MRI and EEG Data
Data Overview |
Data Download |
Data Quality |
Data License |
Notes |
|---|
☰
Data Overview
MRI Protocol
EEG Protocol
- Resting State
- Sequence Learning Paradigm
- WISC Symbol Search (Processing Speed)
- Inhibition/Excitation Paradigm (Surround Suppression) 1
- Visual Perception/Decision-Making Paradigm (Contrast Change) 1
- Video 1
- Visual Perception/Decision-Making Paradigm (Contrast Change) 2
- Video 2
- Visual Perception/Decision-Making Paradigm (Contrast Change) 3
- Video 3
- Inhibition/Excitation Paradigm (Surround Suppression) 2
- Video 4 (The Present)
For EEG data interpretation and organization, click below to download organizational ReadMe files that describe event triggers, behavioral data labels, and EEG channel location.
- EEG README - .zip file containing a general ReadMe file, a ReadMe file for each EEG paradigm, and a WISC Symbol Search key in both excel and .mat format.
- EEG Channel Location - .sfp file containing the channel location of the EEG montage. More information about the EGI to 10-10 Layout Comparison can be found here and here.
Data Download
- Click on the EEG and MRI download links below, you will be prompted to log on to NITRC.
- Register with the 1000 Functional Connectomes Project website on NITRC to gain access to the HBN neuroimaging datasets.
- Use the checkboxes to select which subjects you would like to download or access the data through Amazon Web Services (AWS) and Cyberduck.
- Basic_Info Files contain the following information for each participant with MRI/EEG data:
- Sex: 1 = Female, 0 = Male
- Age
- EHQ_Total: +100 = Fully Right-Handed, -100 = Fully Left-Handed
- Commercial_Use: Whether the participant consented to commercial use of data (Yes/No)
- full_pheno: Whether the participant has a full phenotypic file (Yes/No)
| Release 1.1 (1/31/2018) |
EEG (603 subjects) | MRI (616 subjects) | Basic_Info File |
| Release 2.1 (1/31/2018) |
EEG (156 subjects) | MRI (159 subjects) | Basic_Info File |
| Release 3 (1/31/2018) |
EEG (188 subjects) | MRI (202 subjects) | Basic_Info File | Release 4 (7/31/2018) |
EEG (359 subjects) | MRI (402 subjects) | Basic_Info File |
| Release 5 (11/27/2018) |
EEG (272 subjects) | MRI (205 subjects) | Basic_Info File | Release 6 (02/15/2019) |
EEG (213 subjects) | MRI (174 subjects) | Basic_Info File | Release 7 (09/17/2019) |
EEG (476 subjects) | MRI (438 subjects) | Basic_Info File | Release 8 (01/03/2020) |
EEG (361 subjects) | MRI (309 subjects) | Basic_Info File | Release 9 (12/07/2020) |
EEG (364 subjects) | MRI (243 subjects) | Basic_Info File | Release 10 (04/13/2022) |
EEG (576 subjects) | MRI (700 subjects) | Basic_Info File | Release 11 (11/23/2022) |
EEG (572 subjects) | MRI (459 subjects) | Basic_Info File |
Please note that data in AWS are not organized by Release and that imaging data are released regardless of whether or not participants completed the remainder of the study. Therefore, some participants with imaging data may not have full phenotypic data available until a future release.
Data Quality
Consistent with policies established through our prior data generation and sharing initiatives (i.e., FCP/INDI (Mennes et al. 2013); NKI-Rockland Sample (Nooner et al. 2012)), all imaging datasets collected through the HBN are being made available to users regardless of data quality. This decision is justified by a lack of consensus in the imaging community on what constitutes “good” or “poor” quality data. Also, “lower quality” datasets can facilitate the development of artifact correction techniques and of evaluating the impact of such real-world confounds on reliability and reproducibility. Given the range of clinical presentations in the HBN, the inclusion of datasets of varying qualities creates a unique opportunity to test for associations with participant-related variables of interest beyond age and hyperactivity (e.g., anxiety, autistic traits).
MRI Data
Consistent with recent major FCP/INDI data releases (i.e., the Consortium for Reliability and Reproducibility [CoRR] (Zuo et al. 2014), Autism Brain Imaging Data Exchange 2 [ABIDE 2] (Di Martino et al. 2017)), we made use of the Preprocessed Connectome Project Quality Assurance Protocol (QAP; Shehzad et al., 2013) to assess data quality for core MRI data modalities (i.e., functional MRI, morphometry MRI and diffusion MRI). The QAP includes a broad range of quantitative measures that have been proposed for assessing image data quality (see table below for list of measures and their definitions, adapted from (Di Martino et al. 2017)). See the table below for a description:
Given commonly cited concerns about head motion during MRI scans, particularly during resting state fMRI scans, age-related differences in motion were examined. Head motion was quantified using frame-wise displacement (FD), which is calculated using root mean square deviation (Jenkinson et al. 2002). Mean FD is commonly used to evaluate the impact of movement on a dataset (Power et al. 2015), but it cannot distinguish between occasional large movements and frequent smaller movements, the effects of the former being likely easier to fix using scrubbing or an volume censoring methods. Consistent with this concern, the following figure demonstrates a nonlinear relationship between mean FD and median FD, with the latter providing a better indication of the amount of the data that can be retained after movement correction (e.g., volume censoring).
Consistent with prior work (Power et al. 2012), the 1.5 Tesla mobile scanner in Staten Island and the 3 T fixed scanner at Rutgers University exhibited negative associations between age and head motion for all functional scan types, with children between ages 5 and 8 exhibiting the greatest levels of movement. Median FD tended to be higher during the second half (5 minutes) of the resting state scan than during the first half; this observation resulted in our decision to split the scan into two 5-minute scans. As predicted by recent work highlighting the advantages of naturalistic viewing to minimize head motion (Vanderwal et al. 2015, Vanderwal et al. 2017), we found that head motion was significantly reduced during each of the movie-watching scan sessions (“Despicable Me,” “The Present”) relative to rest.
Beyond the examination of temporal characteristics of the HBN data, we also applied the structural measures included in the PCP QA to each of the core data types collected (functional, diffusion, morphometry). See graphs below for a subset of these measures. More information about the quality control process used for HBN MRI data can be found at the PCP Assurance Protocol webpage.
View GraphsDownload MRI QAP Data
Data License
HBN imaging, EEG, and eyetracking datasets for participants are currently distributed under the Creative Commons, Attribution Non-Commercial Share Alike License, as they were collected prior to the addition of consent for commercial use to the informed consent. Moving forward, HBN datasets are being distributed using Creative Commons BY 4.0 License, which does allow for commercial use of datasets. For the high dimensional phenotypic data, all terms specified by the DUA must be complied with.
Notes
Data losses
Some participants may not be able to successfully complete all components of the HBN protocol due to a variety of factors (e.g., participants experiencing claustrophobia may not be able to stay in the scanner for the full session, a participant with sensory issues may have a more limited ability to participate in the EEG protocol). To prevent data loss when possible, we include exposure procedures such as a mock MRI scanner experience during session 1, and repeat exposures to an EEG cap prior to session 4. Overall, we attempt to collect as much of the data as possible within the allotted data collection intervals and code data losses when they occur.
Handling Head Motion in MRI Data
Head motion presents an unavoidable challenge for developmental and clinical imaging, regardless of MRI modality (fMRI, dMRI, sMRI). Arguably, the most basic strategy for handling motion, short of applying an uncomfortable motion-restricting apparatus, is limiting analyses to high-quality datasets. The Brain Genomic Superstruct data release is an excellent example of the utility of large-scale datasets in supporting such a strategy, as 1570 datasets were selected for analyses from a pool of 3000 individuals following rigorous quality control (Holmes et al. 2015). A limitation of this strategy for psychiatric data is that many phenotypes of interest are inherently more prone to head motion (e.g., children under 9, those with Attention-Deficit/Hyperactivity Disorder), especially those with higher levels of symptomatology. Compounding the downsides of discarding data are the increased costs associated with the recruitment and phenotyping of clinical populations.
For functional MRI, an alternative strategy is to statistically correct the data for movement-induced intensity fluctuations, or remove offending time frames altogether (Power et al. 2015). This can be accomplished by a number of means, ranging from regressing a model of movement from the data (e.g., spike regression (Satterthwaite et al. 2013)), removing the contributions of motion-related spatial patterns from the data (AROMA (Pruim et al. 2015)), attenuating motion spikes using a squashing function, removing offending frames, zeroing out offending frames, or deleting offending frames followed by interpolation. More generalized correction approaches, such as global signal regression and forms of white matter and cerebrospinal fluid regression (e.g., tCompCor, aCompCor (Behzadi et al. 2007; Chai et al. 2012)) can also help to account for motion artifacts. While there is no consensus approach to date, there is a growing literature focused on providing benchmark evaluations of these approaches, as well as their relative merits and weaknesses (e.g., see Ciric et al. 2017; Yan et al. 2013), that can be used to help select among these corrections.
More broadly, group-level statistical corrections can be used to account for the contributions of motion-related artifacts to associations revealed through data analysis (Satterthwaite et al. 2013). In the case of functional MRI, this can be accomplished by including motion parameters as a statistical covariate at the group level. Given the trait nature of head motion (Zuo et al. 2014), some have advocated for using fMRI-derived motion parameters in structural analysis as well. Alternatively, accounting for full-brain differences in measures of interest at the group-level has been shown to be a potentially valuable approach to minimizing the deleterious effects of motion, particularly for fMRI (Yan et al. 2013).
It is our hope that the breadth of the Healthy Brain Network dataset will provide a practical perspective of the challenges of motion for various domains of illness and help to stimulate continued development and testing of novel correction strategies.