Spectral Extremal Connectivity of Two-State Seizure Brain Waves

Mara Sherlin D. Talento, Jordan Richards, Marco Pinto-Orellana, Raphael Huser, Hernando C. Ombao · 2025

Read Spectral Extremal Connectivity of Two-State Seizure Brain Waves by Mara Sherlin D. Talento, Jordan Richards, Marco Pinto-Orellana, Raphael Huser, Hernando C. Ombao (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-04169.txt.

Introduction

Mara Sherlin Talento and co-authors study brain wave links in a 2025 statistics paper on newborn seizures. Standard coherence tools measure average ties between brain signals, but they miss what happens in the extremes, where seizure energy bursts live. The team built a way to read connectivity in the tails of the signal distribution.

Their model handles brain phases that switch over time, contrasting burst-suppression periods with calmer ones. It can flag tail links at key frequency bands tied to energy outbursts that may precede seizures. Findings for epileptic and non-epileptic babies are compared side by side. The authors aim to give clinicians a sharper lens on the moments that matter most in fragile newborn brains.

Catalog record

CREATORMara Sherlin D. Talento, Jordan Richards, Marco Pinto-Orellana, Raphael Huser, Hernando C. Ombao
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE68 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-04169

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Spectral Extremal Connectivity of Two-State Seizure Brain Waves Mara Sherlin D. Talento, Jordan Richards, Marco Pinto-Orellana, Raphael Huser, Hernando C. Ombao Abstract Coherence analysis plays a vital role in the study of functional brain connectivity. However, coherence captures only linear spectral associations, and thus can produce misleading findings when ignoring variations of connectivity in the tails of the distribution. This limitation becomes important when investigating extreme neural events that are characterized by large signal amplitudes. The focus of this paper is to examine connectivity in the tails of the distribution, as this reveals salient information that may be overlooked by standard methods. We develop a novel notion of spectral tail association of periodograms to study connectivity in the network of electroencephalogram (EEG) signals of seizure-prone neonates. We further develop a novel non-stationary extremal dependence model for multivariate time series that captures differences in extremal dependence during different brain phases, namely burst-suppression and non-burst-suppression. One advantage of our proposed approach is its ability to identify tail connectivity at key frequency bands that could be associated with outbursts of energy which may lead to seizures. We discuss these novel scientific findings alongside a comparison of the extremal behavior of brain signals for epileptic and non-epileptic patients. Keywords: coherence analysis; conditional extremes; electroencephalogram data; extreme value theory; spectral analysis; tail dependence 1 Introduction Seizures in neonates are known to disrupt brain development. Compared to adults, study of seizures in infants is more challenging because symptoms in neonates are often more subtle ( Stevenson et al., 2019 ) . To study and assess brain function, various brain imaging modalities, such as electroencephalograms (EEGs), are used to diagnose mental and neurological disorders, e.g., seizures and epilepsy ( Muthuswamy et al., 1999 ; Stevenson et al., 2019 ) . EEGs indirectly measure cortical electrical activity through non-invasive sensors placed on the scalp. In this paper, we analyze EEG recordings for a subset of 79 neonates in a neonatal intensive care unit (NICU) of the BAby Brain Activity Center (BABA Center), Finland ( Stevenson et al., 2019 ) . These neonates were initially recorded as having had a seizure; however, further analysis by experts confirmed seizures in only 39 patients. Figure 1 shows a snippet of EEGs from the temporal lobe channels of one of the neonates. Observe that the data exhibit non-stationarity due to sudden outbursts. This specific pattern, often seen in neonates, is called “burst-suppression" ( Douglass et al., 2002 ) . Although the presence of “burst suppression” patterns (see Figure 1 ) is not always associated with abnormalities in the brain ( Douglass et al., 2002 ) , one characteristic that can delineate hostile patterns is connectivity (dependence between channels) in a brain network. For instance, a recent study of a surgically-removed portion of the temporal region of an epileptic patient revealed a very rare connectivity of axons at the seizure site ( Shapson-Coe et al., 2024 ) . Therefore, developing a statistical model to characterize brain connectivity during burst-suppression and non-burst-suppression patterns (hereafter referred to as the “burst phase" and “non-burst phase", respectively) can provide insights into brain functionality. Thus, in this paper, we develop a novel non-stationary multivariate model for EEG signals, that accounts for differences between states (e.g., brain phases, such as burst-suppression vs. non-burst-suppression). Figure 1: (Left) Channels T3-T6 in the temporal lobe, in green. (Right) Multivariate EEG Signals from channels in the temporal lobe of a neonate with normal ultrasound admitted to the NICU of the BAby Brain Activity Center, Finland. The red lines are signals during a time interval that was flagged as a burst-suppression pattern by our algorithm, described in the Supplementary Material, Section A-1 . EEG data. EEGs are often considered as a “sum" of oscillating waveforms with random amplitudes. Thus, dependence between EEG signals can be characterized as synchronization between these oscillations. In Srinivasan, 2007 , standard analyses decompose EEGs into oscillations at the following five distinct frequency bands, denoted by \Omega_{\ell} for {\ell=1,\dots,5} : (1) Delta, \Omega_{1}:=(0,4] Hertz or Hz hereafter, which is dominant during sleep; (2) Theta, \Omega_{2}:=(4,8] Hz, which is associated with drowsiness; (3) Alpha, \Omega_{3}:=(8,12]\text{ Hz} , which is associated with resting-state; (4) Beta, \Omega_{4}:=(12,30]\text{ Hz} and (5) Gamma, {\Omega_{5}:=(30,50]\text{ Hz}} , which are dominant when performing mental and cognitive tasks ( Kumar and Bhuvaneswari, 2012 ) . To define and estimate between-channel connectivity, the first step is to decompose the original signal into these bands. A general framework for modeling dependence between oscillations is developed in Ombao and Pinto, 2022 which we summarize here. Denote the neonate multi-channel EEGs to be {\{\boldsymbol{X}(t)\}_{t=1}^{T}:=(X_{1}(t),\ldots,X_{p}(t))^{\top}} , where p is the number of channels and T is the observation period. While Figure 1 suggests that the EEGs are globally non-stationary, they can be modeled as weakly stationary within local (short) time blocks ( Adak, 1998 ; Dahlhaus, 2000 , e.g., piecewise stationary or locally stationary;) . Suppose that the EEGs are segmented into B blocks (each of length L ) where the EEGs are approximately weakly stationary within each block. Denote the set of time points in block b to be \mathcal{T}_{b}:=\{(b-1)L+1,\dots,bL\} . Here, denote T=LB to be the total number of time points across all blocks. Each signal, X_{j}(t) , j=1,\dots,p , can be separately decomposed within each time block, \mathcal{T}_{b} , b=1,\dots,B , into the standard oscillations via bandpass linear filtering (e.g., using the Butterworth filter), so that X_{j}(t):=\sum_{\ell=1}^{5}X_{j,\Omega_{\ell}}(t),t\in\mathcal{T}_{b} , where X_{j,\Omega_{\ell}}(t) is the filtered signal for channel j corresponding to \Omega_{\ell} ( Ombao and Pinto, 2022 , see) . We develop a statistical modeling framework and a related cross-channel dependence measure which is motivated by the National Institute of Health ( NIH, 2023 ) definition of seizure-related brain behavior; this defines the onset of a seizure as a concurrent outburst of neuronal electrical discharges across multiple neurons. Instead of modeling the filtered signals (or filtered time series), we study the spectral power (or squared amplitudes), which is a more direct representation of neuronal energy (or signal variance). The data analogue of the spectrum (or spectral power) are the periodograms of each EEG channel ( Brillinger, 2001 ) . As will be formalized in Section 3.1 , where periodograms are directly calculated from the discrete Fourier transform (DFT), the periodogram of \{X_{j}(t)\}_{t\in\mathcal{T}_{b}} at the \Omega_{\ell} -frequency band, denoted as I^{(\ell)}_{j}(b) , may be computed for the b -th block as I^{(\ell)}_{j}(b)=\frac{1}{L}\sum_{t\in\mathcal{T}_{b}}\left[X_{j,\Omega_{\ell}}(t)\right]^{2},\;\;\;j=1,\dots,p,\;\;b=1,\dots,B. (1) Above, it is assumed that X_{j,\Omega_{\ell}}(t) is oscillating around its mean, 0. In this paper, we study dependence between the oscillation amplitudes of signals \{\boldsymbol{X}(t)\}_{t\in\mathcal{T}_{b}} , for all {b=1,\dots,B} , through the periodograms \boldsymbol{I}^{(\ell)}(b)=(I^{(\ell)}_{1}(b),\dots,I^{(\ell)}_{p}(b))^{\top} , \ell=1,\dots,5 , derived from Equation ( 1 ). Guhr and Schell, 2021 propose a multivariate model for non-stationary time series by segmenting the long multivariate time series into distinct blocks and defining a measure for each block (similar to the approach described above). While their study did not introduce a tail dependence measure, it highlighted that non-stationarity, arising from fluctuating correlation matrices across blocks, can lead to a heavy-tailed distribution for these defined measures. Modeling extremal dependence in spectral power. To demonstrate the importance of studying tail dependence, we explore correlation between components of the periodogram (or squared-amplitudes). While our analysis focuses on multi-channel data,

Read the complete text