BACKGROUND: Premature ventricular contractions (PVCs) are cardiac abnormalities that may occur in subjects with/without cardiovascular disorder. Detection is usually performed from electrocardiograms (ECGs); heart activity for a long period of time must be recorded at hospital or with ambulatory electrocardiography. An alternative with a common mobile device would be very interesting, because a simple heart rate sensor should be sufficient. OBJECTIVE: To develop an algorithm to detect PVCs using the RR-interval (distance between consecutive beats) extracted from ECGs or from the heart rate signal captured by mobile devices. METHODS: Feature extraction and classification techniques were included: 1) two timing interval features (prematurity and compensatory pause) were extracted. 2) A linear classifier was applied. To validate the method, the MIT-BIH Arrhythmia Database was used. Considering the existence of unbalanced classes (normal beats and PVCs) at different decision costs, validation was performed with receiver operating characteristic (ROC) analysis. RESULTS: A sensitivity of 90.13% and a specificity percentage of 82.52% were achieved. The area under the ROC curve (AUC) was 0.928. CONCLUSIONS: The method is advantageous since it only uses the RR-interval signal for PVC detection, and results compare well with more complex methods that use ECG recording.
BACKGROUND:Premature ventricular contractions (PVCs) are cardiac abnormalities that may occur in subjects with/without cardiovascular disorder. Detection is usually performed from electrocardiograms (ECGs); heart activity for a long period of time must be recorded at hospital or with ambulatory electrocardiography. An alternative with a common mobile device would be very interesting, because a simple heart rate sensor should be sufficient. OBJECTIVE: To develop an algorithm to detect PVCs using the RR-interval (distance between consecutive beats) extracted from ECGs or from the heart rate signal captured by mobile devices. METHODS: Feature extraction and classification techniques were included: 1) two timing interval features (prematurity and compensatory pause) were extracted. 2) A linear classifier was applied. To validate the method, the MIT-BIH Arrhythmia Database was used. Considering the existence of unbalanced classes (normal beats and PVCs) at different decision costs, validation was performed with receiver operating characteristic (ROC) analysis. RESULTS: A sensitivity of 90.13% and a specificity percentage of 82.52% were achieved. The area under the ROC curve (AUC) was 0.928. CONCLUSIONS: The method is advantageous since it only uses the RR-interval signal for PVC detection, and results compare well with more complex methods that use ECG recording.
Entities:
Keywords:
ECG; Heart rate; biomedical signal processing; linear classifiers; pattern recognition; premature ventricular contractions
Authors: Ka Hou Christien Li; Francesca Anne White; Gary Tse; Timothy Tipoe; Tong Liu; Martin Cs Wong; Aaron Jesuthasan; Adrian Baranchuk; Bryan P Yan Journal: JMIR Mhealth Uhealth Date: 2019-02-15 Impact factor: 4.773