In addition, I happen to maintain the neural
prosthesis
reading group page
The concept of a movable probe has been introduced implicitly in the process of so-called acute recordings, where an experienced operator moves a microelectrode slowly, while interpreting continuously the signal from the electrode, either by looking at a trace from an oscilloscope, or by listening to a sound from an audio monitor. The microelectrode is moved manually until the spikes are well isolated against the background noise. Such a procedure clearly depends on the experience of the operator, and is usually time consuming. Furthermore, the brain tissue tends to move spontaneously (breathing and pulsation) and due to electrode penetration (dimpling and relaxation). Also, a very common scenario includes penetrating through the cell membrane, which results in the death of the isolated cell. Consequently, the operator has to look for another neuron. A control algorithm that will automatically move the microelectrode so as to compensate for the various sources of disturbances, with a little or no human supervision, would be of ultimate importance. Moreover, the advancements in the current technology allow the design of microdrives for multiple electrodes, mounted on a relatively small structure, where the microelectrodes can move independently of each other. Clearly, manual positioning of each electrode would be very tedious, if not impossible, given the time constraint. On the other hand, the increasing computational power would make the eventual control algorithm parallelizable so the extension of the algorithm from a single electrode to multiple electrodes follows readily. More
My secondary research area is in the development of new
decoding schemes that
extract data from neural signals. One of the main unresolved issues within the
neuroscience community is the nature of the neural code: average number of
spikes per time window (rate code) versus precise timing of individual
spikes.
Recently, I developed a new decoding algorithm that attempts to reconcile the two
opposing theories. The algorithm is based on the precise timing of individual
spikes, under the assumption that the spike train is a Markov-1 point process
whose
transition probability densities are to be determined experimentally, either
parametrically or non-parametrically. If the underlying spike train is
renewal i.e. Markov-0 point process, the algorithm is still
applicable, as any renewal process can be treated as a special case of
Markov-1 process. In particular, for Poisson process, the algorithm
reduces to the rate decoder. More
Detection in neural data using wavelet transform
The presence of noise in extracellular recordings of neural activity in the
brain is inevitable. Our current knowledge about the neural code
relies
on the assumption that information is carried in a train of action
potentials that are fired by individual neurons within a population of neurons.
Therefore, it is of ultimate importance to reliably separate signal from noise
under different scenarios that include variable signal-to-noise ratios (SNR),
firing
rates, sampling rates etc. Despite numerous algorithms for detection of
extracellular potentials in a background noise, a fully automated and robust
detection
algorithm is yet to be found. We demonstrate how a continuous
wavelet
transform can be combined with basic decision theory to solve the separation
problem in a robust and unsupervised manner. Moreover, the simplicity of
the proposed solution allows for virtually real-time execution.
More