Robust multichannel QRS detection

Robust multichannel QRS detection

Paper in proceedings (conference paper)

The task of the CinC Challenge 2014 was to develop an algorithm for robust detection of a QRS complex throughout different measurement data. Our proposed approach starts with filtering and detection of the QRS complex in channel with electric activity using standard deviation and defines preliminary QRS annotations. The averaged shape (length of 640 ms) for each channel is found by accumulating data close to preliminary annotations through the whole record (max. 10 minutes). The weight of each averaged shape is computed by average correlation of the shape with its original signal through all preliminary QRS annotations. The correlation of each averaged shape to the corresponding channel is multiplied by its weight and forms a new channel containing the total correlation response channels. The resulting QRS annotations are found by tresholding this total correlation response. The final version of our method leads to results of SE+ = 99.87 and P+ = 99.96 for the Challenge Training data set containing 100 records. The overall score was 73.85 at the end of stage I (using an early version of this method) and 83.73 at the end of stage 111.

The task of the CinC Challenge 2014 was to develop an algorithm for robust detection of a QRS complex throughout different measurement data. Our proposed approach starts with filtering and detection of the QRS complex in channel with electric activity using standard deviation and defines preliminary QRS annotations. The averaged shape (length of 640 ms) for each channel is found by accumulating data close to preliminary annotations through the whole record (max. 10 minutes). The weight of each averaged shape is computed by average correlation of the shape with its original signal through all preliminary QRS annotations. The correlation of each averaged shape to the corresponding channel is multiplied by its weight and forms a new channel containing the total correlation response channels. The resulting QRS annotations are found by tresholding this total correlation response. The final version of our method leads to results of SE+ = 99.87 and P+ = 99.96 for the Challenge Training data set containing 100 records. The overall score was 73.85 at the end of stage I (using an early version of this method) and 83.73 at the end of stage 111.

correlation methods;electrocardiography;filtering theory;medical signal detection;medical signal processing;CinC Challenge 2014;ECG;average correlation;electric activity;electrocardiography;filtering;preliminary QRS annotations;robust multichannel QRS detection;standard deviation;total correlation response channels;Abstracts;Annealing;Indium tin oxide;Shape;Time measurement

correlation methods;electrocardiography;filtering theory;medical signal detection;medical signal processing;CinC Challenge 2014;ECG;average correlation;electric activity;electrocardiography;filtering;preliminary QRS annotations;robust multichannel QRS detection;standard deviation;total correlation response channels;Abstracts;Annealing;Indium tin oxide;Shape;Time measurement

PLEŠINGER, F.; JURČO, J.; JURÁK, P.; HALÁMEK, J.

2017

07.09.2014

CCAL

Cambridge, Massachusetts, USA

978-1-4799-4346-3

Computing in Cardiology

2014

0276-6574

Computers in Cardiology

United States of America

557

560

4

http://ieeexplore.ieee.org/document/7043103/