Multiple Instance Learning Framework Used For ECG Premature Contraction Localization

Multiple Instance Learning Framework Used For ECG Premature Contraction Localization

Paper in proceedings outside WoS and Scopus

We propose the model combining convolutional neural network with multiple instance learning in order to localize the premature atrial contraction and premature ventricular contraction. The model is based on ResNet architecture modified for 1D signal processing. Model was trained on China Physiological Signal Challenge 2018 database extended by manually labeled ground truth positions of premature complexes. The presented method did not reach satisfying results in PAC localization (with dice = 0.127 for avg-pooling implementation). On the other hand, results of localization of PVCs were comparable with other published studies (with dice = 0.952 for avg-pooling implementation).

We propose the model combining convolutional neural network with multiple instance learning in order to localize the premature atrial contraction and premature ventricular contraction. The model is based on ResNet architecture modified for 1D signal processing. Model was trained on China Physiological Signal Challenge 2018 database extended by manually labeled ground truth positions of premature complexes. The presented method did not reach satisfying results in PAC localization (with dice = 0.127 for avg-pooling implementation). On the other hand, results of localization of PVCs were comparable with other published studies (with dice = 0.952 for avg-pooling implementation).

EEICT, ECG, PAC, PVC, CNN, MIL, arrhytmia, localization

EEICT, ECG, PAC, PVC, CNN, MIL, arrhytmia, localization

NOVOTNÁ, P.

2021

23.04.2020

Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií

Brno

978-80-214-5942-7

Proceedings I of the 27th Conference STUDENT EEICT 2021

1

1

5

5