Comprehensive Summary
This study proposes a new model for an AI pupillary-computer interface (PCI) that utilizes the pupil light reflex (PLR) response to binary coded flickering stimuli. Twelve human adult participants were exposed to different levels of light stimulus, and their pupil size variations were captured using an infrared eye tracker. The PLR signals were encoded using the Gramian Angular Field (GAF) transformation method which allowed for usage of the signals as an image to input into a convolution neural network (CNN). Temporal convolution layers (TCNs) were added to account for the spatiotemporal information of the image data. the signals were analyzed, and the PCI was able to achieve high performance with classification accuracies of 98.61%, 93.84%, and 91.84%. Comparative analyses with other learning models showed that this proposed AI-based PCI outperforms others. This model is inexpensive, requires no user training, and remains stable over time.
Outcomes and Implications
PLR responses induced by visual stimuli can be used as functional interfaces that would allow for communication in individuals that are "locked-in" or have other motor impairments. A PCI would also provide a non-invasive alternative to brain computer interfaces (BCIs). PLR classification is also useful for diagnosis of concussions, Parkinson's disease, and Attention-deficit/hyperactivity disorder (ADHD). Further research into the PLR can aid in creation of effective diagnostic tools for these neurological disorders.