The promise of Big Data is that it will be able to learn quickly from huge data sets and find correlations that might escape the notice of a human analyst. If applied to medical applications, this could lead to faster and more accurate diagnoses, which in turn could result in earlier treatments that could have better outcomes at a lower cost. One of the largest sources of big data sets in medicine is from digital imaging. The benefit of these imaging systems is that they give doctors a lot more information to consider; the downside is that there is a lot more information to consider and it is easier to miss something.
DeepMind is the artificial intelligence spinoff of Alphabet (the parent company of Google), and they have an entire division devoted to health applications. In a recent paper published in Nature Medicine, researchers describe their progress in processing eye images rapidly to identify potential disease for referral to physicians. In partnership with Moorfields Eye Hospital in London, UK, they have been developing a system that reads images from optical coherence tomography (OCT) scans. This technology creates a 3D image of the back of the eye, but this is complex information that can be difficult to interpret accurately. A backlog of images can mean that diagnosis and treatment can be delayed for patients, which can result in vision loss that could have been prevented.
The system uses two separate neural networks to analyze the scanned information. The first looks for indications of disease and damage, breaking its findings down into different categories. The second neural network then develops potential diagnoses based on this information. Rather than just present a “black box” result, however, this second system rates the confidence level of its judgment, giving clinicians valuable insight into how the decisions were made. And unlike human analysis, the entire process takes just seconds.
If a diagnosis indicates disease or damage, the results are flagged and referred to human physicians. This facilitates triage procedures so that patients with the most urgent needs are seen sooner. The system can identify 50 different eye diseases and conditions and tests show that it is just as accurate as the best eye doctors in the world. The initial results are still preliminary, but the company hopes to test it through rigorous clinical trials. The ultimate goal is to make the system available at no cost to 30 hospitals in the UK’s National Health System. The refined data set used to train the AI system is also being made available to other researchers for non-commercial studies.
