Trending Science: Images of human faces accurately recreated by monitoring brain cell activity in monkeys

Using functional magnetic resonance imaging (fMRI) and electrophysiological recordings researchers sought to construct an explicit model of face-selective cells as a means of decoding an arbitrary, realistic face from face cell responses. They were also interested in discovering if they could anticipate the firing of cells in response to the presentation of a facial image.

Determining how brain cells compute recognisable images in the higher visual cortex was one of the study’s initial steps. Earlier work conducted by Professor Doris Tsao from the California Institute of Technology (Caltech), and colleagues, used fMRI to identify patches of specialised cells that are active when people are shown images of a face. The neurons, which the team called ‘face cells’, are found in the six areas in the inferior temporal (IT) cortex. These they call ‘face patches’.

Researchers first localised six face patches in two monkeys, using fMRI, by presenting a face localiser stimulus set containing images of faces and non-face objects. Some patches were then targeted for electrophysiological recordings. In order to interpret the activity, the team came up with 50 different dimensions that make up a recognisable face using parameters such as: distance between the eyes, width of hairline, along with non-shape based characteristics such as skin tone.

The team then showed the animals pictures of 2 000 faces and monitored the resulting activity in 205 neurons in the face patches. Their results are published in the journal ‘Cell’ in a paper, titled ‘The Code for Facial Identity in the Primate Brain’. Reporting on the findings, the ‘New Scientist’ quotes Professor Tsao, ‘We’ve cracked the brain’s code for facial identity.’

Professor Tsao explains that although there are an infinite number of faces, they can be described using just 50 dimensions. ‘It’s like computer-generated imagery, except it’s in our brains.’ The research suggests that around 200 neurons each encode different characteristics of a face. But when all are combined, the information contributed by each nerve cell allows the macaque brain to build up a clear image of someone''s face. By using an algorithm the researchers were able to recreate faces the monkeys had seen in the photos. When the intial image was placed next to the reconstruction they were nearly identical. Just 106 cells in one patch and 99 in another were enough to recreate the face accurately.

‘This was completely shocking to us – we had always thought face cells were more complex. But it turns out each face cell is just measuring distance along a single axis of face space, and is blind to other features,’ the BBC reports Professor Tsao as saying. The close relationships between primates suggests a similar mechanism may be at work in the human brain. The paper’s first author, Steve Le Chang, suggests the work may indicate that other objects could be encoded with similar, simple coordinate systems.