Google AI detects diabetic retinopathy

Machine learning, a type of artificial intelligence (AI), is a discipline within computer science that teaches machines to detect patterns in data. Deep learning involves training an algorithm, or ‘neural network’, to perform a specific task by learning from a large set of examples, roving the need to explicitly specify rules.{{quote-A:R-W:450-I:2-Q: Diabetic retinopathy is the fastest growing cause of blindness and that, if caught early, is treatable. -WHO:Ms Lily Peng, Product Manager of Google}}Google trained the algorithm by using 128,175 images (the ‘development set’) obtained from EyePACS in the US and three eye hospitals in India.The performance of the algorithm was assessed by applying it to two data sets (‘the validation sets’), consisting of 9,963 images taken at EyePACS screening sites and 1,748 images that have been publically available.All of the images in the development and clinical validation sets were graded for the presence of diabetic retinopathy, diabetic macular eda and image quality by ophthalmologists.Comparisons of the grades given by the algorithm with those given by the ophthalmologists showed that the performance of the algorithm was “on par” with that of the ophthalmologists.In a Google Research blog post published on the same day as the JAMA article, Ms Lily Peng Google product manager, and Mr Varun Gulshan Google research engineer, and co-authors of the paper, explained that diabetic retinopathy is the fastest growing cause of blindness and that, if caught early, is treatable.“Unfortunately, medical specialists capable of detecting the disease are not available in many parts of the world where diabetes is prevalent. We believe that machine learning can help doctors identify patients in need, particularly among underserved populations,” Peng said.{{image3-a:l-w:400}}Gulshan and Peng noted “there is still a lot of work to do”, adding that they are working with retinal specialists to define even more robust reference standards that can be used to quantify performance.Furthermore, they said that their colleagues at DeepMind Health, which uses AI to bring benefits to medical research, are working on applying machine learning to 3D imaging technology.The research paper for this project is available to view on the Journal of the American Medical Association website.