Scientific model: neural network will detect early signs of Parkinson's disease by EEG

Scientists have developed a neural network to verify the electrical activity of the brain of patients with Parkinson's disease. The model recognizes the frequency anomalies characteristic of this pathology based on the results of electroencephalography (EEG), and its accuracy already reaches 97% today. In the future, it is planned to create the first digital service for early diagnosis of Parkinson's disease based on EEG data. With its help, doctors will be able to quickly and accurately make a preliminary diagnosis and prescribe studies to confirm it.

How is Parkinson's disease diagnosed?

In modern clinical practice, Parkinson's disease is detected based on clinical symptoms — decreased motor activity, hand tremor, muscle stiffness, and others — and the diagnosis is confirmed using neuroimaging techniques. The project of young scientists at Sechenov University aims to identify this pathology at an early stage — even before the first symptoms appear, when treatment can significantly slow down the development of the disease and preserve the quality of life of patients. According to the developers, artificial intelligence methods can help in this.

The essence of the project is to automate the early diagnosis of Parkinson's disease and make it more accessible to patients.

— Currently, EEG is not used to diagnose this disease. However, there are scientific publications by Russian and foreign researchers devoted to the peculiarities of the electrical activity of the brain on electroencephalography of patients with Parkinson's disease and their comparison with the EEG of healthy people. We set ourselves the task of finding out whether it is possible to differentiate this data using a neural network. As it turned out, this is quite realistic," Ekaterina Vakhromeeva, the author of the project, a graduate of the Master's degree in Information Systems and Technologies at Sechenov University, told Izvestia.

Photo: Getty Images/SCIENCE PHOTO LIBRARY

For this work, the scientists also used an open foreign dataset containing anonymized EEG data from patients with Parkinson's disease and healthy participants, manually marked up by a neurologist. The researchers divided the data into samples for training and testing the neural network. The model was trained on the first sample, and on the second sample, which she "saw" for the first time, her ability to detect the EEG of healthy and sick patients was tested. As a result, it turned out that the neural network can recognize frequency anomalies of electroencephalograms and identify patients with Parkinson's disease with an accuracy of 97%.

"The results obtained during this scientific work are very promising," said Denis Andrikov, the project's scientific director and an expert at the Center for Digital Medicine at Sechenov University. — Using a neural network for EEG analysis makes it possible to expand the approach to finding predictors of disease development and can help a doctor make diagnostic decisions.

Neural network for diagnostics

In the near future, the team plans to assemble a large dataset with EEG data from patients with various stages of Parkinson's disease. This will allow the model to be further trained and tested together with neurological experts, the scientists said.

Photo: Getty Images/Andrew Brookes

Artificial intelligence is already capable of diagnosing diseases at a fairly good level, Evgeny Sobolev, co-founder of the biotech studio Scanderm, told Izvestia.

— Such technologies have already proved themselves well. We have developed several AI skin analysis services using a smartphone camera, and there are also a sufficient number of solutions on the market for analyzing research results, such as X-ray images, and, naturally, the scope of AI applications will only expand, so the use of neural networks for EEG analysis is more than a logical step. We can expect that the technology will be able to detect Parkinson's disease at an earlier stage, which will allow us to start treatment on time," the expert said.

Life commitments: Russians may lose access to modern therapy

The reason is the change in the rules of public procurement of medicines.

The model is already achieving 97% accuracy, which indicates its high applicability in practice and potentially solving the problem of early diagnosis of this disease, said Anton Averyanov, CEO of the ST IT Group of companies, TechNet NTI market expert.

— Neural networks have excellent capabilities for analyzing and finding new patterns that were previously unknown to humans. This case confirms this and shows that they are very important for the development of medicine," he stressed.

Photo: Global Look Press/Monika Skolimowska

According to him, in general, the field of application of AI in medicine is increasing, and in the coming years artificial intelligence will open up many opportunities for treating various diseases, the expert concluded.