Visual images that we associate with a familiar concept activate neurons in the medial temporal lobe (MTL) that encode that concept. Now, Cerf, Koch, Fried and colleagues show that when multiple images are viewed simultaneously, humans can use conscious thought to regulate the activity of MTL neurons encoding different concepts, indicating that internal, cognitive processes can override neuronal activation induced by sensory input. The authors used this phenomenon in a brain–machine interface to manipulate the visibility of competing images.

The volunteers in this study were patients with epilepsy who had been implanted with electrodes in several areas of the MTL to aid brain surgery aimed at controlling their seizures. The authors recorded the activity of neurons in response to images of famous people and landmarks, and for each patient identified four neurons that each selectively responded to one image — for example, a picture of Marilyn Monroe. Previous studies have led to the proposal that such neurons represent the memory of a specific concept very familiar to the subject (for example, the concept 'Marilyn Monroe').

internal, cognitive processes can override neuronal activation induced by sensory input.

The experiment consisted of multiple trials during which the activity of the four neurons was continuously recorded. At the beginning of each trial a target image was shown for 2 seconds — this image was the 'preferred' image of one of the four neurons. The subjects then saw a hybrid image that was composed of the target image and one of the other three images. The subjects were told to focus their thoughts on the target image using any strategy. Crucially, the activities of the four neurons were continuously fed into a real-time decoding algorithm that adjusted the image on the screen on the basis of the activity level of each neuron, providing real-time feedback to the subject. Thus, if the neuron corresponding to the target image increased firing relative to the other neurons, the visibility ratio of the two images would change accordingly. This way, the subjects could control the strength of the target image within the composite image, and full visibility of the target was rapidly achieved in 69% of trials.

The authors showed that successful focusing on an image was associated with both increased activity of the neuron corresponding to the target image and simultaneous suppression of activity of the neuron corresponding to the distracter image. Activity in the other two neurons did not change. In addition, when subjects were provided with sham feedback instead of real-time feedback, there was no consistent change in the activities of the 'target' and 'distracter' neurons. Together, this indicates that controlling the visibility of the target image required both a targeted regulation of neuronal activity (rather than a general change in neuronal firing) and feedback provided by the changing neuronal activity.

These findings show that conscious, cognitive processes can regulate the activity of MTL neurons that encode a specific concept. Similar 'cognitive control' of single neuron activity has been reported in the motor cortex, in studies in which individuals could control a prosthetic limb or a cursor on a computer screen by thought. The present study suggests that the MTL could be harnessed for brain–machine interfaces that are based on conscious thoughts about abstract, familiar concepts.