Standardization of electroencephalography for multi-site, multi-platform and multi-investigator studies: insights from the canadian biomarker integration network in depression

Abstract

Subsequent to global initiatives in mapping the human brain and investigations of neurobiological markers for brain disorders, the number of multi-site studies involving the collection and sharing of large volumes of brain data, including electroencephalography (EEG), has been increasing. Among the complexities of conducting multi-site studies and increasing the shelf life of biological data beyond the original study are timely standardization and documentation of relevant study parameters. We present the insights gained and guidelines established within the EEG working group of the Canadian Biomarker Integration Network in Depression (CAN-BIND). CAN-BIND is a multi-site, multi-investigator, and multi-project network supported by the Ontario Brain Institute with access to Brain-CODE, an informatics platform that hosts a multitude of biological data across a growing list of brain pathologies. We describe our approaches and insights on documenting and standardizing parameters across the study design, data collection, monitoring, analysis, integration, knowledge-translation, and data archiving phases of CAN-BIND projects. We introduce a custom-built EEG toolbox to track data preprocessing with open-access for the scientific community. We also evaluate the impact of variation in equipment setup on the accuracy of acquired data. Collectively, this work is intended to inspire establishing comprehensive and standardized guidelines for multi-site studies.

Figure 1

Potential Sources of Variance in Multi-Site EEG Studies. This flowchart summarizes the major phases of an EEG study in chronological order. For each phase, the diagram lists items/parameters/guidelines that need to be standardized in multi-site/multi-project/multi-investigator EEG studies.

Figure 2

Measuring Time Delays in Task-based Multi-Site EEG Studies. (A) Setup of StimTracker system to measure the delay between the exact time of stimulus presentation (on E-Prime PC) and the time at which the stimulus onset was marked by the physiological data system (in EEG PC) (StimTracker diagram adapted from Cedrus website: http://cedrus.com/support/stimtracker/tn1460_st100_pins.htm) (B) Example of delay measurement when an audio signal is directly measured by the EEG system. (C) Example of auditory stimulus delay measurement with StimTracker system. The figure was drawn by authors using Lucidchart (Lucid Software Inc, UT, USA).

Figure 3

A Streamlined Toolbox for Multi-site EEG Data Processing and Archiving. (A) The main graphical user interface (GUI) of the ERPEEG toolbox, with 7 pre-processing steps. Through this main interface, users select the data (by clicking on Working Folder, and Dataset), and navigate through each preprocessing step. Clicking on a processing step opens a new GUI associated with that step, or runs that processing step. Steps that are completed turn green, and uncompleted steps remain red. (B) The view button (corresponding to each step) provides a visual summary of data cleaning processing (e.g., plots the power spectrum). This enables monitoring of data cleaning progress or detecting any major errors and data distortions. (C) The setting tab allows for selection of user-defined parameters for each step. (D) All intermediate steps (files created in completion of each step) are saved in the working folder following a standardized naming convention.

Figure 4

Illustration of Utility of the ERPEEG Toolbox with Pilot Data. The panels illustrate application of ERPEEG toolbox on pilot data from CAN-BIND project 1. Pilot data include EEG collected during affective Go/noGo task in 15 healthy controls and 15 MDD patients. Raw data were imported into MATLAB via EEGLAB, epoched around onset of each stimulus (e.g., Go or NoGo cue superimposed on angry, happy or neutral faces), and preprocessed with ERPEEG toolbox. Panels A to C illustrate the outcomes of steps 2, 5, and 6 of ERPEEG toolbox, respectively, compared visually between MDD and control groups. Panel D, illustrates the final cleaned ERP, outcome of step 7, compared between MDD and controls for NoGo angry face condition. (A,C) In both panels, x-axes depict two groups of controls and MDD, and y-axes depict channel number, while colors illustrate number of trials removed for each channel averaged across subjects following step 2 (panel A), and step 6 (panel C). (B) In both panels, the x-axis depicts the component type (e.g., component related to eye blinks, bad electrodes, EKG, etc), y-axis shows the subject number, and the colors depict number of components removed in control (top) and MDD (bottom). The matrix of deleted trials depicted in each of these images can be used to systematically compare the data processing across controls and patients, but also across different investigators, and projects.

Figure 5

Illustration of Utility of Unbiased Cluster Estimation Technique. Panels A to D illustrate the impact of varying threshold statistic on the outcome of cluster estimation. The figure is adapted with permission from Frehlich et al.. Data were collected from EEG channel CZ and compared statistically between two conditions. Y-axses represent frequency in Hertz and x-axes time in milliseconds with time 0 indicating application of a transcranial magnetic stimulation pulse. Each dark pixel denotes a voxel in the time-frequency space that meets a statistical criteria given by the threshold statistic. The panels illustrate how changing the threshold alters the number, size of clusters, and interpretation of the findings in time-frequency space.