Multiclass arrhythmia classification using multimodal smartwatch photoplethysmography signals collected in real-life settings

2024

Smartwatch-Based Detection of Heart Rhythm Problems

Sample size: 106 publication 10 minutes Evidence: high

Author Information

Author(s): Han Dong, Moon Jihye, Díaz Luís Roberto Mercado, Chen Darren, Williams Devan, Mohagheghian Fahimeh, Ghetia Om, Peitzsch Andrew G., Kong Youngsun, Nishita Nishat, Ghutadaria Ohm, Orwig Taylor A., Otabil Edith Mensah, Noorishirazi Kamran, Hamel Alexander, Dickson Emily L., DiMezza Danielle, Lessard Darleen, Wang Ziyue, Mehawej Jordy, Filippaios Andreas, Naeem Syed, Gottbrecht Matthew F., Fitzgibbons Timothy P., Saczynski Jane S., Barton Bruce, Ding Eric Y., Tran Khanh-Van, McManus David D., Chon Ki H.

Primary Institution: University of Connecticut

Hypothesis

Can a multimodal smartwatch system accurately classify different types of arrhythmias in real-life settings?

Conclusion

The study demonstrated that a multimodal deep learning model can effectively classify normal sinus rhythm, atrial fibrillation, and premature contractions with high accuracy and sensitivity.

Supporting Evidence

The model achieved 97.31% accuracy for AF detection.
The proposed model showed 83.52% sensitivity for PAC/PVC detection.
Results were validated on two independent datasets, demonstrating generalizability.
The model was 14 times more efficient and 2.7 times faster than the best state-of-the-art model.
High macro-averaged AUROC values of 96.22% and 94.17% were achieved on independent datasets.

Takeaway

Researchers used smartwatches to monitor heart rhythms and found a way to tell if someone has a heart problem, like atrial fibrillation, using special computer techniques.

Methodology

The study used a 1D bi-directional Gated Recurrent Unit deep learning model to analyze smartwatch PPG, accelerometer, and heart rate data collected over two weeks.