About 2 years ago, researchers reported the discovery of the so-called “Halle Berry cell” in the human brain. This, and similar cells which respond selectively to other well-known celebrities, famous landmarks or categories of objects are located on the medial surface of the temporal lobe.
The same group of researchers now report that they can decode the activity of these cells to predict what people are seeing. The ability to decode this neural activity will prove to be very useful in the development of brain-computer interfaces for amputees and paralyzed patients.
The cells in question were identified in epileptic patients who did not respond to drugs and were therefore treated surgically. Prior to the operation, clinicians electrically stimulate the brain of the conscious patient in order to determine the focus of the abnormal electrical activity causing the seizures.
The cells are located in the anterior hippocampus, and are likely to be involved in the formation of long-term memories. They carry out higher order processing of visual stimuli; some fire in response to several different images, while others are highly selective.
The Halle Berry cell, for example, becomes active only when different photographs of that actress was shown. The cell fired when the patient saw a photograph of the actress dressed as Catwoman, and even when just her name was presented. This cell therefore appears to encode the abstract concept of “Halle Berry”, rather than visual features which might be common to other images.
Using electrodes which recorded the activity of 100 neurons simultaneously, the researchers recorded the activity of 1,500 individual cells in 11 different epileptic patients. Confirming their previous study, the researchers found that 265 of those cells fired only in response to specific images.
One was found to fire when the patient was shown images of spiders; others fired in response to images of the leaning tower of Pisa, the Eiffel Tower, or the twin towers of the World Trade Center; yet others fired in response to pictures of Pamela Anderson, Jennifer Aniston, Denzel Washington or Saddam Hussein.
It was found that the cells responded in a typical way in response to the appropriate stimulus, by generating a train of 4 action potentials between 300-600 milliseconds after presentation of the stimulus.
The researchers developed a decoding algorithm based on the knowledge of which cells fire in response to which stimuli, and the pattern of neural activity that occurs in response to those stimuli. Using that algorithm, they were able to predict accurately which images the patients were seeing from the activity of the cells.
As before, the responses of some of the cells were very specific. For example, those that fired in resposne to a particular individual also responded to previously unseen images of the same person.
The presence of these cells supports the cardinal cell theory proposed in 1972 by the British neurophysiologist Horace Barlow, which states that complex percepts such faces and objects are encoded explicitly by small numbers neurons (up to 1,000), rather than than by a more generalized implicit code distributed throughout a larger population of cells.
Quian Quiroga, R., et al. (2007) Decoding vidual inputs from multiple neurons in the human temporal lobe J. Neurophysiol. 98: 1997-2007. [Abstract]