ReBeatICG database contains ICG (impedance cardiography) signals recorded during an experimental session of a virtual search and rescue mission with drones. It includes beat-to-beat annotations of the ICG characteristic points, made by a cardiologist, with the purpose of testing ICG delineation algorithms. A reference of synchronous ECG signals is included to allow comparison and mark cardiac events.
Raw data
The database includes 48 recordings of ICG and ECG signals from 24 healthy subjects during an experimental session of a virtual search and rescue mission with drones, described in [1]. Two segments of 5-minute signals are selected from each subject; one corresponding to baseline state (task BL) and the second one is recorded during higher levels of cognitive workload (task CW). In total, the presented database consisted of 240 minutes of ICG signals.
During the experiment, various signals were recorded, but here only ICG and ECG data are provided. Raw data was recorded with 2000Hz using the Biopac system.
Data Preprocessing (filtering)
Further, for the purpose of annotation by cardiologists, data were first downsampled to 250Hz instead of 2000Hz. Further, it was filtered with an adaptive Savitzky-Golay filter of order 3. “Adaptive” refers to the adaptive selection of filter length, which plays a major role in the efficacy of the filter. The filter length was selected based on the first 3 seconds of each signal recording SNR level, following the procedure described below.
Starting from a filter length of 3 (i.e., the minimum length allowed), the length is increased in steps of two until signal SNR reaches 30 or the improvements are lower than 1% (i.e., the saturation of SNR improvement with further filter length increase). These values present a good compromise between reducing noise and over-smoothing of the signal (and hence potentially losing valuable details) and a lower filter length, thus reducing complexity. The SNR is calculated as a ratio between the 2-norm of the high and low signal frequencies considering 20Hz as cut-off frequency.
Data Annotation
In order to assess the performance of the ICG delineation algorithms, a subset of the database was annotated by a cardiologist from Lausanne University Hospital (CHUV) in Switzerland.
The annotated subset consists of 4 randomly chosen signal segments containing 10 beats from each subject and task (i.e., 4 segments from BL and 4 from CW task). Segments of signals with artifacts and very noisy were excluded when selecting the data for annotation, and in this case, 8 segments were chosen from the task with cleaner signals. In total, 1920 (80×24) beats were selected for annotation.
For each cardiac cycle, four characteristic points were annotated: B, C, X and O. The following definitions were used when annotating the data:
– C peak — Defined as the peak with the greatest amplitude in one cardiac cycle that represents the maximum systolic flow.
– B point — Indicates the onset of the final rapid upstroke toward the C point [3] that is expressed as the point of significant change in the slope of the ICG signal preceding the C point. It is related to the aortic valve opening. However, its identification can be difficult due to variations in the ICG signals morphology. A decisional algorithm has been proposed to guide accurate and reproducible B point identification [4].
– X point — Often defined as the minimum dZ/dt value in one cardiac cycle. However, this does not always hold true due to variations in the dZ/dt waveform morphology [5]. Thus, the X point is defined as the onset of the steep rise in ICG towards the O point. It represents the aortic valve closing which occurs simultaneously as the T wave end on the ECG signal.
– O point — The highest local maxima in the first half of the C-C interval. It represents the mitral valve opening.
Annotation was performed using open-access software (https://doi.org/10.5281/zenodo.4724843).
Annotated points are saved in separate files for each person and task, representing the location of points in the original signal.
Data structure
Data is organized in three folders, one for raw data (01_RawData), filtered data (02_FilteredData), and annotated points (03_ExpertAnnotations). In each folder, data is separated into files representing each subject and task (except in 03_ExpertAnnotations where 2 CW task files were not annotated due to an excessive amount of noise).
All files are Matlab .mat files.
Raw data and filtered data .mat files contain „ICG“, „ECG“ synchronized data, as well as “samplFreq“values. In filtered data final chosen Savitzky-Golay filter length (“SGFiltLen”) is provided too.
In Annotated data .mat file contains only matrix „annotPoints“ with each row representing one cardiac cycle, while in columns are positions of B, C, X and O points, respectively. Positions are expressed as a number of samples from the beginning of full database files (signals from 01_RawData and 02_FilteredData folders). In rare cases, there are less than 40 (or 80) values per file, when data was noisy and cardiologists couldn’t annotate confidently each cardiac cycle.
——————-
References
[1] F. Dell’Agnola, “Cognitive Workload Monitoring in Virtual Reality Based Rescue Missions with Drones.,” pp. 397–409, 2020, doi: 10.1007/978-3-030-49695-1_26.
[2] H. Yazdanian, A. Mahnam, M. Edrisi, and M. A. Esfahani, “Design and Implementation of a Portable Impedance Cardiography System for Noninvasive Stroke Volume Monitoring,” J. Med. Signals Sens., vol. 6, no. 1, pp. 47–56, Mar. 2016.
[3] A. Sherwood(Chair), M. T. Allen, J. Fahrenberg, R. M. Kelsey, W. R. Lovallo, and L. J. P. van Doornen, “Methodological Guidelines for Impedance Cardiography,” Psychophysiology, vol. 27, no. 1, pp. 1–23, 1990, doi: https://doi.org/10.1111/j.1469-8986.1990.tb02171.x.
[4] J. R. Árbol, P. Perakakis, A. Garrido, J. L. Mata, M. C. Fernández‐Santaella, and J. Vila, “Mathematical detection of aortic valve opening (B point) in impedance cardiography: A comparison of three popular algorithms,” Psychophysiology, vol. 54, no. 3, pp. 350–357, 2017, doi: https://doi.org/10.1111/psyp.12799.
[5] M. Nabian, Y. Yin, J. Wormwood, K. S. Quigley, L. F. Barrett, and S. Ostadabbas, “An Open-Source Feature Extraction Tool for the Analysis of Peripheral Physiological Data,” IEEE J. Transl. Eng. Health Med., vol. 6, p. 2800711, 2018, doi: 10.1109/JTEHM.2018.2878000.