MapperEEG: A Topological Approach to Brain State Clustering in EEG Recordings

Brittany Story, Zhibin Zhou, Ramesh Srinivasan, Piotr J. Franaszczuk · 2025

Read MapperEEG: A Topological Approach to Brain State Clustering in EEG Recordings by Brittany Story, Zhibin Zhou, Ramesh Srinivasan, Piotr J. Franaszczuk (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-2504-10252.txt.

Introduction

Brittany Story and co-authors introduce MapperEEG, a way to sort brain states in scalp recordings using topology. EEG data are rich but messy, noisy and shifting. Grouping their patterns without labels is hard. Mapper, a tool from topological data analysis, maps high-dimensional data into a connected sketch of clusters.

Blended with classic spectral features, the sketch both clusters states and shows how they link. In a finger-tapping task it beat six rival methods, from hierarchical clustering to autoencoders. In a harder go/no-go shooting task, rival methods failed outright. MapperEEG still revealed structure in the states. The authors offer it as a fresh lens for brain dynamics.

Catalog record

CREATORBrittany Story, Zhibin Zhou, Ramesh Srinivasan, Piotr J. Franaszczuk
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-10252

Browse the full library · About the Tower of Babel

Full text

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

MapperEEG: A Topological Approach to Brain State Clustering in EEG Recordings Brittany Story, Zhibin Zhou, Ramesh Srinivasan, Piotr J. Franaszczuk Abstract Electrical potential scalp recordings (Electroencephalograms-EEGs) are a common tool used to investigate brain activity. EEG is routinely used in clinical applications as well as in research studies thanks to its noninvasive nature, relatively inexpensive equipment, and high temporal resolution. But, EEG is prone to contamination from movement artifacts and signals from external sources. Thus, it requires advanced signal processing and mathematical analysis methods in tasks requiring brain state identification. Recently, tools from topological data analysis have been used successfully across many domains, including brain research, however these uses have been limited to fMRI datasets. We introduce the topological tool MapperEEG (M-EEG) and provide an example of it’s ability to separate different brain states during a simple finger tapping teaming task without any pre-labeling or prior knowledge. M-EEG uses the power spectral density applied to traditional EEG frequency bands combined with the Mapper algorithm from topological data analysis to capture the underlying structure of the data and represent that structure as a graph in two-dimensional space. This tool provides clear separation (clustering) of states during different conditions of the experiment (syncopated vs. synchronized) and we demonstrate that M-EEG outperforms other clustering methods when applied to EEG data. 1 Introduction Characterizing normal and pathological brain states has long been a key component of neuroscience. One way to capture and analyze brain states is through an electroencephalogram (EEG). EEG data is easy to collect, inexpensive, non-invasive, and it provides high time resolution, which makes it a useful tool for researchers across a wide range of disciplines. It has some drawbacks such as low spatial resolution, high noise levels, and a high number of artifacts. Various analytical methods are used for interpretation of EEG data both for clinical and research applications. [ 1 ] . From the earliest EEG studies, clinicians and researchers noticed the presence of oscillatory features in EEG recordings. Over time, the oscillatory components of EEG within different frequency bands were correlated with certain brain functions and pathologies [ 2 ] . Other brain data collection methods, such as fMRI, were introduced later with higher spacial resolution and less noise, however, with lower time resolution and significantly higher costs. Recently, topological approaches have been used by computational neuroscientists to capture and cluster brain states. In [ 3 ] , researchers studied the organization of brain states using an algorithm called Mapper, introduced by Singh et al. in [ 4 ] . Mapper, when applied to fMRI data, was able to organize different brain states categorically without any labeling. But, the types of tasks and states are limited to those that can be collected inside the laboratory setting with all additional restrictions imposed by fMRI. Thus, we look to see how we can use Mapper paired with tools from traditional EEG approaches to cluster EEG data to uncover the underlying structure of brain states. In this work, we introduce MapperEEG (M-EEG), an approach for analyzing EEG data that uses the power spectral density function paired with the Mapper algorithm. We demonstrate how combining these approaches can be used to understand brain state structure by showing it’s efficacy at clustering brain states during a human teaming task. The paper is organized as follows. Section 3 outlines the M-EEG algorithm and why this approach is appropriate for EEG data. We also outline the dataset we used to test our algorithm. Next, Section 4 shows how M-EEG can be used to understand the dataset’s structure and we verify M-EEG’s efficacy and utility. Finally, we discuss how M-EEG can be used with other datasets and consider possible extensions. 2 Background 2.1 EEG Electroencephalography (EEG) has been a popular neuroimaging technique for understanding what goes on in the brain since it’s first human application in 1924. [ 5 ] Non-invasive, inexpensive, and EEG’s high-temporal resolution are a few of advantages it has to offer [ 6 ] . Modern systems can use anywhere from 4 to 256 electrodes, depending on the application and years of research has pinpointed the optimal electrode placement for a wide range of tasks, such as neural speech tracking, epilepsy, and concentration [ 7 , 8 , 9 ] . But there are some disadvantages, just as with any data collection approach. Lack of spatial resolution, individual brain and skull differences, depth restrictions and noise are just a few of the difficulties researchers face [ 10 ] . Over the years, researchers have worked around these obstacles and created tools that clean and parse EEG data. Source localization, band separation, filtering, downsampling, and other signal processing techniques have allowed researchers to find meaningful patterns and information in the initially messy data [ 11 ] . One such task is the clustering of brain states. Understanding how brain states differ across tasks has been explored across a variety of domains such as during transitions, resting, seizures, and mental tasks [ 12 , 13 , 14 , 15 ] . Often, these approaches rely on other types of neuroimaging, such as fMRI. Thus, there is a need for a methodology that can classify different brain states across different task types with EEG data. Traditional methods such as event-related potentials and oscillatory dynamics have long been used to analyze and interpret EEG data [ 2 ] . More modern approaches such as Bayesian networks, support vector machines (SVMs), and random forests, characterize signals and associate them to behavior. For example, Bird et al. have used these methods to classify brain states as neutral, relaxed, or concentrating with an accuracy of roughly 87% [ 16 ] . EEG states have also been used as a method of communication for people with communication issues [ 17 ] . Additionally, EEG has been used to classify teammate trust with a decision tree classifier with an accuracy level of 72% [ 18 ] . Our goal is to develop an EEG clustering/mapping algorithm demonstrates solid accuracy while also capturing the relationship between states. 2.2 Mapper The Mapper algorithm was first developed by Singh, Memoli, and Carlsson in 2007 as a way to visualize and understand the structure of high dimensional data [ 4 ] . Since it’s inception, it has been used in several different domains, such as fraud detection, gerrymandering, and neuroscience [ 19 , 20 , 21 ] . With respect to neuroscience, Mapper has been used to detect and identify transition states, pinpoint causal states, and predict states [ 22 , 23 , 24 ] . Other topological tools, such as persistent homology, have been used to look at simulated EEG data [ 25 ] , cancer research [ 26 ] , and brain injuries [ 27 ] . These applications all leverage topology’s ability to analyze data in a way that is structurally relevant. In this work, we will focus on using Mapper to examine the underlying structure of multiple EEG channels and how each state connects with the rest of the states. 3 Method: Applying Mapper to EEG Data As a way to capture and cluster different brain states from EEG data, we look to the processes outlined in work such as those by

Read the complete text