.png){kind=icon}
Return
A sequential MAE-clustering self-supervised learning method for arrhythmia detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.126379.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In supervised learning, obtaining a substantial quantity of accurately labeled electrocardiogram (ECG) data poses significant challenges. In contrast, self-supervised learning, particularly the masked autoencoders (MAE), shows promise. This study proposes an enhanced MAE approach the sequential MAE-clustering (SMC) self-supervised learning method for analyzing various arrhythmias. Our method comprises a sequential MAE (SM) network, which utilizes sequential masking patches, and a clustering learning component consisting of a symbiotic network and Gaussian clustering. The SMC loss function, a combination of the mean squared error loss from the SM network and the ProtoNCE loss from clustering learning, was developed. The process begins by converting 1D ECG heartbeats into 2-D diagrams using the Gramian angular field technique. These diagrams are subsequently processed using the SMC self-supervised learning method to classify heartbeat categories as per the ANSI/AAMI standard namely, N, V, S, Q, and F of the MIT/BIH arrhythmia database. The SMC method demonstrated effective performance, achieving an average accuracy (acc) of 97.84 %, a positive predictive value (ppv) of 84.25 %, a sensitivity (sen) of 65.51 %, and a specificity (spe) of 94.06 % across the five heartbeat categories of the MIT/BIH arrhythmia database.
Keywords:
Self-supervised learning
Clustering
Sequential MAE
Arrhythmia detection
Journal
IF:
7.5
Papers:
3.0W
Citations:
10.2W
Organization
No organization information available
Cited Papers
ShareSaveShareSave
ShareSaveShareSaveShareSaveShareSaveA Tiny Matched Filter-Based CNN for Inter-Patient ECG Classification and Arrhythmia Detection at the Edge.png)
SENSORS
OAAI
SENSORSIF3.5
OAAIShareSaveShareSave
View more