Researchers at the University of Sydney in Australia and Boston University in the United States have developed an artificial intelligence (AI)-based tool capable of detecting Parkinson’s disease even years before the first symptoms appear. Currently, there are no laboratory tests to diagnose non-genetic cases of Parkinson’s, which means that the disease is diagnosed when physical symptoms, such as resting tremor of the limbs, have already emerged.
To overcome this limitation, the scientists created a machine-learning neural network system that analyzes chemical biomarkers in the blood, known as metabolites, that the body produces as it breaks down food, drugs, and other substances. In a study published in ACS Central Science, they examined blood samples from 39 individuals who were healthy at the time of collection and who developed Parkinson’s 15 years later.
By comparing the metabolites detected in the samples from the Parkinson’s patients with those from the control group, the researchers identified unique combinations of biomarkers that could prevent or alert to a predisposition to develop the disease. They found that samples from patients who developed Parkinson’s had low concentrations of triterpenoids, a neuroprotectant that regulates oxidative stress.
The tool developed, called CRANK-MS, is based on neural networks that use mass spectrometry to identify combinations of metabolites that might have been missed by conventional statistical methods. In initial tests, CRANK-MS was able to detect chemicals associated with Parkinson’s disease with 96% accuracy.
However, the researchers note that larger validation studies are required in different parts of the world before the tool can be used reliably. Although the results are promising, further research is needed to ensure its efficacy and applicability in a broader clinical context.