QT database from "Waveform Segmentation Using Deep Learning" tutorial

Hello,

I have recently started working in the field of cardiac waveform segmentation and I am using the "Waveform Segmentation Using Deep Learning" tutorial. I would like to know how the signals available at https://www.mathworks.com/supportfiles/SPT/data/QTDatabaseECGData.zip have been extracted, i.e. if they are an adaptation of Physionet or how did you come to obtain the .mat files. The documentation available at https://es.mathworks.com/help/signal/ug/waveform-segmentation-using-deep-learning.html# says:

"This example uses ECG signals from the publicly available QT Database [3] [4]. .... The database provides signal region labels generated by an automated expert system [2]. "

, but how the labeled data were actually obtained? Is this automatic labeling system available? I would like to be able to use this data as "gold standard" to compare the results I have obtained with the tutorial with other methods and I am trying to study the QT database labeling.

Thank you in advance,

Carmen

Answer 1

[Prasanna]

Hi Carmen,

It is my understanding that you want more information regarding the extraction and labelling process of ECG signals used in the "Waveform Segmentation Using Deep Learning" tutorial, which utilizes data from the QT Database.

The ECG recordings in the QT Database are publicly available and comprise approximately 15 minutes of two-channel ECG signals from 105 patients, sampled at a rate of 250 Hz. These recordings were acquired by placing two electrodes at specific locations on the patients' chests. The database includes signal region labels, which were generated by an automated expert system.

The methodology for the automated expert system for labelling the data, as described in the associated research paper, involves several steps. Initially, a multilead QRS detector, designed for 15 leads, is applied to each ECG record. Following this, waveform boundaries are estimated independently for each lead, utilizing the differentiated ECG signal and waveform shape information. The detection algorithm applies criteria to determine wave presence and characterizes the patterns of the P wave, QRS complex, and T wave based on the classification by the CSE working party.

The algorithm selects the final wave boundaries from the leads that exhibit the longest detected electrical activity. This approach minimizes the impact of potential noise in the measurements. Subsequent procedures, including QRS detection, fibrillation rejection, and precise waveform boundary location, are then carried out to refine the results.

For a more comprehensive understanding and to explore the availability of the automated labelling system, I suggest reaching out directly to the authors of the research paper. They can provide detailed insights into the automatic labelling process and its implementation: https://www.researchgate.net/publication/15194187_Automatic_Detection_of_Wave_Boundaries_in_Multilead_ECG_Signals_Validation_with_the_CSE_Database

I hope this helps,

Regards,

Prasanna