4 Oct. 2019. In tests with thousands of children’s photographs, a smartphone app shows it can detect early symptoms of serious eye diseases, including cancer. Developers of the White Eye Detector app at Baylor University in Waco, Texas describe their findings in the 2 October issue of the journal Science Advances.
The White Eye Detector app screens for symptoms of leukocoria, a condition in children’s eyes where white light is reflected from the pupil rather than the usual red, something many flash-camera photographers know well. The white reflection is a symptom of abnormal development and lesions in the pupil indicating a disorder such as pediatric cataract, abnormal blood vessels called Coats disease, or a malignant tumor of the retina known as retinoblastoma.
The team developing the White Eye Detector app — the brand name for ComputeR-Assisted Detector of LEukocoria, or Cradle — is led by Baylor biochemistry professor Bryan Shaw and computer scientist Greg Hamerly. The researchers note that family members and friends are more likely to spot white-eye reflections in children before many pediatricians, and administration of diagnostic tests in clinics is uneven. But with the explosion of digital home photography, a wealth of evidence may be readily available in parents’ and friends’ image collections. Not only is a large quantity of images often available, photos of children are taken at different angles and in different environments, with each image a potential test for leukocoria.
Shaw and Hamerly first developed the White Eye Detector app as a test for retinoblastoma — Shaw’s child was diagnosed with the disease. — but since expanded its scope to leukocoria. The app analyzes photos stored in the user’s phone, thus removing the need to upload children’s photos to the cloud and ensuring privacy. The free app, available in Android and Apple/iOS forms, alerts users to evidence of leukocoria by analyzing the stored photos with an algorithm trained by a convolutional neural network, a type of machine learning for image analysis.
In their paper, researchers tested the app’s ability to detect early evidence of leukocoria from family photographs of children. The study team, led by Baylor undergraduate student Micheal Munson, analyzed family photographs from 40 children, 20 children with a leukocoria-related eye disorder and 20 without a disorder. The researchers started with nearly 53,000 family photos of these children from three different smartphone models, two iOS and one Android. From this initial collection, the team analyzed 23,248 images where the child’s face is photographed and eyes are open.
The researchers used White Eye Detector to find evidence of leukocoria in children from their photos over time, with the photos also assessed by three experts for signs of the disorder. For the children with leukocoria, the team also calculated the amount of time between first photographic evidence of the disorder and its clinical diagnosis. The results show the app accurately diagnosed leukocoria in 16 of the 20 children, or 80 percent, considered the “gold standard” threshold for telemedicine. In addition, the app detected leukocoria evidence on average 484 days, or 1.3 years before clinicians spotted the condition.
The app’s developers are refining the White Eye Detector algorithm to reduce false positive readings that can occur from analyzing individual photos. Nonetheless, the results indicate the app is becoming more sophisticated in finding true cases of leukocoria. “We wanted to be able to detect all hues and intensities of leukocoria,” says Shaw in a university statement. “As a parent of a child with retinoblastoma, I am especially interested in detecting the traces of leukocoria that appear as a gray pupil and are difficult to detect with the naked eye.”
More from Science & Enterprise:
- Trial Testing Mobile Telemedicine for Skin Diseases
- New App for Medical Research, Studies Revealed
- Smart Watch, A.I. Analyzing Human Emotions
- Trial to Test App Alerting for Dementia-Linked Drugs
- FDA Clears Wearable App Device for Migraine
* * *
You must be logged in to post a comment.