The NIMH intramural healthy volunteer dataset: A comprehensive MEG, MRI, and behavioral resource

The NIMH intramural healthy volunteer dataset: A comprehensive MEG, MRI, and behavioral resource


Recruitment and online screening

This study is a convenience sample of healthy persons in the DC metropolitan area interested in participating in research. Inclusion criteria for the study require that participants are adults at or over 18 years of age in good health with the ability to read, speak, understand, and provide consent in English. All participants provided electronic informed consent for online screening and written informed consent for all other procedures. Exclusion criteria include a history of significant or unstable medical or mental health condition requiring treatment; current self-injury, suicidal thoughts or behavior; current illicit drug use by history or urine drug screen; abnormal physical exam or laboratory result at the time of in-person assessment; or less than an 8th grade education or IQ below 70. Current employees, or first-degree relatives of NIMH employees are excluded, although other NIH employees may participate. Study participants are recruited through direct mailings, bulletin boards and listservs, outreach exhibits, print advertisements, and electronic media. Supplemental Table 1 gives a comparison of demographics for the current dataset with demographics from Montgomery County, MD, the location of NIMH. The NIMH sample was somewhat younger and more female than the surrounding population.

All potential volunteers first visit the study website (, check a box indicating consent, and complete preliminary self-report screening questionnaires. The study website is HIPAA compliant and therefore does not collect personally identifiable information (PII); instead, participants are instructed to contact the study team to provide their identity and contact information. The questionnaires include demographics, clinical history including medications, disability status (WHODAS 2.0), mental health symptoms (modified DSM-5 Self-Rated Level 1 Cross-Cutting Symptom Measure), substance use survey (DSM-5 Level 2), alcohol use (AUDIT), handedness (Edinburgh Handedness Inventory), and perceived health ratings. At the conclusion of the questionnaires, participants are again prompted to send an email to the study team. Survey results, supplemented by NIH medical records review (if present), are reviewed by the study team, who determines if the participant is likely eligible to participate as a healthy volunteer based on the inclusion/exclusion criteria. These participants are then scheduled for an in-person assessment. Follow-up phone screenings are also used to determine if participants are eligible for in-person screening.

In-person assessments

At this visit, participants undergo a comprehensive clinical evaluation to determine final eligibility to be included as a healthy research volunteer. The mental health evaluation consists of a psychiatric diagnostic interview (Structured Clinical Interview for DSM-5 Disorders (SCID-5), along with self-report surveys of mood (Beck Depression Inventory-II (BD-II) and anxiety (Beck Anxiety Inventory, BAI) symptoms. An intelligence quotient (IQ) estimation is determined with the Kaufman Brief Intelligence Test, Second Edition (KBIT-2) (n = 267). The KBIT-2 is a brief (20–30 minute) assessment of intellectual functioning administered by a trained examiner. There are three subtests, including verbal knowledge, riddles, and matrices.

Medical evaluation includes medical history elicitation and systematic review of systems. Biological and physiological measures include vital signs (temperature, blood pressure, pulse), as well as weight, height, and BMI. Blood and urine samples are taken and a complete blood count, acute care panel, hepatic panel, thyroid stimulating hormone, viral markers (HCV, HBV, HIV), C-reactive protein, creatine kinase, urine drug screen, and urine pregnancy tests are performed. In addition, blood samples that can be used for future genomic analysis, development of lymphoblastic cell lines, or other biomarker measures are collected and banked with the NIMH Repository and Genomics Resource (Infinity BiologiX). Any future assessments on stored samples will be shared as they are available. The Family Interview for Genetic Studies (FIGS) was later added to the assessment in order to provide better pedigree information; the Adverse Childhood Events (ACEs) survey was also added to better characterize potential risk factors for psychopathology. The entirety of the in-person assessment not only collects information relevant for eligibility determination, but it also provides a comprehensive set of standardized clinical measures of volunteer health that can be used for secondary research.

MRI Scan

Participants who were determined to be eligible for inclusion as healthy research volunteers based on the in-person assessment are given the option to consent for a magnetic resonance imaging (MRI) scan, which can serve as a baseline clinical scan to determine normative brain structure as well as a research scan with the addition of functional sequences (resting state and diffusion tensor imaging). Details of scan types are given below and in Table 2.The MR protocol used was initially based on the ADNI-3 basic protocol12, but was later modified to include portions of the ABCD13 protocol. Because there may be small changes in parameters from the standard ABCD/ADNI3 sequences, detailed sequence descriptions are shared in the BIDS sourcedata directory. Additional images collected with parameters inconsistent with the primary dataset are also shared in the sourcedata directory with detailed metadata files so that investigators can include them in analyses at their discretion; numbers for each scan type are given in Supplemental Table 2. High resolution hippocampal scans were originally part of the battery but removed due to time constraints, thus all collected scans are in sourcedata. Some images were acquired with and without GE’s proprietary surface coil intensity correction algorithm applied, these are designated “rec-SCIC” in the repository. Wherever available, both image types were shared. Scan types are as follows, please refer to Table 3 for numbers:

  • The T1 scan from ADNI3 (fSPGR) was initially acquired, but was later replaced the T1 scan from ABCD (MPRAGE)

  • The 2D FLAIR sequence from ADNI2

  • The 3D FLAIR sequence from ADNI3, altered to match the resolution and geometry of the T1 scan (this scan was optional)

  • The ADNI3 T2* weighted scan

  • The 3D T2 weighted scan from the ABCD protocol resolution and bandwidth matched to the T1 scan

  • The ADNI3 pCASL scan, altered to add fat saturation

  • The DTI scan from ADNI3 was modified to include the slice-select gradient reversal method (for 24 directions) and to turn reconstruction interpolation off.

  • The eyes-open resting state from ADNI3 was modified to use a TE of 16.9 ms and was acquired together with 1- phase-encoding reversed volumes

  • Field maps for both DTI and rsfMRI were acquired

On the same visit as the MRI scan, volunteers are administered a subset of tasks from the NIH Toolbox Cognition Battery. The four tasks asses attention and executive functioning (Flanker Inhibitory Control and Attention Task), executive functioning (Dimensional Change Card Sort Task), episodic memory (Picture Sequence Memory Task), and working memory (List Sorting Working Memory Task).

MEG recording

An optional MEG study was added to the protocol approximately one year after the study was initiated, thus there are relatively fewer MEG recordings in comparison to the MRI dataset. All participants eligible for MRI who did not have contraindications such as implanted metal or dental hardware (which would reduce data quality) were offered participation in MEG. MEG studies are performed on a 275 channel CTF MEG system (CTF MEG, Coquiltam BC, Canada), using third-order gradient balancing for noise correction. All datasets were collected at a sampling rate of 1200 Hz, with a quarter-Nyquist filter of 300 Hz. The position of the head is localized at the beginning and end of each recording using three fiducial coils. These coils are placed 1.5 cm above the nasion, and at each ear, 1.5 cm from the tragus on a line between the tragus and the outer canthus of the eye. For 50 participants, photographs were taken of the three coils and used to mark the points on the T1 weighted structural MRI scan for co-registration. For the remainder of the participants (n = 17), a Brainsight neuronavigation system (Rogue Research, Montréal, Québec, Canada) was used to coregister the MRI and fiducial localizer coils in real-time prior to MEG data acquisition.

The MEG task battery was designed to assess multiple cognitive domains which are relevant to neuropsychiatric disorders. The MEG battery is divided into two parts, with additional participant/equipment set-up in between. All tasks are coded in either PsychoPy or Presentation software and are available on GitHub ( In Supplemental Table 3, we give the number of trials per condition and timing for each task, with a summary of the tasks presented here. Accuracy and reaction time can be calculated from the marker timing files for all tasks. Prior to any of the tasks, a brief “artifact” recording was acquired. Participants were asked to blink, move their eyes, breathe deeply, clench their jaw, and swallow. The first half of the MEG session consists of a modified Hariri Hammer task, a modified Sternberg task, and a resting state acquisition. The order is counterbalanced across subjects, with either the Hariri Hammer or Sternberg occurring first or last, and the resting state acquisition always acquired between the two tasks. The Hariri Hammer task was originally developed for fMRI14, and later adapted for MEG15 as a sensitive probe of amygdala function. The original task presents three faces in a triangle, with the target stimuli on the top and match stimuli below. The participant is required to select from the two faces on the bottom to match the emotion of the target stimuli above. We further adapted the task by temporally separating the target and match stimuli with a small delay and fixation cross. The target face is presented first alone, centrally, followed by a fixation cross, and then the two (probe) faces presented centrally as a pair. The delay was incorporated to isolate the evoked response to a single emotional face. This change allows studies investigating whether the trial-to-trial variation in the magnitude or duration of the response to the first face affects the reaction time or accuracy of the choice, and whether this relationship is modulated by emotion. The Sternberg task was developed as a probe of working memory16, and has been previously used in MEG17. The original version sequentially presents a series of digits or letters, and after a delay, presents a single digit or letter, and the participant is asked to indicate whether the letter or digit appeared in the series. The version used here presents the series in its entirety, rather than sequentially, in order to reduce the time required to complete the task. We present two conditions: four-letter strings and six-letter strings. The resting state scan is 6 minutes in duration, and participants are given no specific instructions other than to close their eyes and remain still.

Following a brief break to set up stimulus delivery equipment, participants receive a somatosensory task. A brief tactile stimulus is delivered to the right index finger using pneumatic pressure on a thin flexible membrane. In order to measure both the response to the stimulus as well as the expectation of the stimulus, 15% of all stimuli are omitted. Following the somatosensory task, participants perform either a go/no-go task or a three-stimulus oddball task. Order is counterbalanced across participants, and a naturalistic viewing task is always performed between the two. The go/no-go task was similar to previous implementations for MEG18. Briefly, participants are rapidly presented outlines of shapes, and respond to every shape unless there is an “X” in the middle. The three-stimulus oddball consists of three stimuli presented in one of four randomized orders. There is a standard tone, a higher-pitched rare tone, and a white noise stimulus; participants are asked to respond via button press to the high-pitched rare tone. The naturalistic viewing task consists of the approximately 9-minute short film “Growth,” by Sil van der Woerd ( The film contains audio, but no dialogue, and is presented in its entirety (excluding the credits, which were distributed on paper to all participants after the session).

In addition to the subject datasets, we additionally acquire empty room datasets. Because the CTF sensors are quite stable, we collect these scans approximately monthly. The scans are 100 s in duration, collected with a sampling frequency of 4800 Hz. Although the BIDS specification recommends that these datasets be placed in a directory separate from research participant data, and organized by date, this is unsatisfactory because we have removed the data of scan from the human datasets. In order to maximize ease of use for researchers, we have placed the appropriate empty room dataset in the same directory as the participant MEG data. Because empty room datasets were not acquired before every scan, some subject directories will share the same empty room dataset.

Preprocessing methods

We distribute the data in a minimally processed, raw format. However, in order to facilitate data analysis, the MRI data are converted to NIfTI and transformed into BIDS format19 using Dcm2Bids version 2.1.6 (, which is a wrapper for dcm2niix version 1.0.20211006 ( To preserve subject privacy, structural MRI scans are defaced using AFNI Refacer version 2.31 ( First with AFNI Refacer, a single T1-weighted image was defaced from each of the 155 participants and a defaced image and mask were produced. Next every other image was rigid-body-aligned to its participant’s defaced image with a mutual information cost function using FSL’s FLIRT21,22,23. Last, the deface mask was aligned using the FLIRT alignment matrix to each subject’s other images. The alignment procedure worked well in all cases except one. That one image was instead masked with no registration as it overlaid quite well on the originally defaced image.

The MEG data were transformed to BIDS format using MNE-BIDS ( {Appelhoff, 2019 #36;Niso, 2018 #38} and additional python scripts written locally. For all MEG tasks, we ran custom python scripts that generated marks for all stimulus types for the cognitive tasks (these are available as a git submodule called “hv_proc” in Because the data is shared in the proprietary CTF “.ds” data structure format, these markers are saved within the “.ds” directory in a file entitled Marker.mrk. In addition, the markers are also stored in “events.tsv” sidecar files in the meg directory according to the BIDS MEG specification. While the stimulus delivery software sends triggers indicating the type of stimulus, these are not coincident with the appearance on screen due to delays induced by the refresh rate of the projector. A ProPixx projector was used for all visual presentation, and the upper left pixel of the projected image can be used to encode the precise time at which the stimulus is displayed. Thus, all marks we indicate in the recordings have been adjusted so that they coincide with the precise time of display. In addition, marks for somatosensory and auditory stimuli have been adjusted for delays in stimulus delivery by subtracting the mean measured delay between parallel port onset and stimulus delivery.

We distribute the location of the MEG fiducial coils in the coordinate space of the anatomical MRI. No other pre-processing or filtering was performed.


Source link

Jorge Oliveira