Researchers from China have developed an artificial intelligence (AI) platform that can diagnose congenital cataracts as accurately as ophthalmologists.
In a study published in Nature Biomedical Engineering last month, the researchers say that their platform could be applied to other rare diseases in which missed or mistaken diagnoses are common, especially in developing countries with large populations such as China.
Using a database collected under the Childhood Cataract Program of the Chinese Ministry of Health, researchers trained the AI platform, CC-Cruiser, with a dataset of 410 images of congenital cataracts and 476 control healthy eyes, each independently labelled by 2 experienced ophthalmologists.
CC-Cruiser was tested in several complex, real-world settings, including a multihospital clinical trial, a website-based test and a comparative performance test with ophthalmologists.
The comparative test involved 50 cases of various challenging clinical situations, evaluated independently by CC-Cruiser and ophthalmologists with 3 levels of expertise: expert, competent and novice.
The CC-Cruiser successfully diagnosed all congenital cataract cases, while ophthalmologists at all expertise levels missed several cases and misdiagnosed several false positives. The device also successfully suggested treatment, correctly identifying all the patients in need of surgery, with just 5 false positives.
"Humans tend to be [either] somewhat conservative or radical due to their own experience and personality, and the machine's advantage is its objectivity," said coauthor, Professor Lin Haotian of Sun Yat-Sen University. "We [believe] that deep learning results collaborating with human analysis will achieve a better health care quality and efficiency."
Encouraged by the results, Lin and his team have built a collaborative cloud platform that can be accessed by doctors at hospitals around the country, allowing them to upload patient images into the system. CC-Cruiser will benefit from the continued data collection, further improving the AI platform with a larger dataset.
"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."