A collage of a scan of the human brain and blood samples to detect Parkinson'sShare on Pinterest
New AI tools may aid the early diagnosis of Parkinson’s disease. Design by MNT; Photography by EDUARD MUZHEVSKYI/SCIENCE PHOTO LIBRARY/Getty Images
  • Researchers developed an AI tool to predict Parkinson’s disease from blood samples.
  • The tool can predict Parkinson’s 15 years before symptom onset with 96% accuracy.
  • The underlying technology could possibly be adapted for diagnosing other conditions.

Parkinson’s disease is a neurodegenerative condition characterized by unintended movements such as shaking, stiffness, and difficulty balancing. It is caused by the loss of nerve cells in the brain, leading to reduced levels of dopamine, which plays a key role in movement.

Around 90,000 people in the United States are diagnosed with Parkinson’s, and over 10 million people live with the condition worldwide. Parkinson’s is the second most common neurodegenerative condition after Alzheimer’s disease, and cases are growing more rapidly than other neurological conditions.

Currently, Parkinson’s is diagnosed based on symptoms, medical history, and a physical examination. There are no tests for Parkinson’s.

Tests that can detect Parkinson’s could improve care and management strategies for the condition.

Recently, researchers developed an AI tool that can predict signs of Parkinson’s from patients’ blood samples up to 15 years before symptom onset.

“This tool will be a game changer for Parkinson’s diagnosis by providing an objective, reliable, and highly accurate way to predict conversion in patients at risk of developing the disorder,” Dr. Mya Schiess, neurology professor and director of the Movement Disorders & Neurodegenerative Diseases Program at UTHealth Houston, who was not involved in the study, told Medical News Today.

The study was published by the American Chemical Society.

For the study, the researchers gathered healthcare data from 78 individuals from Spain. Each participant provided a blood sample between 1993 and 1996 and was followed for 15 years. Among the participants, 39 eventually received a diagnosis of Parkinson’s, whereas 39 did not.

The researchers next utilized an AI tool called CRANK-MS to analyze the data. In particular, they compared different combinations of metabolites—chemicals the body creates when breaking down food, drugs, chemicals, and its own tissue—among those who developed Parkinson’s and those who didn’t.

“Typically, researchers using machine learning to examine correlations between metabolites and disease reduce the number of chemical features first, before they feed it into the algorithm,” W. Alexander Donald, associate professor in the School of Chemistry at the University of New South Wales Sydney, Australia, one of the study’s authors, said in a press release.

“But here, we feed all the information into CRANK-MS without any data reduction right at the start. And from that, we can get the model prediction and identify which metabolites are driving the prediction the most, all in one step. It means that if there are metabolites which may potentially have been missed using conventional approaches, we can now pick those up,” he added.

Using CRANK-MS, the researchers identified metabolite combinations unique to participants who developed Parkinson’s. They identified 96% of people who developed Parkinson’s cases within 15 years from these combinations alone.

The researchers noted that their new tool was more accurate in diagnosing Parkinson’s than existing clinical assessments by movement disorder specialists, which have an accuracy of 80%.

The researchers also reported that CRANK-MS could also diagnose Parkinson’s 84.3% of the time from skin sebum samples when they tested the algorithm with a second cohort of 274 NHS patients.

MNT spoke with Daniel Truong, neurologist and medical director of The Parkinson’s and Movement Disorder Institute at MemorialCare Orange Coast Medical Center, who was not involved in the study, about how CRANK-MS could improve Parkinson’s care.

“The AI tool demonstrated the ability to detect signs of Parkinson’s disease up to 15 years before the onset of symptoms, providing the opportunity for intervention and treatment at an earlier stage,” he noted.

He continued that the tool’s high level of accuracy could also allow potentially reliable diagnostic tools for early identification of risk factors and that the tool may have the potential for diagnosing other conditions too.

Dr. Schiess called the results exciting but called for caution.

“[C]onfirming these findings in larger cohorts is crucial before this can be widely used in clinical practice. Furthermore, although the CRANK-MS tool is publicly available, physicians and researchers must familiarize themselves with their use and data interpretation,” she said.

Dr. Truong added that collecting and processing blood samples may not always be convenient or feasible in certain clinical or practical settings and that accessibility to blood samples for early detection in asymptomatic individuals may be challenging.

“While the AI tool can identify unique combinations of metabolites that may serve as potential markers for Parkinson’s disease, the specific biological mechanisms and associations of these metabolites with the disease are not yet fully understood. Further research is needed to validate and elucidate the underlying metabolic pathways and causative relationships,” he continued.

Dr. Julie Pilitsis, a board certified neurosurgeon at Marcus Neuroscience Institute, established at Boca Raton Regional Hospital, part of Baptist Health, also spoke with MNT about the study’s limitations.

She noted that the study uses ‘ideal data’—or data that is collected over time and that has known outcomes that are often not available to AI tools.

“How the tool would perform in a lesser set of data or in a set of data where outcomes have not been determined would need to be assessed,” she explained.

“Since we know the outcome of the patients [15 years after their blood samples were taken], we are able to see that the tool is superb in predicting who may develop Parkinson’s,” said Dr. Pilitsis.

“Most notably there were correlations with lower levels of triterpenoids in patients that developed Parkinson’s and higher levels of polyfluoroalkyl substances—which are seen in industrial compounds—in patients with Parkinson’s. This type of data will allow us as clinicians in the future to guide patients towards what they can do to lower their risk and augment their ability to protect themselves from developing Parkinson’s.”
— Dr. Julie Pilitsis

Dr. Truong noted that the implications of this test range from more personalized treatment plans and targeted public health interventions to the discovery of new biomarkers for Parkinson’s.

He added that the tool could also provide further research opportunities.

The availability of the CRANK-MS tool to researchers allows for further exploration and validation of its performance across different populations and settings. This presents an opportunity for collaborative research efforts, data sharing, and refinement of the tool’s algorithms, potentially advancing our understanding of diseases,” he concluded.