In America, congenital cataracts—a clouding of the eye lens at birth that can lead to blindness—are vanishingly (and thankfully) uncommon. Like tooth decay or tetanus, better screening and technologies have led to earlier diagnoses, and the problem can largely be cured with surgery. But in developing countries, a lack of widespread expertise and resources mean that hundreds of thousands of children are now blind due to this treatable disease.
"Missed or mistaken diagnoses, as well as inappropriate treatment decisions, are common among rare-disease patients and are contrary to the goals of precision medicine, especially in developing countries with large populations, such as China," write a group of Chinese researchers in a study published Monday in the journal Nature Biomedical Engineering.
These researchers aim to fix that preventable treatment gap by using eagle-eyed AI. The researchers outline an artificial intelligence program that can diagnose congenital cataracts more accurately than human doctors, and report that the data it collects could help spur new research on how to treat this rare disease.
Aging is the most common cause of cataracts, but roughly 5 to 20 percent of childhood blindness is caused by congenital cataracts. Though the disease is curable with surgery, if not fixed soon enough, it can lead to lazy eye as the brain and eye don't work properly together while the child grows. In China, roughly 30 percent of childhood blindness is due to this form of the disease.
In 2010, the cataract crisis in China prompted the founding of the Childhood Cataract Program of the Chinese Ministry of Health, according to study co-author Haotin Lin. The program has collected data on thousands of cases of congenital cataracts, Lin said, but the dataset had yet to reach its full potential. So, inspired by the DeepMind project that built an AI program that could beat professional players at classic video games, Lin and his team decided to use their data to an AI opthamologist.
"Since AI can play games against human players, why not create an AI that could act equally as a qualified human doctor?" Sun Yat-Sen University ophthalmology researcher Lin said of his team's thinking.
Working with a team from Xidian University for two years, the researchers were able to build CC-Cruiser, an AI program trained to scrutinize images of eyes to detect the presence of cataracts and recommend whether surgery is necessary. In a test alongside human ophthalmologists, CC-Cruiser successfully identified every case of congenital cataracts out of a group of 50 images of patients. Meanwhile, the ophthalmologists missed several cases and misdiagnosed several false positives, the researchers report in their new study.
"Humans tend to be [either] somewhat conservative or radical due to their own experience and personality, and the machine's advantage is its objectivity," Lin says. "We [believe] that deep learning results collaborating with human analysis will achieve a better health care quality and efficiency."
But Lin and his team’s vision goes further: They see the CC-Cruiser as a model for harnessing the power of big data to help improve research and treatment of congenital cataracts.
Because congenital cataracts can present in a variety of ways, pooling data from cases worldwide can give computers and doctors a better sense of how to approach the disease, the researchers report. Thus, the researchers have built CC-Cruiser as a cloud-based AI that could be accessed by doctors at hospitals around the country. Doctors would be able to upload patient images into the system, and the AI would evaluate the images to diagnose or rule out congenital cataracts.
If the AI detects the disease and determines that immediate surgery is required, an emergency notification would be sent to CC-Cruisers creators to confirm the diagnosis, which would then be sent back to the patient's doctor. Meanwhile, CC-Cruiser would continue to collect data that doctors and scientists could use to further improve the AI and use to study variations and treatment options for congenital cataracts.
Moreover, the CC Cruiser could pave the way for sussing out even rarer diseases when countries and institutions lack specific expertise. "The limited resources of patients and the isolation of the data in individual hospitals represent a bottleneck in data usage," Lin said. "Building a collaborative cloud platform for data integration and patient screening is an essential step."