EyePACS Helps Google Rethink the Future of Eye Health

When Google was founded in 1998 “to organize the world’s information and make it universally accessible and useful,” few would have predicted the role it might play in detecting diabetic retinopathy (DR), the leading cause of blindness among working age adults. This week EyePACS CEO Jorge Cuadros, OD, PhD, was featured as an “AI Hero” at the tenth annual Google I/O conference in Mountain View, California as an example of how AI can be used to help tackle important challenges in healthcare such as preventable blindness.

JC Il 2 (1).jpg

                                                                                                                    Each year since 2008 Google has used the I/O Conference to dramatically introduce new features, new tools, and breakthroughs. In 2018, Google invited some individuals outside of Google to share their stories, relating how they were using technology from Google AI to solve important human problems.

The Google-EyePACS partnership goes back to 2015, when a team from Google research used 128,000 retinal images to train a new neural network optimized for image classification. Each image was first graded by at least three U.S. licensed ophthalmologists. Then these graded retinal images were used to “teach” a computer how to detect and grade retinopathy. Tens of thousands of those images, completely de-identified, had been contributed from the EyePACS database which now houses more than two million such images.

Google’s work with EyePACS should come as no surprise. The Google AI web site explains the company’s intention to bring “the benefits of AI to everyone” by “conducting research that advances the state-of-the-art in the field, applying AI to products and to new domains, and developing tools to ensure that everyone can access AI.” Referring specifically to AI’s application to health and bioscience, the site states that “AI is poised to transform medicine, delivering new, assistive technologies that will empower doctors to better serve their patients. Machine learning has dozens of possible application areas and tools that we hope will dramatically improve the availability and accuracy of medical services.”

 Justin Burr of Google (center) gathers the 2018 “AI Heroes” around him. All have employed TensorFlow, Google’s open source machine learning framework, to tackle an important problem. In addition to Dr. Cuadros of EyePACS on the left are Topher White and Ari Siburt in back and Shaza Mehdi and Nile Ravenell in front. White, the founder of Rainforest Connection, invented “The Guardian” device, a system of upcycled cell phones to prevent illegal deforestation in the Amazon. Siburt, a PhD student at Penn State, is using TensorFlow to uncover the origins of the solar system. Shaza and Nile are high school students who have developed PlantMD, an app that lets you figure out if your plant is diseased. 

Justin Burr of Google (center) gathers the 2018 “AI Heroes” around him. All have employed TensorFlow, Google’s open source machine learning framework, to tackle an important problem. In addition to Dr. Cuadros of EyePACS on the left are Topher White and Ari Siburt in back and Shaza Mehdi and Nile Ravenell in front. White, the founder of Rainforest Connection, invented “The Guardian” device, a system of upcycled cell phones to prevent illegal deforestation in the Amazon. Siburt, a PhD student at Penn State, is using TensorFlow to uncover the origins of the solar system. Shaza and Nile are high school students who have developed PlantMD, an app that lets you figure out if your plant is diseased. 

                                                                                                                                                        Working with healthcare providers like EyePACS, the Google team trained a computer algorithm “on par with U.S. board-certified ophthalmologists” that is easily placed in the humblest medical care site anywhere in the world. There a diabetes patient, possibly with little means, faces a cascade of issues in trying to manage a multifaceted disease. A digital retinal camera in the hands of one trained medical assistant is now able to immediately detect signs of retinopathy, raise a red flag in the clinic, and very possibly get that patient to an eye specialist who can actually prevent the blindness so closely associated with diabetes 

As Dr. Cuadros explained in an interview at the Mountain View site, “Since preventable blindness is growing fast globally and is projected to triple by 2050, soon we will simply not have enough trained doctors available to protect the world’s population from blindness.” In his presentation to conference attendees and global news outlets, Cuadros showed images of retinas displaying varying degrees of retinopathy. He referenced the work EyePACS has done in countries around the world, including the Republic of Djibouti, Armenia, Guyana, Latin America, and of course, the United States. He explained how the algorithm can cut hours and even days required for human reading and grading, motivating the patient, who probably has no symptoms of vision loss at this point, to seek preventive treatment.

EP included in Verily.jpg

Verily, the healthcare and life sciences arm of Alphabet, is partnering closely with Google AI to develop a medical device based on its machine learning model and will work with regulators to secure approval for use by physicians. In a Google I/O presentation, Verily discussed the importance of cross-industry collaboration with organizations like EyePACS to bring these tools to market. Google AI and Verily have twice published reports on the use of artificial intelligence to detect DR through its work with EyePACS. The 2016 report, “Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photography” (JAMA December 2016), stated that “an algorithm based on deep machine learning had high sensitivity and specificity for detecting referable diabetic retinopathy. Further research is necessary to determine the feasibility of applying this algorithm in the clinical setting and to determine whether use of the algorithm could lead to improved care and outcomes compared with current ophthalmologic assessment.”

Since then the algorithm has been tested under clinical trial by EyePACS and other research groups along with Google. A 2018 report by Google Research, “Grader Variability and the Importance of Reference Standards for Evaluating Machine Learning Models for Diabetic Retinopathy” (Ophthalmology 2018), announced that “adjudication reduces the errors in DR grading. A small set of adjudicated DR grades allows substantial improvements in algorithm performance. The resulting algorithm's performance was on par with that of individual U.S. Board-Certified ophthalmologists and retinal specialists."

Dr. Cuadros commented on the future of AI and EyePACS to the folks at Google and Verily: “We are very excited about the opportunity to work with Google and we appreciate adapting Google’s tremendous expertise to help us on the front lines of preventing vision impairment.”