PhD Scientific Days 2026

Neurosciences

From Noise to Physiology: Assessing Sleep Neural Time Series Using Multi-Scale Entropy and Power-Law Modulated Multi-Scale Entropy

Name of the presenter

Magyar, Tárek Zoltán

Institute/workplace of the presenter

Institute of Behavioural Sciences, Semmelweis University

Authors

Tárek Zoltán Magyar^1,2, Dr. Martin Dresler³, Dr. Orsolya Szalárdy¹, Dr. Yevgenia Rosenblum³, Prof. Dr. Róbert Bódizs¹
1: Institute of Behavioural Sciences, Semmelweis University, Hungary
2: Selye János Doctoral College for Advanced Studies, Semmelweis University, Hungary
3: Donders Institute for Brain, Cognition, and Behaviour, Radboud University, Netherlands

Text of the abstract

Besides canonical sleep electroencephalography (EEG) features, advances in signal processing have enabled the extraction of aperiodic (non-oscillatory) and complexity-based metrics that capture complementary properties of neural activity. Among these, multi-scale entropy (MSE) quantifies signal irregularity across temporal scales, whereas power-law modulated multi-scale entropy (pMSE) extends this approach by incorporating the aperiodic, power-law characteristic of EEG signals. As a result, pMSE emphasizes the balance between randomness and order that characterizes healthy biological systems. In this study, we compare MSE vs pMSE in differentiating between sleep signals and the consistency of the results over datasets.
MSE and pMSE were computed from EEG recordings recorded by the human Healthy Brain Study (n = 863), and a mouse sleep deprivation experiment (n = 8), as well as simulated signals with controlled spectral exponents ranging from 1.5 to 3.5. We evaluated the ability of MSE and pMSE to differentiate signals with varying spectral exponents over a range of time scales up to 500 ms. Likewise, we compared their ability to capture sleep homeostatic adjustments relative to baseline with ANOVA.
In simulations, both MSE and pMSE curves were sensitive to changes in spectral exponents, indicating that entropy estimates depend on the power-law structure of the signal. However, only pMSE showed a clear correspondence between physiological and simulated signals with matching spectral exponents across time scales up to 400 ms. In addition, pMSE, within the 300-350 ms time scale, captured homeostatic adjustments induced by sleep deprivation (F(1, 7) = 7.034; p = .024; ηp = .501), whereas MSE did not.
By incorporating power-law spectral dynamics into multiscale entropy estimation, pMSE more accurately maps physiological sleep EEG signals and better captures homeostatic changes observed in a rodent sleep deprivation paradigm compared to conventional MSE.
Funding: Research supported by the Dutch Research Council (NWO), Alzheimer Nederland Early Career Grant, the Hungarian Ministry of Innovation and Technology (TKP2021-EGA-25 and TKP2021-NKTA-47), and the National Research, Development and Innovation Office of Hungary (K146792,2023-1.2.1-ERA_NET-2023-00004), as well as the EU Joint Programme Neurodegenerative Disease (JPND2022-120)