Publication detail

Deep-Learning Premature Contraction Localization in 12-lead ECGfrom Whole Signal Annotations

NOVOTNÁ, P. VIČAR, T. RONZHINA, M. HEJČ, J. KOLÁŘOVÁ, J.

Original Title

Deep-Learning Premature Contraction Localization in 12-lead ECGfrom Whole Signal Annotations

English Title

Deep-Learning Premature Contraction Localization in 12-lead ECGfrom Whole Signal Annotations

Type

conference paper

Language

en

Original Abstract

Since common electrocardiography (ECG) diagnosticsapproaches are time-consuming and arrhythmia-type sen-sitive, deep-learning methods are state-of-the-art for theirdetection accuracy. However, premature ventricular con-tractions’ (PVC) localization via common deep-learningapproaches requires large training set, therefore MultipleInstance Learning (MIL) framework was applied, wheremodel is trained from whole-signal annotations. ProposedMIL framework is based on 1D Convolutional Neural Net-work (CNN), with global max-pooling in the last layer. Thedetection of PVCs’ positions was done by the peak detectorwith specified parameters – threshold, minimal distanceand peak prominence. Our method was tested on databasecontaining 1590 ECGs, including 672 signals with PVCs.Dice coefficient reaches 0.947. This simple deep-learningmethod for the localization of PVC achieves a promisingperformance while being trainable from the whole-signalannotations instead of positional labels.

English abstract

Since common electrocardiography (ECG) diagnosticsapproaches are time-consuming and arrhythmia-type sen-sitive, deep-learning methods are state-of-the-art for theirdetection accuracy. However, premature ventricular con-tractions’ (PVC) localization via common deep-learningapproaches requires large training set, therefore MultipleInstance Learning (MIL) framework was applied, wheremodel is trained from whole-signal annotations. ProposedMIL framework is based on 1D Convolutional Neural Net-work (CNN), with global max-pooling in the last layer. Thedetection of PVCs’ positions was done by the peak detectorwith specified parameters – threshold, minimal distanceand peak prominence. Our method was tested on databasecontaining 1590 ECGs, including 672 signals with PVCs.Dice coefficient reaches 0.947. This simple deep-learningmethod for the localization of PVC achieves a promisingperformance while being trainable from the whole-signalannotations instead of positional labels.

Keywords

ECG, electrocardiogram, arrhythmia, localization, global, annotation, PVC, premature ventricular contractions

Released

30.09.2020

Publisher

Computing in Cardiology 2020

Location

Rimini, Italy

ISBN

0276-6574

Periodical

Computers in Cardiology

State

US

Pages from

1

Pages to

4

Pages count

4

Documents

BibTex


@inproceedings{BUT165491,
  author="Petra {Novotná} and Tomáš {Vičar} and Marina {Ronzhina} and Jakub {Hejč} and Jana {Kolářová}",
  title="Deep-Learning Premature Contraction Localization in 12-lead ECGfrom Whole Signal Annotations",
  annote="Since common electrocardiography (ECG) diagnosticsapproaches are time-consuming and arrhythmia-type sen-sitive, deep-learning methods are state-of-the-art for theirdetection accuracy.  However, premature ventricular con-tractions’  (PVC)  localization  via  common  deep-learningapproaches requires large training set, therefore MultipleInstance  Learning  (MIL)  framework  was  applied,  wheremodel is trained from whole-signal annotations. ProposedMIL framework is based on 1D Convolutional Neural Net-work (CNN), with global max-pooling in the last layer. Thedetection of PVCs’ positions was done by the peak detectorwith  specified  parameters  –  threshold,  minimal  distanceand peak prominence. Our method was tested on databasecontaining 1590 ECGs, including 672 signals with PVCs.Dice coefficient reaches 0.947.  This simple deep-learningmethod for the localization of PVC achieves a promisingperformance while being trainable from the whole-signalannotations instead of positional labels.",
  address="Computing in Cardiology 2020",
  booktitle="Computing in Cardiology 2020",
  chapter="165491",
  edition="47",
  howpublished="online",
  institution="Computing in Cardiology 2020",
  year="2020",
  month="september",
  pages="1--4",
  publisher="Computing in Cardiology 2020",
  type="conference paper"
}