Method for assessing visual saliency in children with cerebral/cortical visual impairment using generative artificial...

Cerebral/cortical visual impairment (CVI) is a leading cause of pediatric visual impairment in the United States and other developed countries, and is increasingly diagnosed in developing nations due to improved care and survival of children who are born premature or have other risk factors for CVI. Despite this, there is currently no objective, standardized method to quantify the diverse visual impairments seen in children with CVI who are young and developmentally delayed. We propose a method that combines eye tracking and an image-based generative artificial intelligence (AI) model (SegCLIP) to assess higher- and lower-level visual characteristics in children with CVI. We will recruit 40 CVI participants (aged 12 months to 12 years) and 40 age-matched controls, who will watch a series of images on a monitor while eye gaze position is recorded using eye tracking. SegCLIP will be prompted to generate saliency maps for each of the images in the experimental protocol. The saliency maps (12 total) will highlight areas of interest that pertain to specific visual features, allowing for analysis of a range of individual visual characteristics. Eye tracking fixation maps will then be compared to the saliency maps to calculate fixation saliency values, which will be assigned based on the intensity of the pixel corresponding to the location of the fixation in the saliency map. Fixation saliency values will be compared between CVI and control participants. Fixation saliency values will also be correlated to corresponding scores on a functional vision assessment, the CVI Range-CR. We expect that fixation saliency values on visual characteristics that require higher-level processing will be significantly lower in CVI participants compared to controls, whereas fixation saliency values on lower-level visual characteristics will be similar or higher in CVI participants. Furthermore, we anticipate that fixation saliency values will be significantly correlated to scores on corresponding items on the CVI Range-CR. Together, these findings would suggest that AI-enabled saliency analysis using eye tracking can objectively quantify abnormalities of lower- and higher-order visual processing in children with CVI. This novel technique has the potential to guide individualized interventions and serve as an outcome measure in future clinical trials.