Researchers from Google have used deep machine learning to create an algorithm for the automated detection of diabetic retinopathy in retinal photographs that reportedly performs on par with ophthalmologists, achieving both high sensitivity and specificity.
According to researchers, deep learning is a type of machine learning that allows an algorithm to program itself by learning the most predictive features directly from the images. For the study, the Google team used deep learning to create an algorithm for automated detection of diabetic retinopathy and diabetic macular edema in retinal fundus photos.
A specific type of network optimized for image classification was reportedly trained using a data set of over 128,000 retinal images previously graded for diabetic retinopathy, diabetic macular edema, and image gradability by a panel of U.S. licensed ophthalmologists and senior residents. Additionally, the algorithm was validated using two separate data sets, both graded by at least seven U.S. board-certified ophthalmologists.
Upon completion of the study, the algorithm reportedly achieved high sensitivities and specificities for detecting referable diabetic retinopathy in both data sets. Google researchers plan to continue their work, including examining how the algorithm could be used for assessing 3D images.
The study was recently published in the Journal of the American Medical Association (JAMA).
Click here to read the full press release.
Click here to read Google's blog post regarding the study.
Source: The JAMA Network & Google