Using LSTM network to detect R-peaks from noisy ECG signals
Laitala, Juho (2020-08-17)
Using LSTM network to detect R-peaks from noisy ECG signals
(17.08.2020)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2020091669899
https://urn.fi/URN:NBN:fi-fe2020091669899
Tiivistelmä
Electrocardiogram (ECG) is one of the most important signals that can be measured from the human body. It contains lots of important information from the function of the heart, which can be utilized e.g. in medical diagnosis. Also, one of the vital signs, the heart rate can be derived from the ECG. Because of these reasons, ECG is extensively used by researchers and medical professionals. However, usage of ECG is not only limited into medical field as it is often collected e.g. during sport activity to track the heart rate. Utilization of the ECG is becoming even more widespread, today some of the newest smart watches have capability to measure it.
Detection of QRS complexes or R-peaks from the ECG signal is a prerequisite for heart rate calculation. Over time, numerous rule-based algorithms have been proposed for the task. Many of them work well when ECG signal has good quality, but their performance can drop in the presence of noise. Recently Laitala et al. [2] proposed robust R-peak detection algorithm that is based on Long Short-Term Memory (LSTM) network. The work of Laitala et al. is extended in this thesis. More detailed description is given from the LSTM based R-peak detection algorithm. It is also evaluated with new additional dataset and more QRS detection algorithms are used as reference.
Results are in line with the original work, LSTM based detector has the best general performance from the all of the evaluated detectors. However, results are not so striking as before. One reference detector bested the LSTM based detector with the new dataset. The major strength of the LSTM based detector is its robustness to noise. Another strong point is its ability to do sample precise R-peak detection. Majority of the reference detectors lack this ability and they often induce a lag to their predictions.
Detection of QRS complexes or R-peaks from the ECG signal is a prerequisite for heart rate calculation. Over time, numerous rule-based algorithms have been proposed for the task. Many of them work well when ECG signal has good quality, but their performance can drop in the presence of noise. Recently Laitala et al. [2] proposed robust R-peak detection algorithm that is based on Long Short-Term Memory (LSTM) network. The work of Laitala et al. is extended in this thesis. More detailed description is given from the LSTM based R-peak detection algorithm. It is also evaluated with new additional dataset and more QRS detection algorithms are used as reference.
Results are in line with the original work, LSTM based detector has the best general performance from the all of the evaluated detectors. However, results are not so striking as before. One reference detector bested the LSTM based detector with the new dataset. The major strength of the LSTM based detector is its robustness to noise. Another strong point is its ability to do sample precise R-peak detection. Majority of the reference detectors lack this ability and they often induce a lag to their predictions.