Identifying Independent Components of Mobile Brain Imaging

Introduction

This page outlines and introduces the objective of our research study Independent Components of Mobile Brain Imaging (ICMoBI). This study requires a large amount of data to yield the most successful outcome; therefore, we, the research group of Human Neuromechanics Laboratory (HNL) at the University of Florida and Berlin Mobile Brain/Body Imaging Laboratory (BeMoBIL) at the Technical University of Berlin, are kindly requesting to collect and use various datasets from different research groups.

Objective

The primary aim of this study is to automatically identify neural and artificial components in mobile electroencephalography (EEG) data. EEG acquired during active human motion are often contaminated with a variety of unwanted signals such as muscle, eye, and cardiac signals. Current classifiers for labeling ICs were trained with stationary EEG data, which does not fully represent the data content of mobile EEG studies (e.g., ICLabel, MARA, IC_MARC). Training a component classifier with mobile EEG data may lead to better model performance than existing classifiers. Therefore, we will be customizing a new machine learning model using data contributed from a large variety of MoBI studies. The goal is to reduce subjectivity when selecting ICs to report by automatically classifying brain, muscle, and other types of ICs. We intend to eventually integrate this classifier into EEGlab so it can easily be applied to any MoBI dataset. Additionally, we are exploring the possibility of implementing deep learning to pioneer a novel decomposition method in separating neural and non-neural signals. Data may be used for this secondary aim as well.

Use of Data

We will request data with a private DropBox link. The received data will be promptly moved and stored on a network drive hosted by the University of Florida to ensure confidentiality from other research groups. We will be using this data to train and test machine learning algorithms to separate neural signals from artifacts. Using the received datasets, we will be applying feature extraction methods for EEG signal analysis and feature selection. The model performance will be evaluated using the outcomes from previous steps by comparing new classifier results with the original IC labels. This comparison will be made by having the ICs available on our customized website for experts to label. Additionally, the IC features from datasets will be published on the website to collect crowdsourced labels. After evaluating results, any algorithms and models created during this project will be published publicly. Note: those who contribute data will not be given authorship unless they have dedicated a significant amount of time to this project (e.g. >50 hours of expert component labeling).

Dataset Criteria

The following is the list of desired and non-desired features of the target population:

Inclusion Criteria

• Neurologically intact (can include elderly population)
• Normal or correct to normal vision
• Age ≥ 12 years

Exclusion Criteria

• Neurological or locomotor (muscular/musculoskeletal) deficits (Parkinson’s Disease, Multiple Sclerosis, Post-stroke hemiparesis, Spinal Cord Injury, Cerebral Palsy)

Requirements

• De-identified participant/patient data
• Number of EEG electrodes ≥ 64
• Sampling rate ≥ 250 Hz
• Must already have an ICA decomposition and dipole model (any ICA decomposition method and head model can be used)

Preferences

• BIDS format
• PowPowCAT results
• Other forms of data acquired during the use of EEG
  • Electromyography (EMG)
  • Electrooculogram (EOG)

How to Contribute Data

If you are interested in contributing data, please fill out this Google form for each study folder you plan to upload. Additionally, please fill and sign one Data Usage Agreement form per lab either using your institution’s template or the one provided and send to icmobiproject@gmail.com. Upon receiving these forms, we will send you a private DropBox link where you can upload your data. Each folder you upload should contain data from only one study, but can contain as many participants as desired. If you wish to add more than one study, please upload each study as a separate folder and fill out the Google form for each study folder uploaded. Make sure both the .set and .fdt files from each subject are in the folder, and compress/zip your folder prior to uploading.

We would greatly appreciate your data. If you have any questions, please contact Noelle Jacobsen at icmobiproject@gmail.com or jacobsen.noelle@ufl.edu.

Thank you,

BeMoBIL, TU Berlin, P.I. Dr. Klaus Gramann

HNL, University of Florida, P.I. Dr. Dan Ferris