Artificial intelligence for detection of ventricular oversensing: Machine learning approaches for noise detection within nonsustained ventricular tachycardia episodes remotely transmitted by pacemakers and implantable cardioverter-defibrillators

Strik, M; Sacristan, B; Bordachar, P; Duchateau, J; Eschalier, R; Mondoly, P; Laborderie, J; Gassa, N; Zemzemi, N; Laborde, M; Garrido, J; Matencio Perabla, C; Jimenez-Perez, G; Camara, O; Haïssaguerre, M; Dubois, R; Ploux, S

doi:10.1016/j.hrthm.2023.06.019

Arrhythmias & electrophysiology

Artificial intelligence for detection of ventricular oversensing: Machine learning approaches for noise detection within nonsustained ventricular tachycardia episodes remotely transmitted by pacemakers and implantable cardioverter-defibrillators

Heart RhythmPublished 3 Jul 2023Diagnostic accuracy study

Read the original article ↗PubMed

Cite

Authors' conclusion

Machine learning algorithms were accurate in detecting nonphysiological signals within electrograms transmitted by PMs and ICDs.

Study context

Population: Pacemakers (PMs) or implantable cardioverter-defibrillators (ICDs) (Boston Scientific) from 4 French hospitals with ≥1 remotely transmitted nonsustained ventricular tachycardia (NSVT) episode
Participants: 807 devices

Comparison: 4 machine learning algorithms vs remote monitoring team analysis
Outcome: Classification of NSVT episodes as physiological or nonphysiological, by F2 score (favoring sensitivity over PPV)

Methodological detailsmulticentre, independent test set

Multicentre
Independent test set

Performance

As reported in the abstract

F2 score for nonphysiological signals

Test or model	F2 score	Sensitivity	Specificity
Remote monitoring team	0.932	95%	99%
2DTF-CNN	0.914
2D-DenseNet	0.906
2DTF-VGG	0.863
1D-AgResNet	0.791

F2 score

Remote monitoring team0.932

2DTF-CNN0.914

2D-DenseNet0.906

2DTF-VGG0.863

1D-AgResNet0.791

Sensitivity

Remote monitoring team95%

2DTF-CNNNot reported in the abstract

2D-DenseNetNot reported in the abstract

2DTF-VGGNot reported in the abstract

1D-AgResNetNot reported in the abstract

Specificity

Remote monitoring team99%

2DTF-CNNNot reported in the abstract

2D-DenseNetNot reported in the abstract

2DTF-VGGNot reported in the abstract

1D-AgResNetNot reported in the abstract

Further findings

Devices transmitting nonphysiological signals

Devices with nonphysiological NSVT recordings: 87 (10.8%)
NSVT recordings with nonphysiological signals: 544

Reference

Authors: Strik M, Sacristan B, Bordachar P, Duchateau J, Eschalier R, Mondoly P, et al.
All 17 authors
Strik M, Sacristan B, Bordachar P, Duchateau J, Eschalier R, Mondoly P, Laborderie J, Gassa N, Zemzemi N, Laborde M, Garrido J, Matencio Perabla C, Jimenez-Perez G, Camara O, Haïssaguerre M, Dubois R, Ploux S
Journal: Heart Rhythm
Published: 3 Jul 2023
DOI: 10.1016/j.hrthm.2023.06.019
PubMed: 37406873

Topic: Arrhythmias & electrophysiology

Diagnostic definitions

Remote monitoring team: analysis by the remote monitoring team of the Bordeaux University Hospital
2DTF-CNN: machine learning algorithm
2D-DenseNet: machine learning algorithm
2DTF-VGG: machine learning algorithm
1D-AgResNet: machine learning algorithm

Key notes

Only 1D-AgResNet had significantly different labeling from that of the remote monitoring team

Original publication

Artificial intelligence for detection of ventricular oversensing: Machine learning approaches for noise detection within nonsustained ventricular tachycardia episodes remotely transmitted by pacemakers and implantable cardioverter-defibrillators. Heart Rhythm. Published 3 July 2023. doi:10.1016/j.hrthm.2023.06.019

Read the original article ↗

Summarised from the published abstract; every number is checked against it