target eye disease

The potential for artificial intelligence (AI) to diagnose and treat health conditions continues to gain momentum, as a new study shows how the technology can speed diagnosis and treatment of eye diseases.

A paper published February 22 in the journal Cell describes how AI can be applied to patients with retinal diseases. The research, led by Kang Zhang, MD, PhD, professor of ophthalmology at Shiley Eye Institute at the University of California in San Diego, demonstrates that a computer could learn to accurately and reliably recognize such common eye diseases as macular degeneration and diabetic retinopathy.

“This is about trying to teach a computer what an image is and how to make a decision about what they’re seeing,” Dr. Zhang explains. “The goal is for the computer to be as good as the specialist who went to medical school and is highly trained in medical diagnostics and treatment.”

While it can take a specialist decades of practical experience to reach the highest levels of expertise, he adds, “we’re seeing a computer can recognize these things after a few days.”

The paper follows other recent studies that show deep-learning computers may have a legitimate place in healthcare, says Rahul Khurana, MD, an ophthalmologist in Mountain View, California, and a clinical spokesperson for the American Academy of Ophthalmology.

“This kind of technology is very accurate for patients with certain types of conditions,” Dr. Khurana says. “That is creating some excitement in the field.”

Diagnosing Macular Degeneration, Diabetic Retinopathy

In the new paper, Zhang and his colleagues in China, Germany, and Texas first fed images of the eye disorders into the computer. The images were taken with an imaging technique known as optical coherence tomography. This newer, revolutionary diagnostic technology uses light waves to take high-resolution, cross-section images of the eye to give doctors a way to map and measure the retina in detail.

The scans are used to help spot common conditions like macular degeneration, in which a part of the retina called the macula deteriorates, and diabetic retinopathy, a complication of diabetes that causes the blood vessels in the retina to swell and leak fluid. Both are dangerous conditions that can cause blindness if they are not diagnosed and treated promptly.

Current computational approaches require millions of images to train a computer. Zhang’s research used an AI-based “convolutional neural network” that required a much smaller dataset of only 200,000 optical coherence imaging scans.

“The computer is learning the normal map of the eye,” Zhang says. “We give it a variety of pictures to learn and memorize. We teach, for example, ‘if this spot is here, it’s going to be macular degeneration.’ The beauty of this is instead of having the computer learn by itself, we can tell them what to look for. This is about designing computer software to make computers think like a human being.”

The computer was able to generate a decision on whether a patient should be referred for treatment within 30 seconds and with 95 percent accuracy.

The study demonstrates that neural networks can assist physicians and perhaps even outpace them with the ability to remember so much data. Such technology will have uses throughout the world, Zhang predicts. In resource-rich countries like the United States, it can hasten the critical time between signs of disease and treatment.

“A patient with possible macular degeneration may need to be treated within a month, but referrals and appointments can end up taking several months. That can delay the diagnosis and treatment,” he says.

Treating Patients Where Specialists Are Scarce

In resource-poor areas, the technology can help patients who might otherwise receive no care because of the scarcity of physicians. Zhang and his colleagues will take their neural network to Haiti this summer to assess its utility. The region has a large population of people with diabetes who are at risk for retinopathy, but it has fewer than 60 ophthalmologists.

“The ability to do this will, hopefully, give more patients access to the healthcare system because we can diagnose conditions earlier,” Khurana says, noting there are approximately 415,000 people living with diabetes worldwide who are at risk for diabetic retinopathy. “Whenever we have new and improved technology to allow us to make diagnoses faster, better, and make care more accessible to the broader population, it’s a win-win for patients and doctors.”

Getting Doctors to Trust Computers

Challenges remain in implementing AI-based networks in healthcare, Zhang notes. Doctors have to trust their computer assistants. In the study, Zhang and his colleagues also asked the computer to explain its diagnosis, identifying the regions of the eye that were recognized and were the basis for the machine’s conclusion.

“The computer doesn’t just spit out a diagnosis. It explains why it made the diagnosis and recommendation it did,” he says. “That makes this more transparent and helps the physician trust the computer more. That way, this isn’t just a black box, and you have no idea why it gives the diagnoses it does.”

Other Uses for Artificial Technology

AI-based networks have vast potential in healthcare imaging. Zhang also showed that the system could distinguish between viral and bacterial pneumonia in children by examining X-rays. While viral pneumonia may require no treatment, a patient with bacterial pneumonia requires prompt antibiotic treatment to prevent serious complications of the disease.

“We’re seeing a variety of medical fields where artificial intelligence is being used more and more,” Khurana says. “I think it’s a very exciting time for the field of artificial intelligence and its applications in medicine.”


Source: https://www.everydayhealth.com

For more helpful eye care and vision care tips, please visit our main blog page.

Leave a Reply

Your email address will not be published. Required fields are marked *