Parkinson’s Disease (PD) is the fastest-growing neurological disease in the world, with more than one million people living with it in the United States alone. Unfortunately, reliable and effective progression biomarkers to track the severity of PD over time have yet to be developed, leading to greater reliance on patient self-reporting or qualitative rating by a clinician. Various measures, such as cerebrospinal fluid and blood biochemical values for the diagnosis of PD, have been put forward in the literature, but further research is needed.
Interestingly, a relationship between PD and breathing was noted as far back as 1817 by James Parkinson himself, who structured his findings into what is now known as Parkinson’s Law:
“The disease begins with barely perceptible shake which gradually increase and accumulate until every part of the body sympathetic to this disorder becomes affected.”
His work was later refined by other scholars and scientists until today, when we understand that tremors or rigidity are indeed early signs of PD. There have been no effective biomarkers for diagnosing PD, but that may be about to change through digital biomarkers found by an artificial intelligence model. The same logic is spreading beyond neurology: in cardiology, AI can now read a routine ECG image to flag a hidden heart disease.
Artificial Intelligence Assessment
According to a promising new study, breathing patterns could be remotely tracked to identify Parkinson’s disease. The study, conducted by MIT researchers, reveals an AI system that uses radio waves to track breathing while a person is sleeping.
The link between breathing and the brain runs deeper than that. A 2026 study found that the brain follows the shape of each individual breath, with neural activity in memory and emotion regions mirroring every breath’s curve, which helps explain why breathing-based detection can work at all.
Dina Katabi, the study’s primary investigator, said the study was motivated by findings made by James Parkinson, the first doctor to clinically document symptoms of the degenerative neurological condition, 200 years ago.
Katabi explained:
“A relationship between Parkinson’s and breathing was noted as early as 1817 in the work of Dr. James Parkinson. This motivated us to consider the potential of detecting the disease from one’s breathing without looking at movements. In addition, some medical studies have shown that respiratory symptoms manifest years before motor symptoms, meaning that breathing attributes could be promising for risk assessment prior to Parkinson’s diagnosis.”
The datasets used for model training in this study vary significantly in origin and content. The Mayo Clinic, Massachusetts General Hospital (MGH) sleep lab, observational PD clinical trials sponsored by the Michael J. Fox Foundation (MJFF), and the National Institutes of Health (NIH) Udall Center were among the numerous sources from which data was collected. Additionally, public sleep datasets from the National Sleep Research Resource, such as the Sleep Heart Health Study (SHHS) and the MrOS Sleep Study (MrOS) 26, were included in the collection of data.
The first phase involved examining the breathing patterns of 757 Parkinson’s disease patients and roughly 7,000 healthy control volunteers throughout nearly 12,000 nights. Next, the MIT team trained its algorithm through a neural network using this sizable dataset of nighttime breathing.
With just one night’s worth of data, the AI model could diagnose Parkinson’s patients with 86% accuracy. In addition, the study discovered that a diagnosis of Parkinson’s could be made with an average accuracy of 95% after 12 nights of continuous tracking. Who would have thought that a synthetic device modeled after the human brain would eventually use sensors to detect clinical diagnosis in an actual human brain?
Early AI Detection for Parkinson’s
What’s more intriguing is that the system may be able to detect Parkinson’s disease before any motor symptoms manifest. The analyzed dataset contained information on participants before and after a Parkinson’s diagnosis. The two sleep sessions were roughly six years apart. Before a patient was diagnosed with Parkinson’s, the AI model could predict Parkinson’s in the undiagnosed population with 75% accuracy from the first sleep data set.
The MIT researchers said:
“Currently, PD diagnosis is based on clinical motor symptoms, which are estimated to develop after 50–80% of dopaminergic neurons have already degenerated. Our system shows initial evidence that it could potentially provide risk assessment before clinical motor symptoms.”
Of course, further research is required to confirm the system as an early diagnosis tool, but monitoring illness progression may be a more immediate use for the technology. For example, other data analyzed in the study demonstrated that the AI model could monitor a Parkinson’s patient over a year and link changes in breathing patterns to escalating disease severity.
The Future Role of AI in Medicine
According to Katabi, this may be useful in several situations, such as detecting other neurological diseases, enhancing clinical care for patients who live in remote areas, and helping researchers determine the efficacy of novel pharmacological treatments in clinical trials.
Katabi said:
“In terms of drug development, the results can enable clinical trials with a significantly shorter duration and fewer participants, ultimately accelerating the development of new therapies. Finally, in terms of clinical care, the approach can help assess 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.”
Even though it is in its early stages, the researchers have already created a wall-mounted device that may be used to monitor patients at home. In the end, this type of gadget could serve as an early warning system for those with a higher-than-average risk of acquiring Parkinson’s disease or for patients in the early stages of the disease who desire close monitoring of the disease’s development.
The researchers speculated:
“We envision that the system could eventually be deployed in the homes of PD patients and individuals at high risk for PD (for example, those with LRRK2 gene mutation) to monitor their status and provide feedback to their provider. For example, suppose the model detects severity escalation in PD patients or conversion to PD in high-risk individuals. In that case, the clinician could follow up with the patient to confirm the results via telehealth or a visit to the clinic.”
The study was published on August 22, 2022, in Nature Medicine.
