The “red reflex test” is used to screen children for leukocoria (“white eye”) in a standard pediatric examination, but is ineffective at detecting many eye disorders. Leukocoria also presents in casual photographs. The clinical utility of screening photographs for leukocoria is unreported. Here, a free smartphone application (CRADLE: ComputeR-Assisted Detector of LEukocoria) was engineered to detect photographic leukocoria and is available for download under the name “White Eye Detector.” This study determined the sensitivity, specificity, and accuracy of CRADLE by retrospectively analyzing 52,982 longitudinal photographs of children, collected by parents before enrollment in this study. The cohort included 20 children with retinoblastoma, Coats’ disease, cataract, amblyopia, or hyperopia and 20 control children. For 80% of children with eye disorders, the application detected leukocoria in photographs taken before diagnosis by 1.3 years (95% confidence interval, 0.4 to 2.3 years). The CRADLE application allows parents to augment clinical leukocoria screening with photography.

We use Convolutional Neural Networks to detect leukocoria, or white-eye reflections, in recreational photography. Leukocoria is the most prominent symptom of retinoblastoma, a solid-tumor cancer of the eye that occurs most often in young children. We trained several networks for the task, using training images downloaded from Flickr. We achieved low error rates (<3%) for classification of eye images into three classes: normal, leukocoric, and pseudo-leukocoric. We also provide a method for tuning the outputs of a trained network to match desired true-positive/false-positive rates.

Retinoblastoma is a pediatric ocular cancer typically indicated by leukocoria (white-eye pupillary reflex). Early detection of leukocoria can improve health outcomes when it indicates disease, and it can be easily seen in recreational photographs. As part of a system for automatic leukocoria detection, we propose an image processing algorithm for detecting the exact location and radius of the smallest circle containing the iris in an eye image. Our algorithms use both median filters and two-dimensional stationary wavelet transforms and achieve low error rates.