Data

Data description

The competition dataset includes EEG recordings from over 3,000 participants across six distinct cognitive tasks, divided into passive and active categories (see table below). Each participant’s data is accompanied by four psychopathology dimensions (internalizing, externalizing, attention, and p-factor) derived from a bifactor model of parent-reported questionnaire responses to the Child Behavior Checklist (CBCL). These psychopathology factors represent orthogonal dimensions of mental health and serve as target variables for the regression component of the competition.

More details about the dataset download links, task-specific procedures, video recordings of the experiment, and more can be found on the HBN-EEG dataset page.

Overview of HBN-EEG tasks used in the competition.

Task-specific procedures

Resting State. In the Resting State task, participants rested their heads on a chin rest, looking forward at the computer screen with a fixation cross at the center of the screen. A pre-recorded voice instructed them to keep their eyes closed or open and fixate on the central cross during ‘eyes open’ periods. The timeline of the event markers for the task is shown in the figure.

Surround Suppression. Participants observed a modified Surround Suppression stimulus sequence. Four flashing peripheral disks in the foreground were presented with a contrasting background in two runs, each ~3.6 minutes long. We imported and synchronized foreground and background stimulus parameters from the corresponding behavioral recording for each subject to complete the stimulus event description. Figure 5 shows an example of an actual event-marker stream during the Surround Suppression task.

Movie Watching. The final passive task involved watching movies. Participants were asked to watch four short movies with different themes and aims: (1) ‘Despicable Me’, (2) ‘Diary of a Wimpy Kid’, (3) ‘Fun with Fractals’, and (4) ‘The Present’ [1]. Details and rationale for the movie choices and links to access the presented movies are available in the top-level _eeg.json files. The recorded event markers for each movie presentation include only the start and stop times. We will soon introduce methods and remodeling tools to insert custom event sequences for each movie relative to these event anchors (movie presentation start and stop) so researchers will be able to plug in an event stream of choice.

Contrast Change Detection. In this task, two co-centric flickering grated disks (one with left-leaning gratings and one with right-leaning gratings) appear on the screen. After a randomly selected time, one of the two flickering grated disks on the screen changes contrast. Participants were asked to identify the left-leaning or right-leaning grated disk with dominant contrast as soon as they could identify the disk. Based on their responses, feedback (a smiley face or sad face) was presented. We imported participant responses from the behavioral recordings and assembled HED descriptions for the main task and the respective subject responses.

Sequence learning. Participants were shown a sequence of 10 (or 7 if <8 years old) flashed circles among 8 (or 6) possible and predetermined targets positioned on the periphery of a larger invisible circle on the screen (Figure 7). The same sequence was repeated five times, and after each repetition, participants were asked to repeat the sequence using a computer mouse. The participant’s response sequence was recorded for each repetition, but unfortunately, the timestamps of these responses were not recorded. Because of the difference in the target and sequence counts, we divided this task into two tasks: a six-target task and an eight-target task. The task is based on a similarly designed sequence learning experiment by Steinemann [24].

Symbol search. The final task in the interactive battery of experiments involves an emulation of a standard neuropsychological test (a subset of the Wechsler Intelligence Scale for Children IV, WISC-IV [25]) on the computer screen, in which participants were asked to confirm if either of the target symbols in each row was present among the five search symbols in the same row (Figure 8). The test was presented to the participants in 15-row sequences, and participants were asked to go to the next set of sequences once they responded to all 15 rows. Since each of the 15 questions was presented simultaneously, only the times when participants pressed the mouse button to respond were recorded.

Psychopathology Dimensions and Demographics

The p-factor and psychopathology dimensions are broad, dimensional measures of mental health derived from the Child Behavior Checklist (CBCL), a parent-report tool for assessing behavioral and emotional traits in children and adolescents.

Understanding the Four Psychopathology Dimensions

P-factor (General Psychopathology Factor): The p-factor captures a general tendency for various psychological traits to co-occur, reflecting overall emotional and behavioral vulnerability.

Internalizing: This dimension reflects inward-focused traits such as emotional sensitivity, withdrawal, or somatic complaints.

Externalizing: This dimension describes outward-directed traits like impulsivity, activity level, or rule-challenging behaviors.

Attention: This dimension measures traits related to focus, distractibility, and sustained attention.

Bifactor Model and Advantages

These four dimensions are derived using a bifactor model, which separates general psychopathology (p-factor) from specific, independent trait clusters. This statistical approach provides a nuanced, robust, and privacy-preserving summary of mental health by transforming raw questionnaire responses into anonymized, stable factor scores. This method reduces noise, increases stability and generalizability, and enhances statistical power, while protecting participant privacy by avoiding the use of raw, potentially identifiable responses. Researchers can thus study neural correlates of mental health traits with greater confidence in both confidentiality and scientific rigor.

Demographic Information

Demographic variables available in the dataset include age, sex (male/female), and handedness as assessed by the Edinburgh Handedness Questionnaire. These variables enable researchers to control for or examine the effects of developmental stage, biological sex, and lateralization (such as biases in right-hand and left-hand button presses) in their analyses.

Reference:

• Seyed Yahya Shirazi, Alexandre Franco, Mauricio Scopel Hoffmann, Nathalia Esper, Dung Truong, Arnaud Delorme, Michael Milham, and Scott Makeig. HBN-EEG: The FAIR implementation of the healthy brain network (HBN) electroencephalography dataset. bioRxiv, page 2024.10.03.615261, 3 October 2024. doi: 10.1101/2024.10.03.615261.