Toolboxes Bouquet

In the field of cognitive and clinical neuroscience, handling and analyzing EEG and MEG data from various acquisition systems is often a technical challenge due to heterogeneous formats and processing requirements. To address this, we developed a software tool designed to provide intuitive visualization, flexible pre and post processing, and interoperability with current neuroimaging standards. The software is fully compatible with the Brain Imaging Data Structure (BIDS) for EEG/MEG, enabling structured data organization and seamless integration with standard analysis pipelines. By integrating with FreeSurfer and GARDEL(shown in the 2nd part) the software allows visualization of brain activity on subject-specific cortical surfaces. We will present various processing operations:
• Bandpass and notch filters
• Independant component analysis
• Power Spectral Density (PSD) estimation
• Time-frequency analysis (wavelets, STFT)
• BIDS interaction with SEEG activity mapping (raw signal or ICA topographies) and Interactive visualization

In a second part we will present GARDEL, a companion tool for co-registration of CT and MRI scans and semi-automatic detection of intracranial electrodes. GARDEL facilitates accurate anatomical localization of implanted electrodes by segmenting and mapping them onto brain structures.

We will also present the ability to create and execute custom analysis modules written in MATLAB or Python directly from within the software. This offers flexibility for users who rely on existing scripts or research pipelines.
A quick demonstration/tutorial will expose how to create a plugin in MATLAB/Python.

We will finish by making a demo of available plugins for signal processing such as Delphos module, which detects spikes and fast oscillations in EEG signals—an essential tool for clinical and research applications in epilepsy.

Prerequisite: None

Braindecode is an open-source toolbox for training and deploying deep learning decoding models on M/EEG signals. It contains a large collection of state-of-the-art deep learning models.

In this session, we will see how to harness deep learning methods for M/EEG decoding.

Prerequisite: Python

This three session workshop will demonstrate how to use my free open-source Matlab EEG analysis suite (Dien, 2010) to analyze ERP data, with an emphasis on its strengths for performing cutting edge artifact correction (Dien, 2024), robust ANOVA (Dien, 2017), and two-step PCA (Dien, 2012). Each session will consist of a brief presentation of the core concepts, followed by a demonstration of how to perform them using the EP Toolkit, and ending in a short hands-on period allowing for questions and answers.

Prerequisite: None

Documenting Events in Time Series Data using HED Tools
Cutting M/EEG Tutorials,
October, 2025

Cognitive neuroscience and functional neuroimaging seek to relate brain dynamics to experience and to behavior. Cognitive neuroimaging seeks to model the relationship of recorded time series data to concurrent sensory experience and behavior of the imaged participants. Documenting details of this experience and behavior is thus essential. In the current era of growing public data archives, documenting sensory, behavioral, and other events in neuroimaging data using common terms and syntax allows efficient data search, retrieval, and joint analysis (including AI-empowered mega-analysis). Unfortunately, a common system for event annotation has not been adequately addressed in emerging data storage standards (e.g., BIDS or NWB).
A dozen years ago, Nima Bigdely-Shamlo at UCSD proposed development of a standard for annotating events occurring during time series recordings, naming it the system of Hierarchical Event Descriptors (HED). Following a decade of development, the HED standard and its growing array of associated user tools was accepted (in 2024) by the INCF as the (now only) international standard for event annotation of time series data.
The HED tutorial will consist of compact lectures on HED purpose and structure, the process of HED annotation, and using HED annotations in M/EEG data analysis. Example analyses will use EEGLAB and Fieldtrip. These will alternate with HED tool demonstrations and periods for attendees to test applying the demonstrated tools to readily downloaded data. Tutors will be available to answer attendee questions (making use of whatever videochat options are available). HED tools now include a HED annotation assistant using AI resources.
We also hope to be able to report on a proposed NeurIPS competition using a very large (~3k subject) EEG dataset (Healthy Brain Network data, available on NEMAR.org). We hope that this competition will provide stimulating examples of using HED annotations to mine M/EEG data.

Prerequisite: Some understanding of current data archiving systems (BIDS or NWB).

Websites
https://www.HEDtags.org
https://www.youtube.com/@HierarchicalEventDescriptors

References
Makeig, S. and Robbins, K., 2024. Events in context—The HED framework for the study of brain, experience and behavior. Frontiers in Neuroinformatics, 18, p.1292667.
Robbins, K., Truong, D., Jones, A., Callanan, I. and Makeig, S., 2022. Building FAIR functionality: annotating events in time series data using hierarchical event descriptors (HED). Neuroinformatics, 20(2), pp.463-481.
Robbins, K., Truong, D., Appelhoff, S., Delorme, A. and Makeig, S., 2021. Capturing the nature of events and event context using hierarchical event descriptors (HED). NeuroImage, 245, p.118766.

The Human Neocortical Neurosolver (HNN) is a user-friendly neural modeling software designed to provide a cell- and microcircuit-level interpretation of macroscale magneto- and electroencephalography (M/EEG) signals (https://hnn.brown.edu, Neymotin et al. 2020). The foundation of HNN is a biophysically-detailed neocortical model, representing a patch of neocortex receiving thalamic and corticocortical drive. The HNN model was designed to simulate the time course of primary current dipoles and enables direct comparison, in nAm units, to source-localized M/EEG data, along with layer-specific cellular activity. HNN workflows are constructed around simulating commonly measured Event Related Potentials (ERPs) and low-frequency oscillations. The HNN model can be accessed through a user-friendly interactive graphical user interface (GUI) or through a Python scripting interface.

The foundation of HNN, referred to as HNN-core (Jas et al. 2023), is a Python package containing all of the core functionality of HNN, and is implemented with a clear application programming interface (API). A new GUI has recently been implemented. Tutorials on how to simulate ERPs and low-frequency oscillations in the alpha, beta, and gamma bands are distributed for both the interactive GUI and the Python API. HNN was created with best practices in open-source software to allow the computational and human neuroscience communities to understand and contribute to its development. The HNN API contains additional functionality beyond that accessible through the GUI, including the ability to modify local network connectivity, perform parameter optimization, and simulate layer-specific local field potential signals and current source density. The package is available to install with a single command on PyPI (“pip install hnn_core”), is unit tested, and extensively documented. HNN is additionally accessible through computing resources offered by the Neuroscience Gateway (NSG), enabling large simulation workloads. Overall, HNN is a one-of-a-kind, openly-distributed tool designed for a broad community to develop and test hypotheses on the multiscale origins of localized human M/EEG signals.

In this session, we will begin with a didactic overview of the background and development of HNN. We will then introduce users to the GUI and Python API through lectures and demo investigations of ERPs.

Prerequisite: basic neuroscience background

Precise anatomical localization of intracranial electrodes is crucial for interpreting invasive recordings in clinical and cognitive neuroscience research. The open-source iElectrodes toolbox offers a fast, semi-automated, and robust solution for localizing subdural grids, depth electrodes, and strips from MRI and CT images, supporting automatic anatomical labeling. iElectrodes was initially introduced in Blenkmann et al. (2017), and has been updated with major methodological innovations in Blenkmann et al. (2024). To date, it has >2000 downloads.
In this 90-minute session, I will first provide an introductory lecture on the core functionalities of iElectrodes, including image pre-processing steps, semi-automatic electrode localization, brain shift compensation, and standardized anatomical registration. We will cover the recent major upgrades to the toolbox: the GridFit algorithm for robust localization of SEEG and ECoG electrodes under challenging conditions (e.g., noise, overlaps, and high-density implants), and CEPA (Combined Electrode Projection Algorithm) for smooth compensation methods for grids, addressing brain deformations based on mechanical modeling principles. These developments significantly enhanced the robustness and precision of intracranial electrode localization.

In the second part of the session, we will move into a hands-on tutorial, where participants will learn how to use the toolbox through practical exercises. Using real patient datasets (anonymized), we will cover:

• Preprocessing MRI and CT images.
• Semi-automatic detection and localization of electrode coordinates using clustering and GridFit algorithms.
• Brain shift correction using CEPA.
• Automatic anatomical labeling of electrodes.
• Generation of an iElectrodes localization project file.
• Exporting electrode coordinates into formats compatible with Fieldtrip, EEGLAB, and text reports.
• Integration with further analysis workflows.

This session is intended for both clinical and cognitive neuroscience research users working with SEEG or ECoG. Attendees will leave with practical skills for reliable and reproducible electrode localization, ready to apply to their own datasets.
Required Materials:

• Participants should install MATLAB (requires a license) and download the open-source iElectrodes toolbox (available at https://sourceforge.net/projects/ielectrodes/) ahead of the session.
• Example datasets of pre-processed images will be provided before the event.
References:
• Blenkmann AO, et al. (2017). iElectrodes: A Comprehensive Open-Source Toolbox for Depth and Subdural Grid Electrode Localization. Frontiers in Neuroinformatics, 11:14. doi:10.3389/fninf.2017.00014
• Blenkmann AO, et al. (2024). Anatomical registration of intracranial electrodes. Robust model-based localization and deformable smooth brain-shift compensation methods. Journal of Neuroscience Methods, 404:110056. doi:10.1016/j.jneumeth.2024.110056

Prerequisite: Matlab license required

Have you ever wondered what will happen to the results if you change a parameter of your analysis method? Did you want to test whether your results could be explained by a trivial effect? Or did you need to generate a toy example to illustrate your idea in a presentation? For all these questions, simulated M/EEG data can be of great help! In fact, it brings more fun if the simulations can be assembled easily and in a flexible way. This is exactly the aim of the MEEGsim toolbox, which provides building blocks for simulations, mostly focusing on connectivity (for now): template waveforms of source activity, simulation of phase-phase coupling, and adjustment of the signal-to-noise ratio. Come to the session to learn more about the toolbox and try it out in your (maybe even first) simulation! In the meantime, feel free to read more about the toolbox in the documentation: https://meegsim.readthedocs.io/en/stable/.

Prerequisite: Python >= 3.9 as well as MNE-Python and MEEGsim packages

This workshop introduces participants to advanced M/EEG analysis techniques focused on quantifying neural signal complexity using modified Multiscale Entropy (mMSE). Drawing on the FieldTrip toolbox and an extended entropy analysis pipeline, I will explore how mMSE can capture meaningful temporal irregularities in brain activity that are often missed by traditional analysis approaches like spectral power or event-related potentials.
The session is structured into two main components: a conceptual lecture and a hands-on demonstration (~45 minutes each). In the lecture portion, I will begin by reviewing the theoretical basis of entropy as a measure of brain signal variability, emphasizing its relevance for characterizing brain states that dynamically change over time. I will then introduce the mMSE algorithm, a method designed to estimate moment-to-moment brain signal entropy over multiple timescales using discontinuous EEG data. Compared to conventional multiscale entropy (MSE), mMSE allows for computation of entropy over time by concatenating short segments of interest, using a similar sliding window approach as is done in traditional time-frequency (spectral) analysis. This makes mMSE especially useful for investigating short-lived cognitive processes as they occur in event-related designs.
I will highlight recent applications of mMSE in cognitive and systems neuroscience to demonstrate its utility in linking entropy profiles to individual differences and task-related neural variability. The lecture will also cover key practical considerations, including preprocessing steps such as high-pass filtering, removal of the event-related potential, and segment selection, all of which are critical for reliable entropy estimation.
In the practical demonstration, participants will follow a step-by-step walkthrough of an mMSE analysis using a FieldTrip-based pipeline in MATLAB. I will begin with loading and preprocessing sample EEG data, proceed to entropy computation using the ft_entropyanalysis function, and finish with result visualization and interpretation. Special attention will be given to configuring the mMSE-specific parameters (e.g., time scales, pattern length and pattern similarity), and understanding how different parameter choices affect entropy estimates. Participants will gain hands-on experience that enables them to apply entropy-based analyses to their own FieldTrip-preprocessed datasets.
This session is aimed at M/EEG researchers and students with basic familiarity in EEG data processing and MATLAB. By the end of the workshop, attendees will have a solid conceptual and practical foundation for using mMSE as a tool to investigate brain signal variability. All code, example data, and links to the mMSE toolbox will be provided. See https://shorturl.at/NWhso for the online tutorial, and https://github.com/LNDG/mMSE for the MATLAB function.

Prerequisite: N/A

In this lesson, we will explore PhysioEx, an open-source Python library designed to facilitate explainable deep learning for automated sleep staging. The session will guide participants through the core design principles of PhysioEx and demonstrate how it supports the complete deep learning sleep staging pipeline, from data loading and preprocessing to model training, evaluation, and explainability.
We will begin by discussing the motivation behind PhysioEx: the growing need for standardized, modular, and accessible tools to develop and evaluate sleep staging models that are both accurate and interpretable. Emphasis will be placed on how PhysioEx integrates Explainable AI (XAI) methods directly into the pipeline, bridging the gap between raw physiological data (EEG, EOG, EMG) and clinically meaningful decisions.
The lesson will then walk through the structure of the library, covering its extensible API and command-line interface to train, test and finetune deep learning models on a large variety of datasets. We will detail how PhysioEx manages (big-)data loading and preprocessing with a focus on how the library allow to dynamically merge multiple.
We will explore the training and testing workflow, highlighting how PhysioEx supports a variety of state-of-the-art neural architectures for sleep staging with a focus on models resembling the sequence-to-sequence framework, such as SeqSleepNet, TinySleepNet and SleepTransformer. Particular attention will be given to cross-dataset training and generalization experiments, showing how PhysioEx enables fair and reproducible evaluation of model robustness across domains.
In the final part of the lesson, we will focus on explainability, introducing the set of post-hoc XAI algorithms implemented in the library and suited for time-series classification. These include techniques for saliency mapping, relevance propagation, and concept-based explanations that help interpret model predictions in alignment with AASM-defined sleep staging rules. We will show how these tools can provide meaningful insights into model behavior, promote transparency, and support clinical adoption.
By the end of the lesson, participants will have a comprehensive understanding of PhysioEx’s capabilities and its role in promoting reproducible, interpretable, and clinically aligned research in sleep medicine.

Prerequisite: Good familarity with Python and Pytorch

Presentation spike sorting using the spikeinterface pipeline.

Prerequisite:

When testing analysis pipelines, comparing different analysis approaches, or validating statistical methods, one often needs EEG data with a known ground truth. Simulating EEG data based on known parameters addresses this need.

Here, we present UnfoldSim.jl, a free and open-source Julia package designed for simulating continuous EEG data based on (potentially overlapping) event-related potentials. Using regression formulas, users can flexibly determine the relationship between the experimental design and the response functions. UnfoldSim.jl also provides support for multi-channel simulations via EEG-forward models and allows for the simulation of both single-subject and multi-subject data. One of its core design principles is modularity, enabling users to tailor the simulation to their specific research applications.

In this session, we will introduce UnfoldSim.jl and provide a brief overview of its key features. The core part of the session will guide you through a simple simulation example to introduce the different simulation ingredients and illustrate the simulation workflow. Afterwards, there will be a hands-on part in which you can explore the effect of different parameters on the simulated data and create your own simulations.

This workshop uses the Julia programming language and the practical part will be conducted using Pluto.jl notebooks. While programming experience in Julia is not required, experience in MATLAB, R, or Python is recommended.

If you want to learn more about UnfoldSim.jl already, have a look at its documentation (https://unfoldtoolbox.github.io/UnfoldDocs/UnfoldSim.jl/stable/) or our JOSS paper (https://joss.theoj.org/papers/10.21105/joss.06641).

Prerequisite: For the hands-on session, please install Julia (at least version 1.11) and Pluto.jl on your computer. We recommend you to follow this installation guide (from our Unfold workshop earlier this year): https://www.s-ccs.de/workshop_unfold_2025/installation.html

Linear mixed models are versatile and increasingly popular in cognitive psychology to analyze behavioral datasets with within-subject trial-repetitions. Some brave researchers have already applied these hierarchical models to EEG data, typically on the averaged space/time region of interest.

Prerequisite: Basics of multiple regression