DeepMind's AI detects over 50 eye diseases with 94% accuracy, study shows
- Google's DeepMind artificial intelligence system can detect and recommend how patients should be referred for more than 50 eye diseases with the same accuracy as a physician, according to clinical results published online in Nature Medicine.
- Google partnered with Moorfields Eye Hospital in the U.K. two years ago to explore whether machine learning technology can be used to analyze 3D eye scans, leading to earlier diagnosis and interventions for degenerative eye diseases.
- The results, while early, show the software is capable of handling the wide range of patients that are seen in routine clinical practice and hasten analysis of scans, over 1,000 a day at Moorfields, a DeepMind Health blog post says.
The research, a collaboration of DeepMind, Moorfields Eye Hospital and University College London, involved training the software on 14,844 scans from about 7,500 patients with sight-threatening retinal diseases. These were then analyzed alongside expert physician diagnoses. All of the patients had been treated at Moorfields-operated facilities.
The AI system recommended the proper referral for over 50 eye diseases with 94% accuracy, on par with eye disease experts.
"Our work opens up the possibility of testing the clinical applicability of this approach in other global settings and clinical pathways, such as emergency macular assessment clinics in the UK NHS, triage and assessment in community eye care centers and the monitoring of disease during treatment regimes," the authors wrote.
The next step, they said, is to test the technology in a randomized clinical trial.
This is some good news for AI in healthcare. As more digital health technologies become available, providers are looking for clinical evidence that new tools like machine learning do what their developers profess they do and actually improve patient outcomes.
When it comes to digital health, "highly prized qualities would be solutions that save time, decrease the physician workload, improve interaction with technologies and harness technologies to aid the clinician in diagnosis or treatment that ultimately leads to better outcomes for patients," Kirti Patel, a physician and digital health startup adviser based in Massachusetts, told Healthcare Dive via email last year.
The DeepMind results come as Google is focusing on healthcare. In December, the tech giant launched Deep Variant, an open source tool that uses AI to create an image of a person's genetic blueprint using sequencing data. The goal is to pinpoint specific genes or gene mutations that can help providers more effectively manage disease states.
The company is also working on projects to release sterilized mosquitoes to reduce the spread of infectious diseases, improve diabetes management and solve the interoperability conundrum by powering a healthcare data infrastructure layer, according to a recent CB Insights report. And through acquisitions, it is in a position to have an impact on conditions ranging from cancer to behavioral health and aging.