Non-Invasive Cognitive State Estimation using Near-Infrared Spectroscopy

Just beyond our perceivable range of light lies the near-infrared (NIR) spectrum. This band of light, found between 700 nm and 1400 nm, holds special properties. First, the absorption spectra of near-infrared light by hemoglobin varies depending on its oxygenation state within blood cells. Second, near-infrared light is able to penetrate the skull. Through the use of a multispectral camera, we can take an image of someone in eight different bands of NIR light. Combined with face tracking technology, we can precisely isolate a targeted area on the forehead which is located directly in front of the frontal lobe. The level of NIR light absorbance in this area strongly correlates with the quantity of blood present in the brain, which in turn is directly influenced by brain activity. We collect spectroscopic data on test subjects for three different cognitive states. Baseline, where the subject is at rest. Low, in which the subject solves addition problems at a slow pace. Then the high state, where the subject solves the same addition problems but at a much faster pace. Using the collected data, we trained a deep neural network to predict one’s cognitive state with 83.7% accuracy on new data. This is compared to previous methods which required subjects to wear multiple sensors and achieved an accuracy of 64.3%. This technology is intended to be used in adaptive user-interfaces for astronauts and pilots, dynamically changing the amount of information provided depending on the cognitive state of the user.