- Diabetic retinopathy is
the leading cause of blindness in adults with type 2
- Diabetic eye disease is screened using retinal
photography with manual interpretation.
- The evaluation of
retinal images using an algorithm based on deep machine learning can
improve early detection and treatment of diabetic retinopathy,
Adults with type 2
are at high risk of
developing diabetic retinopathy - most common diabetic eye disease. People with
diabetes are screened for diabetic retinopathy using retinal photography.
shows that evaluation of retinal photographs using an algorithm based on deep
machine learning had high sensitivity and specificity for detecting referable
‘Algorithm-based on deep machine learning for detection of diabetic retinopathy can improve care and outcomes compared with the current ophthalmologic assessment.’
Grading of Diabetic Retinopathy
is a discipline within computer science that focuses on the development of
computer programs on teaching machines to grow and change when exposed to new
data. Machine learning has been leveraged for a variety of classification tasks
including automated classification of diabetic retinopathy.
An algorithm is
needed to maximize the use of automated grading and detect referable diabetic
retinopathy. Deep learning is a machine learning technique that allows an
algorithm to program itself from the given set of data.
Lily Peng, M.D., Ph.D., of Google Inc., Mountain View, Calif., and colleagues
applied deep learning to create an algorithm for automated detection of
diabetic retinopathy and diabetic macular edema in retinal fundus photographs. The
algorithm was designed to detect specific lesions or predict the presence of
any level of diabetic retinopathy.
Automated Grading of Diabetic Retinopathy
- Early detection and treatment of
- Increases efficiency and coverage of
- Reduces barriers to
- Improves patient outcomes
used a data set of 128,175 retinal photographs for image classification. The
images were graded 3 to 7 times for diabetic retinopathy, diabetic macular
edema by a panel of 54 U.S. licensed ophthalmologists and ophthalmology senior
residents between May and December 2015.
was validated using 2 separate data sets (EyePACS-1, Messidor-2), both graded
by at least 7 U.S. board-certified ophthalmologists.
Both the data
sets were used to analyze the prevalence of referable diabetic retinopathy
(RDR), which is defined as moderate and worse diabetic retinopathy, referable
diabetic macular edema, or both.
data set consisted of 9,963 images from 4,997 patients and 8 percent of fully
dataset had 1,748 images of 874 patients and 15 percent of fully gradable
achieved high sensitivities for detecting referable diabetic retinopathy (97.5
percent [EyePACS-1] and 96 percent [Messidor-2]) and specificities (93 percent
and 94 percent, respectively) for detecting referable diabetic retinopathy.
algorithm for the detection of diabetic retinopathy offers several advantages
such as consistency of interpretation, high sensitivity, specificity and near
instantaneous reporting of results.
noted that, "These results demonstrate that deep neural networks can be
trained, using large data sets and without having to specify lesion-based
features, to identify diabetic retinopathy or diabetic macular edema in retinal
fundus images with high sensitivity and high specificity."
research is necessary to determine the feasibility of applying this algorithm
in the clinical setting and to determine whether the use of the algorithm could
lead to improved care and outcomes compared with the current ophthalmologic
The study is
published in JAMA.
is a condition caused by damage
to the blood vessels of the retina. More than 285 million people have
diabetes and over one-third have signs of diabetic retinopathy.
About 29 percent
of the Americans with diabetes have diabetic retinopathy. Adults with no
retinopathy or mild diabetic retinopathy should go for annual screening. Adults
with moderate diabetic retinopathy should go for screening every six months.