For our latest research visit our new website at www.le.ac.uk/csn.
In collaboration with Christof Koch at Caltech and Itzhak Fried at UCLA we study multiple single-neurons and local field potentials in conscious humans during different tasks. The data comes from patients suffering from epilepsy, which do not improve with medication and are implanted with intracranial electrodes to determine the focus of the epileptic seizures and evaluate the potential outcome of curative surgery.
There has been a very long controversy in Neuroscience about how information is represented by neurons in the brain. In particular, researchers have been arguing whether concepts, such as the identity of a person, are represented by the common activity of an immense neural network or by just by a few ‘abstract’ neurons. At this respect, we found a remarkable type of neurons that responded to the identity of a given individual, disregarding the visual details of different pictures of the same person. For example, one neuron responded to different pictures of Jennifer Aniston, but not to other persons, objects, etc. Remarkably, these neurons also responded to the written name of the particular person and not to other names. We believe these neurons play an important role in the transformation of complex visual percepts into long-term and more abstract memories. They are reminiscent of ‘Grandmother cells’ (it is the closest neuroscience got to something like this), but they are far from being ‘the one and only cell’ representing one (and only one) single concept. See the discussion in our original Nature article.
This finding opened a completely new set of questions we are currently targeting with follow up experiments. The main goal is to study how the activity of these neurons correlates with conscious visual perception and the formation of memories. This research gives the extraordinary opportunity to study neuronal responses in conscious humans during different behavioural tasks and brain states.
Invariant visual representation by single-neurons in the human brain.
R. Quian Quiroga, L. Reddy, G. Kreiman, C. Koch and I. Fried
Nature, 435: 1102-1107; 2005. Suppl. Inf. (4.5 Mb)
The activity of neurons in the vicinity of the implanted microwires is reflected in spikes. Each microwire detects the spikes of all neurons in its surroundings and each neuron fires spikes of a particular shape. In order to understand how these neurons encode information in the brain, we need a reliable way of detecting and sorting the spikes (i.e. separating the spikes from the different neurons based on their shapes). At this respect, we developed “Wave_clus” an unsupervised and fast method to do this. It is based on using the wavelet transform to extract the features that best separates the different spike shapes and superparamagnetic clustering, a method from statistical mechanics that doesn’t assume any particular shape or distribution of the clusters. Moreover, we proposed a very simple trick to improve the detection of the spikes by using the median of the signal.
Although we already got excellent results with the current algorithm, we are always trying further optimizations. Our goal is to identify as many neurons as possible from each electrode in an unsupervised way.
Unsupervised spike sorting with wavelets and superparamagnetic clustering.
R. Quian Quiroga, Z. Nadasdy and Y. Ben-Shaul
Neural Computation, 16: 1661-1687; 2004.
How much information is represented by a population of neurons can be quantified in an objective manner by reconstructing the stimulus or any particular behavior from the pattern of responses of the neurons in question.
In collaboration with Richard Andersen at Caltech, we have been studying the decoding of different movement plans using the activity of neurons in the posterior parietal cortex of monkeys. The possibility of predicting movement intentions from neuronal activity has applications in brain-machine-interfaces, especially for the development of neural prosthesis for paralyzed patients.
Our general approach is to use decoding strategies to understand how the brain extracts features and deciphers information encoded in the activity of population of neurons (in contrast to the standard cell-by-cell analysis). We can then use this information to move robot devices for potentials clinical applications.
Extracting information from neural populations: Information theory and decoding approaches
It is common practice to study electroencephalographic (EEG) responses to different types of sensory stimulation. These evoked potentials (EPs) are very small in comparison with the ongoing EEG and are barely visible in the individual trials. Therefore, most EP research relies on the identification of different waves after averaging several presentations of the same stimulus pattern. Although ensemble averaging improves the signal-to-noise-ratio, it implies a loss of information related to systematic or unsystematic variations between the single-trials. This information may be crucial to study the time course of dynamic cognitive processes, simple and complex behavioral patterns and cognitive dysfunctions in pathological conditions.
We developed a denoising method based on the wavelet transform, “EP_den” to visualize the single-trial evoked responses and study single trial latency and amplitude changes. The use of “EP_den” allows the study of the dynamics of cognitive processes using single-trial evoked potentials. For example, in collaboration with Marijtje Jongsma from the University of Nijmegen, we developed a new paradigm –the “learning-oddball” – to track learning processes from the single-trial evoked responses. With Marijtje and colleagues from the University of Bergen, we used the learning-oddball together with single trial analysis to integrate the information from the EEG and fMRI, allowing the study brain processes with fine temporal (EEG) and spatial (fMRI) resolutions (see Eichele et al, PNAS 2005).
Single-trial event-related potentials with Wavelet Denoising.
In collaboration with Peter Grassberger, at the Research Center Juelich in Germany, we have been working on new measures of non-linear synchronization. These methods are based on non-linear dynamical systems theory and they have the main advantage of being sensitive to non-linear interactions. For coupled chaotic systems and for real data we showed a better performance in comparison with conventional linear methods, such as cross-correlation and coherence. Moreover, since the measures are asymmetric, it is in principle possible to establish driver-response relationships of two given signals. The possibility of establishing driver-response relationships has a wide variety of applications, such as the determination of the source of a market crash, an infectious disease, an earthquake, an epileptic seizure, a machine break, etc.
We also developed “event-synchronization”, a quantitative measure of synchronization that is especially suited for point processes. This measure deals better with problems such as sparseness of events, which are troublesome for conventional measures. Moreover, it gives a straightforward visualization of the time evolution of the delay and synchronization level with an excellent resolution. Event-synchronization has been successfully used for the study of EEG data from epileptic patients in order to localize the source of the seizures. This technique has the appeal of being very simple and fast, thus being suitable for on-line implementations.
Performance of different synchronization measures in real data: a case study on electroencephalographic signals.
Quian Quiroga R, Kraskov A, Kreuz T and Grassberger P.
Phys. Rev. E, 65: 041903; 2002.
In collaboration with Dr. Sandra Dudley at Museum Studies and Dr. David Barrie, former director of the Art Fund, we study principles of visual perception of art in the museum environment. There has been historically very little interaction between arts and science, but clearly artists and neuroscientists have complementary knowlege and expertise about visual perception. Our goal is to link both fields to start understanding, using Eye-tracker recordings and antropological data, the underlying mechanisms involved in the perception of visual art and how these are affected by the museum environment, the available information, expertise, etc.