Database Open Access
Motion Artifact Contaminated fNIRS and EEG Data
Published: March 3, 2014. Version: 1.0.0
Sweeney KT, Ayaz H, Ward TE, Izzetoglu M, McLoone SF, Onaral B. A Methodology for Validating Artifact Removal Techniques for Physiological Signals. IEEE Trans Info Tech Biomed 16(5):918-926; 2012 (Sept).
Please include the standard citation for PhysioNet:
(show more options)
Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P. C., Mark, R., ... & Stanley, H. E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation [Online]. 101 (23), pp. e215–e220.
This data collection, contributed to PhysioBank by Kevin Sweeney and colleagues at the National University of Ireland in Maynooth, contains examples of functional near-infrared spectroscopy (fNIRS) and electroencephalogram (EEG) recordings that have been created for evaluating artifact removal methods.
In each such recording, one or two pairs of similar physiological signals have been acquired from transducers in close proximity. (The EEG recordings contain one pair of signals. The fNIRS recordings contain two pairs of such signals, acquired at 690 nm and 830 nm wavelengths.)
In each case, one of the two transducers has not been disturbed, while the other has been manipulated to produce motion artifacts of variable duration within each 2-minute interval of the recording. The motion of the disturbed transducer, and the lack of motion of the other transducer, are documented in each case by simultaneously recorded outputs of 3-axis accelerometers affixed to each transducer. Comparison of the "ground truth" signals obtained from the undisturbed transducer with the intermittently noisy signals obtained from the other transducer reveals high correlation during motion-free intervals and lower correlation during artifact-contaminated intervals. To evaluate the efficacy of artifact-removal methods for "cleaning" the contaminated signals, one may similarly compare the "ground truth" signals with the "cleaned" signals.
The first reference above (Sweeney 2012) describes how these recordings were made in detail, and illustrates their use in evaluation of a wide variety of previously described artifact removal methods. The second reference above (Sweeney 2013) presents a novel method, also evaluated using these recordings.
The original data recorded by Sweeney and colleagues are available here within a pair of zip archives in CSV format. Note that the fNIRS (or EEG) signals and the accelerometer signals were recorded simultaneously using independent recording systems; a software-generated trigger signal was recorded by both systems. In order to synchronize the accelerometer signals with the other signals, it is necessary to line up the corresponding transitions in the recorded trigger signals.
Two versions of the trigger signals were produced. In the full version, recorded together with the fNIRS signals and with the accelerometer signals, the trigger signal initially rises to indicate the start of the experiment. The trigger signal transitions to a low value during intervals when motion artifact was produced, and returns to a high level in the clean intervals. The final transition from high to low signals the end of the experiment.
The version of the trigger signal recorded together with the EEG signals includes only the initial rise and final fall (marking the start and end of the experiment); it does not include the intervening transitions marking the noisy and clean intervals of the recording.
The EEG signals and the accompanying trigger signal were digitized at 2048 Hz; the fNIRS signals and their trigger signal, at 25 Hz [*]; and the 3-axis accelerometer signals and their trigger signal, at 200 Hz.
[*] Analysis of the trigger signals in the fNIRS recordings suggests that the actual sampling frequency was slightly slower, and was approximately 24.99305 Hz. This value was used to resynchronize the fNIRS signals with the accelerometer signals in the PhysioBank-compatible versions of these recordings (see below).
fNIRS CSV files
Each of the 9 trials included in fNIRS-csv.zip consists of 2 channels of highly correlated fNIRS data recorded from the pre-frontal cortex. Therefore each trial presents 4 signals (2x690 nm and 2x830 nm recordings). Note: In trials 5 and 8, the 690 nm recordings are poor and were not used in the studies referenced above and below. Columns in the TrialX.csv files represent:
Column 1: Sample index Column 2: Channel 1 : Raw 690 nm intensity : sampled @ 25 Hz Column 3: Channel 1 : Raw 830 nm intensity : sampled @ 25 Hz Column 4: Channel 2 : Raw 690 nm intensity : sampled @ 25 Hz Column 5: Channel 2 : Raw 830 nm intensity : sampled @ 25 Hz Column 6: Trigger data for fNIRS data : sampled @ 25 Hz Column 7: Accelerometer 1 : X-axis : sampled @ 200 Hz Column 8: Accelerometer 1 : Y-axis : sampled @ 200 Hz Column 9: Accelerometer 1 : Z-axis : sampled @ 200 Hz Column 10: Accelerometer 2 : X-axis : sampled @ 200 Hz Column 11: Accelerometer 2 : Y-axis : sampled @ 200 Hz Column 12: Accelerometer 2 : Z-axis : sampled @ 200 Hz Column 13: Trigger data for accelerometer data : sampled @ 200 Hz
EEG CSV files
Each of the 23 trials included in EEG-csv.zip consists of 2 channels of highly correlated EEG data recorded from the pre-frontal cortex. Columns in the TrialX.csv files represent:
Column 1: Sample index Column 2: Channel 1 : Raw EEG : sampled @ 2048 Hz Column 3: Channel 2 : Raw EEG : sampled @ 2048 Hz Column 4: Trigger data for EEG data : sampled @ 2048 Hz Column 5: Accelerometer 1 : X-axis : sampled @ 200 Hz Column 6: Accelerometer 1 : Y-axis : sampled @ 200 Hz Column 7: Accelerometer 1 : Z-axis : sampled @ 200 Hz Column 8: Accelerometer 2 : X-axis : sampled @ 200 Hz Column 9: Accelerometer 2 : Y-axis : sampled @ 200 Hz Column 10: Accelerometer 2 : Z-axis : sampled @ 200 Hz Column 11: Trigger data for accelerometer data : sampled @ 200 Hz
The *.hea, *.dat, and *.trigger files available here were derived from the contents of the CSV files described above. The signals are presented in the same order as for the columns of the original CSV files. The locations of the transitions in the trigger signals were used to resynchronize the data in each recording. The short intervals before the initial rise and after the final fall of the trigger signals were not included in these files. The *.trigger files are annotation files that mark the locations of the transitions in the synchronized trigger signals; R annotations mark rises in the trigger signal, and F annotations mark falls in the trigger (the beginnings of artifact-contaminated intervals). The fNIRS signals were digitally resampled at 200 Hz to match their accompanying accelerometer signals in these records. In the EEG records, the accelerometer signals were digitally resampled at 2048 Hz to match the EEG signals.
- Sweeney KT, Ayaz H, Ward TE, Izzetoglu M, McLoone SF, Onaral B. A Methodology for Validating Artifact Removal Techniques for Physiological Signals. IEEE Trans Info Tech Biomed 16(5):918-926; 2012 (Sept).
- Sweeney KT, McLoone SF, Ward TE. The Use of Ensemble Empirical Mode Decomposition With Canonical Correlation Analysis as a Novel Artifact Removal Technique. IEEE Trans Biomed Eng 60(1):97-105; 2013 (Jan).
Anyone can access the files, as long as they conform to the terms of the specified license.
License (for files):
Open Data Commons Attribution License v1.0
Total uncompressed size: 649.9 MB.
Access the files
- Download the ZIP file (649.9 MB)
- Access the files using the Google Cloud Storage Browser here. Login with a Google account is required.
- Access the data using the Google Cloud command line tools (please refer to the gsutil documentation for guidance):
gsutil -m -u YOUR_PROJECT_ID cp -r gs://motion-artifact-1.0.0.physionet.org DESTINATION
- Download the files using your terminal:
wget -r -N -c -np https://physionet.org/files/motion-artifact/1.0.0/