Jiaheng-Wang/ZJU-Children-Emotion

ZJU Children Emotion Dataset

Dataset Description

ZJU Children Emotion Dataset (ZCED) is a multi-modal multi-group children emotion dataset aimed for the research of disease classification and emotion recognition in children with neurodevelopmental disorders. The ZCED dataset contains both behavioral and physiological recordings based on a video-based emotional stimulation paradigm for four groups of children including 19 TD, 15 ASD, 20 ADHD, and 18 ASD+ADHD. Data are synchronously recorded in the .xdf format using the LSL protocol. Preprocessed EEG epochs are also provided in the .mat format. There are a total of 12 emotional videos categorized by positive (happy), neutral, and negative (fear) emotion conditions. The detailed illustration of the experimental protocol can be referred in the paper ** **.

How to load the dataset

from datasets import load_dataset

ds = load_dataset("Jiaheng-Wang/ZJU_Children_Emotion")

Dataset Structure

data
├── TD
│   ├── TD_2024-02-23_01
│   │   ├── TD_2024-02-23_01.xdf
│   │   └── preprocessed.mat
│   └── ...
├── ADHD
│   ├── ADHD_2023-11-11_01
│   │   ├── ADHD_2023-11-11_01.xdf
│   │   └── preprocessed.mat
│   └── ...
├── ASD
│   ├── ASD_2023-10-09_01
│   │   ├── ASD_2023-10-09_01.xdf
│   │   └── preprocessed.mat
│   └── ...
├── ASD+ADHD
│   ├── ASD+ADHD_2023-10-02_01
│   │   ├── ASD+ADHD_2023-10-02_01.xdf
│   │   └── preprocessed.mat
│   └── ...
├── MetaData
│   ├── BioSemi32.xyz
│   └── ...

xdf file format

Pysiological signals are first transmitted by App-BioSemi and App-TobiiPro using the LSL protocol. Then data are synchronously recorded with the Labrecorder software using the LSL protocol. In the [group]_[date]_[id].xdf file, we store physiological signals along with event markers. We recommend to use EEGLAB for the primary data preprocessing in MATLAB. Also, pyxdf along with MNE is a good choice in Python.

• EEG: EEG data is stored as part of the LSL stream named 'BioSemi'. 32-channel EEG signals correspond to Channel_2-33. Chanel_34 and 35 stand for the left and right mastoid references, respectively. The electrode layout is provided in Metadata/BioSemi32.xyz.

• ECG: ECG data is stored as part of the LSL stream named 'BioSemi'. It is represented by the differential signal of Channel_36 and Channel_37.

• Eye-tracking: Eye-tracking data is stored as the LSL stream named 'Tobii'. There are six channels representing the metrics of left-eye X, left-eye Y, left-eye pupil, right-eye X, right-eye Y, and right-eye pupil, respectively.

• Event Markers: Event markers are stored as the LSL stream named 'Emotion_Markers'. Events are encoded by unique int values. Video IDs are sent at the beginning of each video trial, while video types and self-reported ratings are sent at the end of each video trial.

  Event Code	Descripition
  1-12	Video ID
  101, 102, 103	Video types (positive, neutral, negative)
  201, 202, 203	Self-reported ratings (negative, neutral, positive)

mat file format

In the preprocessed.mat file, we store EEG epochs along with emotion labels extracted through a basic preprocessing pipeline. Briefly, filtering, downsampling, bad channel interpolation, common average reference, epoching, bad trial rejection, independent component analysis (ocular artifact removal), and amplitude thresholding are performed sequentially. The data contains two fields.

• EEG_data: [32×1024×N] 32 channels, 4 s with 256 Hz sampling rate, N epochs.

• labels: [N] N epochs.

facial videos

For personal privacy and security considerations, facial videos are only available from the corresponding authors on reasonable requests.

Citation

Under review.

Misc.

There is only an illustration sample of TD at present. The whole dataset will be released soon.

We are grateful to the participating families and all staff involved in the data collection. If you find any problems, please contact with us by [email protected]