System for automatic analysis of sleep patterns developed by UNN scientists
A system for automatic analysis of sleep patterns in patients with epilepsy has been developed at Lobachevsky University. The software package will allow doctors to extract maximum information from electroencephalography (EEG) recordings of patients with suspected epilepsy, predict the course of the disease, and find personalised treatment approaches.
UNN researchers have trained artificial intelligence models to automatically detect and analyse sleep spindles in EEG recordings from healthy individuals and patients with epilepsy. High-precision EEG analysis is crucial for uncovering hard-to-detect epilepsy symptoms in children.
"According to medical protocols, threshold values of brain rhythms serve as signals for starting treatment. However, these data are often evaluated in a fragmented and subjective manner. It is challenging for an individual specialist to analyse many hours of EEG recordings in full. To assist epileptologists and neurologists in conducting a more accurate and rapid analysis, this work needs to be automated," explained Tatiana Levanova, author of the study and senior researcher at the UNN Laboratory for Artificial Intelligence in Cardio- and Neuroscience.
One of the key stages of this project was dedicated to the analysis of neural signals during the second phase of sleep. Brain rhythms recorded during this state in healthy individuals are known as "sleep spindles". These spindles serve as biomarkers for important cognitive processes such as the formation of long-term memories and the assimilation of new information. In epilepsy, the characteristics of sleep spindles, including their density, frequency, and duration, may be altered. Automatic analysis of these characteristics can serve as an important biomarker for monitoring the progression of the disease in individuals with an established diagnosis. Nizhny Novgorod researchers have developed a method that allows for the description and analysis of these changes.
"The EEG data used to train our artificial intelligence models, as well as the results obtained from the prototype of our future software package, have been verified by our colleagues, doctors from Moscow and Nizhny Novgorod. We continue to collect data on various types of epileptic activity in order to further train our AI and improve the accuracy of its recommendations," said Albina Lebedeva, author of the study and senior researcher at the UNN Laboratory for Artificial Intelligence in Cardio- and Neuroscience.
"The global goal of our research is to minimise subjectivity in evaluating electroencephalography data, simplify the labelling of epileptiform activity and key physiological patterns, and achieve a unified calculation of parameters such as frequency, amplitude, and occurrence. This will make it possible to make informed clinical decisions regarding both diagnosis and optimal treatment," commented neurologist Artyom Sharkov, researcher at the Laboratory for Artificial Intelligence and Big Data Processing at UNN.
According to scientists, neural network approaches for assessing sleep spindles and background brain activity can serve as an additional tool for automatically diagnosing suspected epilepsy. Software development for testing the technology in partner clinical centres is currently underway.
The results of the study were published in the international journal Technologies 2025. The research was carried out by a team of neuroscientists, mathematicians, and programmers from UNN as part of the Artificial Intelligence Centre project, with the support of the UNN IITMM Laboratory for Artificial Intelligence and Big Data Processing.
In 2025, the researchers who created Russia's first open medical database of labelled EEG recordings were awarded the Gravity International University Prize. The database allows for accurate, quick, and high-quality diagnosis of epilepsy in both adults and children using artificial intelligence technology.
