Classification of 12 Lead ECG Signal Using 1D-Convolutional Neural Network With Class Dependent Threshold

Rohit Pardasani, Navchetan Awasthi

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

5 Citaten (Scopus)


The goal of the proposed work is to classify the ECG signals into 24 different classes using the data obtained from the 12 Lead ECG signal. As part of the PhysioNetlComputing in Cardiology Challenge 2020, an approach based on 1 Dimensional - Convolutional Neural Network (1D-CNN) with class dependent threshold was developed for identifying cardiac abnormalities from 12-lead electrocardiogram (ECG). The method uses 1D-CNNs stacked in parallel with each CNN tuned to identify one of the classes. Each of these CNNs have same architecture comprising of convolutional layers, batch normalizations, activation layers and a dense layer with added regularizations and dropouts. The class dependent threshold gives the benefit of optimizing the model for each of the class individually without the need of training separate models for each category. This property of the network makes it ideal for real time setup where one inference run of this model is sufficient for multi-label and multi-class classification. The class dependent thresholds were chosen based on the ROC curve for each of the class respectively. Our approach achieved a challenge validation score of 0.342, and full test score of 0.077, placing our team (AI Strollers) 32 out of 41 in the official ranking.
Originele taal-2Engels
Titel2020 Computing in Cardiology, CinC 2020
Aantal pagina's4
ISBN van elektronische versie9781728173825
ISBN van geprinte versie978-1-7281-1105-6
StatusGepubliceerd - 13 sep. 2020
Evenement47th Computing in Cardiology, CinC 2020 - Rimini, Italië
Duur: 13 sep. 202016 sep. 2020
Congresnummer: 47


Congres47th Computing in Cardiology, CinC 2020
Verkorte titelCinC


Duik in de onderzoeksthema's van 'Classification of 12 Lead ECG Signal Using 1D-Convolutional Neural Network With Class Dependent Threshold'. Samen vormen ze een unieke vingerafdruk.

Citeer dit