By applying machine learning and computer vision expertise, the researchers designed a new deep learning algorithm that can interpret indications of Diabetic retinopathy (DR) in retinal photographs, enabling doctors to screen more patients in settings with limited resources.

With the support of doctors in India and the US, the researchers created a development dataset of 128,000 images that were each assessed by three to four ophthalmologists from a panel of 54 ophthalmologists.

The dataset has been used to train a deep neural network to detect referable diabetic retinopathy.

Later, the researchers tested the algorithm’s performance on two separate clinical validation sets, including 12,000 images. The majority decision from a panel of seven or eight US board certified ophthalmologists serving as the reference standard.

The ophthalmologists selected for the validation sets were the ones, which showed high consistency from the original group of 54 doctors.

According to Google, the two complementary methods are expected to be used together to assist doctors in the diagnosis of a wide spectrum of eye diseases in the future.

The researchers are currently working with doctors and researchers to study the entire process of screening in settings across the globe.

In addition, the researchers are working with the FDA and other regulatory agencies to further assess these technologies in clinical studies.

Image: Examples of retinal fundus photographs that are taken to screen for DR. Photo: courtesy of Google.