---
license: pddl
tags:
- eeg
- medical
- clinical
- classification
- mtbi
- tbi
- oddball
---
# Cavanagh2019: EEG mTBI Classification Dataset with Auditory Oddball Task
The Cavanagh2019 dataset includes EEG recordings collected during a 3-stimulus auditory oddball paradigm in participants with mild traumatic brain injury (mTBI) and matched healthy controls. A total of 96 participants took part: 45 sub-acute mTBI patients (tested within 2 weeks post-injury), 26 healthy controls, and 25 chronic TBI patients (mild to moderate severity). Sub-acute mTBI and control participants completed two or three EEG sessions - at 3-14 days after the injury, and again after approximately 2 months - while chronic TBI participants completed a single session.

The task involved 260 trials: 70% standard tones (440 Hz), 15% target tones (660 Hz), and 15% novel naturalistic sounds. Stimuli were presented binaurally, and participants were instructed to count target tones while ignoring the others. EEG was recorded from 60 channels at a 500 Hz sampling rate.
## Paper
Cavanagh, J. F., Wilson, J. K., Rieger, R. E., Gill, D., Broadway, J. M., Remer, J. H. S., Fratzke, V., Mayer, A. R., & Quinn, D. K. (2019). **ERPs predict symptomatic distress and recovery in sub-acute mild traumatic brain injury**. _Neuropsychologia_, 132, 107125.

DISCLAIMER: We (DISCO) are NOT the owners or creators of this dataset, but we merely uploaded it here, to support our's ([EEG-Bench](https://github.com/ETH-DISCO/EEG-Bench)) and other's work on EEG benchmarking.
## Dataset Structure
- `data/` contains the annotated experiment EEG data.
- `scripts/` contains MATLAB scripts that produced the paper's results.
- `scripts/BigAgg_Data.mat` contains information about the subjects.
- `scripts/QUALITY_CHECK.xlsx` and `scripts/QUINN_QUALITY_CHECK.xlsx` contain information about bad quality recordings.

A `.mat` file can be read in python as follows:
```python
from scipy.io import loadmat
mat = loadmat(filepath, simplify_cells=True)
```
(A field "fieldname" can be read from `mat` as `mat["fieldname"]`.)

Subject information can be read from `scripts/BigAgg_Data.mat` from the following fields (among others):
- `DEMO`: information about mTBI and control subjects
    - `ID`: subject IDs, as included in the filename of the corresponding EEG recording under `data/`
    - `Group_CTL1`: for each subject, whether it belongs to the control group (which is the case if and only if the corresponding `Group_CTL1`-entry is `1`) or not
    - `Sex_F1`: gender of the subject (`1` means female, everything else means male)
    - `Age`: age of the subject
- `Q_DEMO`: information about chronic TBI subjects
    - `URSI`: subject IDs, as included in the filename of the corresponding EEG recording under `data/`
    - `Sex_F1`: gender of the subject (`1` means female, everything else means male)
    - `Age`: age of the subject
- `NP`: mTBI and control subjects' TOMM, TOPF, HVLT and other scores
- `Q_NP`: chronic TBI subjects' TOMM, TOPF, HVLT and other scores
- `QUEX`: mTBI and control BDI and other scores
- `Q_QUEX`: chronic TBI BDI and other scores
- `TBIfields`: information about mTBI subjects' injury
- `Q_TBIfields`: information about chronic TBI subjects' injury

### Filename Format

```
[PID]_[SESSION]_3AOB.mat    (or [PID]_[SESSION]_QUINN_3AOB.mat for chronic TBI participants)
```
PID is the patient ID (e.g. `3001`), while SESSION distinguishes different days of recording (can be `1`, `2` or `3` for patients with mTBI or control patients and is always `1` for patients with chronic TBI). 

### Fields in each File
Let `mat` be an EEG `.mat` file from the `data/` directory.
Then `mat` contains (among others) the following fields and subfields
- `EEG`
    - `data`: EEG data of shape `(#channels, trial_len, #trials)`. E.g. a recording of 247 trials/epochs with 60 channels, each trial having a duration of 4 seconds and a sampling rate of 500 Hz will have shape `(60, 2000, 247)`.
    - `event`: Contains a list of dictionaries, each entry (each event) having the following description:
        - `latency`: The onset of the event, measured as the index in the merged time-dimension `#trials x trial_len` (note `#trials` being the _outer_ and `trial_len` being the _inner_ array when merging). The duration of each event is 200ms. Hence, with a 500 Hz sampling rate, the EEG data `event_data` corresponding to the `i`-th event is
            ```python
            start_index = mat["EEG"]["event"][i]["latency"]
            event_data = numpy.transpose(mat["EEG"]["data"], [1, 2]).reshape([num_channels, num_trials * trial_len])[:, start_index:start_index+100]    # shape (#channels, 100)
            ```
        - `type`: The type of event. Can be `"S200"` (660 Hz tone), `"S201"` (440 Hz tone) or `"S202"` (naturalistic).
    - `chanlocs`: A list of channel descriptors
    - `nbchan`: Number of channels
    - `trials`: Number of trials/epochs in this recording
    - `srate`: Sampling Rate (Hz)

## License
By the original authors of this work, this work has been licensed under the PDDL v1.0 license (see LICENSE.txt).