New study reveals why some visual cues remain subliminal

A new study by HBP researchers in Science sheds light on the fine line separating what we see and what we miss

Understanding conscious perception is a major challenge for neuroscience. A new study published on March 22nd in Science now shows why weak visual stimuli of the same strength are sometimes detected and other times remain subliminal: As the signals travel from visual areas to the prefrontal cortex, fluctuating brain dynamics can make the difference, elevating some signals above the threshold of conscious experience and stopping others in their tracks. The findings enabled the creation of a model combining elements of two major theories of consciousness and perception. Central to the model is an event called “global ignition”, which allows information to become sustained long enough to make its mark on our inner stage. In simulated trials, the model produces results highly similar to real-life measurements, including the occasional “false alarm”. Senior author Pieter Roelfsema from Netherlands Institute of Neuroscience is working in the European Human Brain Project within the areas of Human Brain Organisation and Visuo-Motor Integration.

Much of what enters through our eyes, we never actually see. Many images stay subliminal, leading a response in brain activity in the visual areas, but not entering our conscious mind. A curious finding has been that identical weak signals sometimes get through and sometimes stay below the threshold of perception. For neuroscientists led by Pieter Roelfsema, this phenomenon has provided a window to better understand the influence of internal brain dynamics on conscious perception. “With very weak stimuli close to the threshold of perception, we can learn a lot from comparing brain activity when everything about the stimulus environment is the same, except one time it is consciously experienced and the other time it isn’t”, Roelfsema explains.

The team, which also included scientists from CEA Neurospin in France, Harvard Medical School and the Italian Istituto Italiano di Tecnologia, investigated the route between primary visual areas V1 and V4 and a higher processing area, the so-called dorsolateral Prefrontal Cortex. Rhesus monkeys were trained to react to visual cues in low contrast images as their brain activity was measured in these three areas. For the experiment a small, increasingly dim light spot would flicker on a screen for 50 milliseconds. About half a second later the monkey would either report to have seen the stimulus by making an eye movement to its previous location, or report not seeing a stimulus by looking to a grey dot on the other side of the screen. Reliably, conscious reactions would only occur when there was strong and sustained activity in the frontal cortex. Cues that went unnoticed had weaker activity which quickly decayed. “Surprising was that much of the loss already happened on the way from V1 to V4, which is really not far apart,” Roelfsema says.

The theorists in the team then devised a mathematical model to simulate these dynamics. It combined elements from two established theories: “Signal Detection Theory”, which describes the presence of constant fluctuating activity in the brain, that an added stimulus can

push above a perceptual threshold, and “Global Neuronal Workspace Theory”, or GNWT. According to GNWT, an event called “global ignition” has to occur in the brain for information to become consciously accessible. In this event, the signal starts being broadcasted back and forth between brain areas, leading to an overall increase in activation. “It’s sort of an explosion in activity for a while, like an avalanche”, Roelfsema explains. The model proposes that ignition is caused by strong reciprocal interactions between the parietal and frontal cortex, which is line with observations that these interactions are weakened when unconsciousness is induced by anesthesia.

“The analysis by Stanislas Dehaene in Paris was very sophisticated, but the resulting model is really quite simple”, Roelfsema says, adding that one reason for the simplicity is that the modeled detection task is so basic. “It contains different feed-forward connections, self-connections within the areas and feedback connections. What’s really remarkable is how closely the model output resembles the real live measurements – even the occasional false alarm.”

For Roelfsema, these questions have a very practical dimension: He wants to construct a visual prosthetic for the blind. Connected to a camera, it would electrically stimulate the cortex in just the right places to induce internal visual impressions. In the Human Brain Project, his findings are integrated into the HBP Human Brain Atlas, a massive integration effort of neuroscience data into a common framework. “We contribute functional information on visual areas and processes for the atlas, but the atlas also informs our work in return. There is a big need of something like this,” the scientist says. He also collaborates with theorists, computing and robotics experts in HBP on a project to build a realistic model of visuo-motor integration. “While this study is certainly not yet the full picture about visual thresholds and conscious perception, the results are very exciting for our future work, both on visual prosthetics and our modeling efforts in the Human Brain Project.”