A multiclass arrhythmia classification method based on photoplethysmography signals of smartwatches (MGERB)

MGERB: multiscale group convolution + efficient channel attention blocks inside a ResNet (spatial features) fused with a BiGRU (temporal features) to classify six arrhythmia types from smartwatch PPG. Overall accuracy 91.11% on an unnamed public dataset and 95.57% on a self-collected dataset. Dataset identity, class list and split protocol not in the abstract.

• Form factor: wrist

Currently, cardiovascular diseases remain one of the major challenges facing humanity globally. Various types of arrhythmias caused by cardiovascular disease can affect the effective heart contraction and blood ejection, leading to serious complications such as thrombosis. Therefore, convenient detection and rapid classification of arrhythmias is very necessary and has a good application prospect. In this article, we investigate a classification method of multiclass arrhythmia based on photoplethysmography (PPG) signals of smartwatches. The arrhythmia classification method, utilizing the proposed multiscale group convolution and efficient channel attention enhanced residual network and bidirectional gated recurrent unit (MGERB) model, can identify six different types of arrhythmias. By combining the multiscale group convolution and efficient channel attention (MGE) mechanisms, we design a new MGE block to improve the residual blocks of the residual network (ResNet). The spatial features of PPG signals are extracted by stacking the MGE blocks. Subsequently, the feature vectors are input into the bidirectional gated recurrent unit (BiGRU) to extract the temporal features of PPG signals. Finally, the spatial features from the MGE blocks and the temporal features from the BiGRU are fused to classify the types of arrhythmias. Experiments show that the proposed method achieved an overall accuracy of 91.11% on the public dataset and 95.57% on the self-collected dataset, which verifies the accuracy and effectiveness of the proposed method.

Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: https://creativecommons.org/licenses/by/4.0

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.