Key result
A deep learning model using raw PPG data discriminated AF from sinus rhythm with an AUC of 0.983 (sensitivity 0.985, specificity 0.880), outperforming models using heart rate alone.
Why the study?
Smartphones and smartwatches detect AF using PPG heart rate patterns, but enhanced accuracy is needed to reduce false positives in screening populations.
Does a deep learning algorithm using raw PPG waveforms improve the detection of atrial fibrillation compared to algorithms using heart rate alone in patients presenting for cardioversion?
RCT (n=51)
randomly assigned
Does a deep learning algorithm using raw PPG waveforms improve the detection of atrial fibrillation compared to algorithms using heart rate alone in patients presenting for cardioversion?
Deep learning applied to raw PPG waveforms significantly improves the accuracy of smartwatch-based atrial fibrillation detection compared to heart rate-based algorithms.
No takes yet. Share an insight, caveat, or question.
Raw PPG DL models may enhance wearable AF detection; leaves open validation in ambulatory populations before clinical use.
Aschbacher et al. (2020) conducted an RCT in Atrial fibrillation (n=51). Deep convolutional-recurrent neural net (DNN) using raw PPG data vs. Logistic regression of heart rate variability and LSTM model using heart rate data was evaluated on Discrimination of atrial fibrillation from sinus rhythm. A deep learning model using raw PPG data discriminated AF from sinus rhythm with an AUC of 0.983 (sensitivity 0.985, specificity 0.880), outperforming models using heart rate alone.