Cradle app scans children’s photos to spot eye disorders
A smartphone app designed to look for early indicators of eye disorders in children has been able to spot the signs more than a year before actual diagnosis, according to the scientists who created it.
The app, called Cradle, automatically scans through photos stored in the phone to spot leukocoria – an abnormal white reflection from the retina of the eye.
Leukocoria is one of the primary signs of retinoblastoma – a rare form of cancer in the retina.
It is also an indicator for other eye disorders in children, such as infant cataract and Coats disease – an abnormality of the blood vessels in the back of the eye.
The researchers said their findings could lead to speedy diagnosis and treatments by catching these signs more frequently than routine checks.
They added that the app, which is available on Android and iOS, is “totally private” and does not track the users’ activities.
Dr Bryan Shaw, an associate professor at Baylor University in Texas, US, told the PA news agency: “The software is downloaded to the phone, so you don’t upload images to us to check them on a cloud server.
“We have no way of tracking you or seeing what you are doing with the app.
“If the app does detect white eye, we give you the option to upload the photo to us for research purposes.”
Cradle was first released in 2014 but the researchers said their work, published in the Science Advances journal, is “the first real world test of the app on pictures of kids with eye disease”.
Dr Shaw said: “We go back and look at family photo libraries of 20 kids with different disorders collected over several months or years.”
GPs and opticians look for signs of eye disorders during children’s routine appointments by performing red reflex tests – where an instrument is used to shine light into the pupils to spot eye problems.
A red reflection means everything is normal but if a white reflection is seen, it could be a sign of an eye problem.
However, sometimes these routine checks can fail to spot white eye as the pupil needs to be dilated to allow a clear view of the retina at the back of the eye.
Ophthalmologists are better at spotting white eye as they use eye drops to increase the size of pupils when carrying out the red reflex tests.
Dr Shaw came up with the idea for the app after his son was diagnosed with retinoblastoma in both his eyes when he was four months old.
He told PA: “His mum noticed ‘white eye’ in his pictures and told the paediatrician. He was diagnosed later that afternoon by an ophthalmologist but up to that point he had passed all of his red reflex tests.”
His son lost his right eye and received radiation therapy to his left.
Dr Shaw added: “We went back and started looking through all of our baby pictures and we noticed that white eye emerged in pictures taken at 12 days old.
“If we had caught the cancer then, he wouldn’t have lost his eye or had his other eye radiated.
“His vision would be much better because the tumours would have been smaller when they were discovered.”
The team led by Michael Munson, a teaching assistant at Baylor University, developed Cradle’s machine learning system to detect white eye.
The software was tested on 52,982 photos of 40 children – 20 with confirmed eye disorders and 20 control subjects.
The system detected white eye in the photos of 16 children with eye disorders an average of 1.3 years before actual diagnosis, the researchers said.
But they noted their app has some limitations – the main one being the inability to distinguish between “physiologic leukocoria”, where white eye is a result of light reflecting off the optic nerve in healthy people and does not indicate any underlying condition, and “pathologic leukocoria”, which is white eye associated with eye disorders.
Dr Shaw told PA: “We find that people with eye disorders exhibit pathologic leukocoria 10 times more frequently than healthy people exhibit physiologic leukocoria.
“But the app can’t distinguish the two – they often look identical.”
Dr Shaw hopes the next version of the app will solve the problem by including a “rate detector”, where the system will be able to ignore the instances of physiologic leukocoria.