MIT Media Lab, E14-633
In order to map the dynamics of neural circuits in mammalian brains, we need tools that can record activity over large volumes of tissue and correctly attribute the recorded signals to the individual neurons that generated them. High-resolution neural activity maps will be critical for the discovery of new principles of neural coding and neural computation, and to test computational models of neural circuits. Extracellular electrophysiology, a method for sensing the electrical currents generated in the brain by neural activity, has been neuroscience's primary method of observing the activity of large populations of neurons, yet after decades of tool development, modern tools fall short of the demands of neural circuit mapping by several orders of magnitude. Additionally, well-known problems with signal attribution pose an existential threat to the viability of further system scaling, as analyses of network function become more sensitive to errors in attribution, and alternative technologies, such as optical sensing, receive greater attention in the tool development community. However, electrophysiology maintains certain advantages over alternative methods, especially as it is currently the only method approved and suited for human brain-machine interfaces. As these tools are currently being used for restoration of function in paraplegia and treatment of neural disorders such as Parkinson's disease, epilepsy, and severe depression, and these therapies have much to benefit from more complete mapping of neural circuit activity, it is of utmost importance to discover a path for scalable electrophysiology.
One of the primary challenges to developing electrophysiology tools is that neural recording systems are comprised of modules utilizing a wide range of engineering disciplines, including MEMS, microelectronics, electrochemistry, embedded electronics, software engineering, and data analysis, and tool development has traditionally been pursued within groups focused on individual modules within their domain of expertise. Unfortunately, interdependencies between system modules have limited the effectiveness of these approaches, and we believe tool development must be understood and directed at the level of complete recording systems in order to unlock the potential of modern engineering capabilities. A key insight is that blind-source separation algorithms such as Independent Component Analysis may ameliorate problems with signal attribution. These algorithms are dependent on recording signals at much finer spatial resolutions than existing probes have accomplished, and these probes place demands on the size and bandwidth of downstream signal conditioning and data acquisition modules. Fortunately, all of the modules in modern neural recording systems are built with electronics microfabrication methods which are the subject of continuous research and development by the electronics industry, so there is a great opportunity to utilize modern tools to achieve rapid increases in recording system scale, and there exists a dependable roadmap for the development of more advanced tools into the future.
We present several advances to technologies in neural recording systems, and a complete neural recording system designed to investigate the challenges of scaling electrophysiology to whole brain recording. We have developed close-packed microelectrode arrays with the highest density of recording sites yet achieved, for which we built our own data acquisition hardware, developed with a computational architecture specifically designed to scale to over several orders of magnitude. We present results from validation experiments using colocalized patch clamp recording to obtain ground-truth activity data. This dataset provides immediate insight into the nature of electrophysiological signals and the interpretation of data collected from any electrophysiology recording system. This data is also essential in order to optimize probe development and data analysis algorithms which will one day enable whole-brain activity mapping.