and conferences and are used by researchers in a number of fields.
Exploitation Route Our algorithmic findings can be used for analysing neural data, but also other kinds of time series and network data.
Sectors Healthcare

 
Description This study develops new algorithmic and experimental techniques for understanding the cortical basis of motor control, and the application thereof to neural prosthetic devices. A number of new algorithms were developed with use both in the analysis of neural data and in the larger context of Big Data problems for time series, networks, and other kinds of data. High impact papers have been published both in machine learning (JMLR, NIPS, ICML, AISTATS, PAMI) and in Neuroscience (e.g. in Nature, Nature Neuroscience, J Neurophysiol, and J Comp Neuro, and Neuron).
First Year Of Impact 2013
Sector Healthcare
Impact Types Societal