Summary: Researchers have developed an artificial intelligence model that can detect Parkinson’s disease by analyzing a person’s breathing during sleep. The model also estimates disease severity and monitors progression over time.
Source: MIT
Parkinson’s disease is often hard to diagnose early because current diagnoses rely mostly on visible motor symptoms—tremor, rigidity, and slowed movement—that typically appear years after the disease begins.
Dina Katabi, the Thuan (1990) and Nicole Pham Professor in the Department of Electrical Engineering and Computer Science at MIT and principal investigator at the MIT Jameel Clinic, led a team that created an artificial intelligence model capable of detecting Parkinson’s from a person’s breathing patterns during sleep.
The core of the approach is a neural network trained to recognize subtle differences in nocturnal breathing. The model was developed and trained by MIT Ph.D. student Yuzhe Yang and postdoctoral researcher Yuan Yuan and can determine both the presence of Parkinson’s and the disease’s severity. It can also track changes over time, allowing for continuous, passive monitoring.
Yang and Yuan are co-first authors of the new paper, published in Nature Medicine, with Katabi as the senior author. The research team includes collaborators from Rutgers University, the University of Rochester Medical Center, the Mayo Clinic, Massachusetts General Hospital, and Boston University College of Health and Rehabilitation.
Previous efforts to detect Parkinson’s have explored biomarkers in cerebrospinal fluid and neuroimaging, but these approaches are invasive, expensive, and limited to specialized centers. They are therefore unsuited for frequent or large-scale testing that could enable earlier diagnosis or ongoing monitoring.
The MIT team demonstrated a noninvasive alternative that can be used nightly at home without any contact. Their device resembles a home Wi‑Fi router but emits harmless radio signals and analyzes their reflections to extract a person’s breathing pattern while they sleep. That breathing signal is fed into the neural network, enabling passive assessment without any effort from the patient or caregiver.
“A relationship between Parkinson’s and breathing was noted as early as 1817 by Dr. James Parkinson,” Katabi says. “Medical studies have shown that respiratory changes can appear years before motor symptoms, which suggests breathing could be useful for early risk assessment before a clinical diagnosis.”
Parkinson’s is the fastest-growing neurological disorder worldwide and the second most common after Alzheimer’s disease. In the United States it affects over one million people and carries a large economic burden. The study tested the device and AI model on thousands of participants—more than 7,600 individuals in total—including 757 people diagnosed with Parkinson’s disease.
Katabi highlights two major implications. For drug development, continuous, objective measures could shorten clinical trials and reduce the number of participants needed, accelerating the evaluation of new therapies. For clinical care, a contactless, at-home tool could extend access to objective assessments for patients in underserved or remote communities and for those with limited mobility or cognitive challenges.
Ray Dorsey, professor of neurology at the University of Rochester and a co-author, notes the importance of evaluating patients in their natural environment. “We have been limited by sporadic, clinic-based assessments,” he says. “This contactless sensor offers objective, real-world measures of how people are doing at home and could reveal aspects of the disease that are otherwise hard to observe.”
About this AI and Parkinson’s disease research news
Author: Anne Trafton
Source: MIT
Contact: Anne Trafton – MIT
Image: The image is in the public domain
Original Research: Open access. “Artificial intelligence-enabled detection and assessment of Parkinson’s disease using nocturnal breathing signals” by Yuzhe Yang et al., Nature Medicine
Abstract
Artificial intelligence-enabled detection and assessment of Parkinson’s disease using nocturnal breathing signals
Reliable biomarkers for diagnosing Parkinson’s disease and tracking its progression are currently lacking. This study presents an AI model trained to detect Parkinson’s and monitor its progression using nocturnal breathing signals.
The model was evaluated on a large dataset collected from multiple U.S. hospitals and public datasets, involving thousands of participants. It achieved an area under the curve (AUC) of 0.90 on held-out test data and 0.85 on external test sets for PD detection. The model’s severity estimates correlate strongly with the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (R = 0.94, P = 3.6 × 10−25).
An attention layer in the model enables interpretation of predictions in relation to sleep architecture and electroencephalogram features. The model can also operate in a fully contactless, at-home setting by extracting breathing signals from radio waves that reflect off the sleeping body.
This work demonstrates the feasibility of objective, noninvasive, at-home assessment for Parkinson’s disease and provides initial evidence that nocturnal breathing patterns may offer value for early risk assessment before clinical diagnosis.