Sometime in the not-too-distant future, your doctor may be using artificial intelligence to help out with your visit. Sounds futuristic, right? Well, Google just made it happen.
An algorithm created by Google recently learned how to detect diabetic retinopathy, a condition that causes loss of vision in diabetes patients.
The algorithm learned how to spot the disease on its own using the same method Google uses to index images on the Web.
A Breakthrough for Medical AI
The tech giant trained the algorithm with more than 128,000 retinal images, which had been classified by ophthalmologists. The researchers then tested the system and found it recognized the disease and rated its severity as well as, and sometimes even better than, the human specialists.
Google published its findings in the Journal of the American Medical Association, an unprecedented sign of acceptance of medical research from tech companies by the medical community. The journal even published an additional article addressing medical professionals, detailing the study and its potential benefits.
About Diabetic Retinopathy
Diabetic retinopathy is the leading cause of vision loss for people with diabetes and all working-age adults. The condition affects the retina, the layer of tissue at the back of the eye that’s sensitive to light. It can cause significant vision loss and even blindness.
The disease usually doesn’t have symptoms in its early stages, meaning regular eye exams are often required to detect it. Early detection can reduce the risk of serious damage. Because AI has now been shown to aid in this detection, it could be especially useful for patients with diabetic retinopathy.
How AI Could Improve Medical Care
Now that Google has shown deep learning to be effective for this particular condition, others can start testing it in additional areas. The potential uses in medicine for AI are plentiful:
- AI algorithms could be useful for analyzing other types of medical imagery too, freeing up doctors to spend more time with patients as opposed to looking through scans.
- Some experts say artificial intelligence could also eventually make medical examinations more accurate by reducing the risk of human error and increasing objectiveness.
- The technology could be especially useful in areas where the required expertise is hard to find. Because of AI, remote, poor or otherwise disadvantaged areas could in the future more easily get the specialized medical care they need.
Challenges to AI in Medicine
This study helps to prove AI can be as effective as a human doctor for detecting diabetic retinopathy, but convincing medical professionals it is consistent and dependable remains one of researchers’ biggest hurdles.
Currently, AI has no way of explaining how it came to a particular conclusion, making it difficult to double-check its accuracy. Google researchers say they’re working on this issue.
Before it’s implemented in a medical setting, the technology will need to pass clinical trials as well. These tests can be quite stringent. For example, a drug must be the subject of an approved marketing application before it’s distributed across state lines. Google says it’s currently working with authorities in the U.S. to design testing.
Similar Projects
Google is also working on algorithms for detecting other types of eye disease and on using deep learning in cancer treatment.
Google isn’t the only one working on projects like these. Another California company, Flow Health, just signed a five-year agreement with the U.S. Department of Veteran Affairs on a medical deep learning project. Through this project, the two groups plan to use AI to analyze data from 22 million veterans in order to improve diagnoses, inform personal care plans and learn more about the link between genomes, phenotypes and health.
It will take a bit of time for AI to begin to be integrated into healthcare, but it may happen sooner than many think. There’s still plenty of work to be done, but the medical community is beginning to open its eyes to artificial intelligence’s potential benefits.
Edited by
Alicia Young
