Arrhythmias & electrophysiology
Continuous Atrial Fibrillation Monitoring From Photoplethysmography: Comparison Between Supervised Deep Learning and Heuristic Signal Processing
Authors' conclusion
DNNs perform at least as well as SP models, classify more data, and thus may be better for continuous PPG monitoring.
Study context
- Population
- Free-living patients with 4 weeks of continuous PPG and ECG monitoring
- Participants
- 204 patients
- Comparison
- Convolutional deep neural network (DNN) vs signal processing (SP) heuristic for continuous atrial fibrillation (AF) detection from PPG
- Outcome
- AF detection performance (ROC AUC, sensitivity, specificity) and proportion of data classified
Methodological detailsexternal validation
- External validation
Performance
As reported in the abstract
AF detection, holdout validation
| Test or model | AUC | Sensitivity | Specificity | Data classified |
|---|---|---|---|---|
| SP | 0.972 | 99.6% | 94.4% | 62% |
| DNN | 0.973 | 92.2 | 95.5% | 95% |
AUC
SP0.972
DNN0.973
Sensitivity
SP99.6%
DNN92.2
Specificity
SP94.4%
DNN95.5%
Data classified
SP62%
DNN95%
Further findings
AF detection, external validation cohort (different population and PPG sensor)
AUC SP 0.989 · DNN 0.994
Data analysed
AF detection, external validation cohort (different population and PPG sensor)
- DNN performed better and analysed more signal
Original publication
Continuous Atrial Fibrillation Monitoring From Photoplethysmography: Comparison Between Supervised Deep Learning and Heuristic Signal Processing. JACC Clin Electrophysiol. Published 7 February 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →