A team of researchers at IBM has used deep learning and visual analytics technology to detect diabetic retinopathy, with 86% accuracy as to the severity of the condition.
The results of the research were presented at the IEEE's International Symposium on Biomedical Imaging in Melbourne last week. The paper was written by a team consisting of Pallab Roy, Ruwan Tennakoon, Khoa Cao, Suman Sedai, Dwarikanath Mahapatra, Stefan Maetschke, and Rahil Garnavi.
Diabetic retinopathy is a leading cause of blindness and affects one in three of the 422 million who suffer from diabetes globally.
Without treatment, it can lead to permanent blindness but early detection and treatment can reduce the risk of blindness by 95%.
{loadposition sam08}There are five recognised levels of diabetic retinopathy: no DR; mild; moderate; severe; and proliferative diabetic retinopathy.
Color-fundus images of diabetic retinopathy: (a) mild, (b) moderate, (c) severe non-proliferative and (d) proliferative.
Dr Joanna Batstone (below, right), vice-president and lab director of IBM Research Australia, told iTWire that the training was a long process, beginning with training the algorithms developed to recognise the right eye from the left eye.
This was in order that the technology could recognise the progress of the disease in one eye, and was not misled in the event that an image of the other eye was presented.
Then one had to train the system to identify the optic cup and blood vessels, what normal eye looked like, how to compare a normal eye with one that had a mild form of the disease and one that had a severe form.
Dr Batstone said more than 35,000 images from EyePACS, a Web-based application for exchanging eye-related clinical information, were used for training.
Ophthalmologists were able to identify the disease at various stages with an accuracy rate of 91%, she said, and thus what had been achieved using algorithms was reason for celebration.
Since the technology can quickly and accurately identify both the presence and severity of the disease, it could potentially help doctors and clinicians have a better view of disease progression and determine treatment.
Dr Batstone said while she was unable to give any kind of timeline as to when the technology would be available for widespread use, it was likely to be supplied as a service from the cloud when it was ready.
Since the human eye was the same no matter the ethnicity and geography of an individual, she said that the technology could be used worldwide with no need for anything beyond what was done to get it ready for first use.
IBM has 12 research labs worldwide and Dr Batstone said the Australian lab, which was set up six years ago, had been concentrating on retinal and skin imaging. The lab team had been in place for three or four years.
She said the approach taken by the lab was a multidisciplinary one, and computer experts worked in tandem with clinicians on projects such as this.
No special computing power was needed to train the algorithms, Dr Batstone said, adding that it was all done on ordinary laptops.
Images: courtesy IBM Research Australia.