In Parkinson’s disease, brain cells become damaged or die in the part of the brain that produces dopamine, a chemical needed to produce smooth, purposeful movement. Over time, this damage leads to unintended or uncontrollable movements such as shaking, stiffness, and difficulty with balance and coordination.
Symptoms of Parkinson’s disease usually begin gradually and worsen over time. Currently, there aren’t any markers that can be easily measured in the blood or in imaging tests to diagnose the condition. This has greatly slowed the development of treatments. It also means that people may wait years to get a diagnosis.
NIH-funded researchers led by Dr. Dina Katabi from the Massachusetts Institute of Technology have been testing ways to use artificial intelligence to diagnose Parkinson’s disease.
In a new study, they designed a computer program based on a model of how the brain works (called a neural network) to analyze breathing patterns collected during sleep. The areas of the brain that control breathing and sleep tend to be affected early in the course of Parkinson’s disease. The team tested whether differences in nighttime breathing patterns could be used to distinguish people with the disease from those without.
The researchers used two types of sleep data—breathing patterns and brain activity—to test their program. In the most common test, called a polysomnogram (PSG), people wore a chest belt while they slept to measure breathing patterns. These are often done to help determine why someone is having trouble sleeping. A second group of people participated in tests of a wireless sleep monitoring system. This system bounces radio signals off the body during sleep and uses that information to record breathing patterns without physical contact.
Using sleep breathing data from more than 7,600 people, including 757 with Parkinson’s disease, the team tested their program’s ability to diagnose and track Parkinson’s disease in several datasets from multiple hospitals. The results were published on August 22, 2022, in Nature Medicine.
Using a single night of PSG breathing data, the program correctly identified the people with Parkinson’s about 80% of the time. That number rose to 86% when the program used a single night of wireless breathing data. Adding additional nights of wirelessly-collected breathing data increased accuracy. With 12 nights of data, the program’s ability to identify Parkinson’s disease rose to 95%.
In a small group of people who participated in at least two sleep studies, one before they were diagnosed with Parkinson’s disease, the program identified three-quarters of them as having the disease from the data collected before their official diagnosis.
The team also tested the ability of the program to track whether the disease got worse over time. The scales currently used to measure disease progression in the clinic are relatively insensitive. They can also provide different results when used by different doctors. Compared with two different scales, the program was better at identifying small changes in Parkinson’s symptoms.
If these results are confirmed, this program could help in the early detection of Parkinson’s disease and enable shorter clinical trials with fewer participants, accelerating the development of new therapies. More work will be needed to test the program in larger, diverse populations. Katabi points out, “The approach can potentially help in the assessment of Parkinson’s patients in traditionally underserved communities, including those who live in rural areas and those with difficulty leaving home due to limited mobility or cognitive impairment.”
