End-to-end sensor fusion and classification of atrial fibrillation using deep neural networks and smartphone mechanocardiography
Executive Summary
This study introduces a deep convolutional-recurrent neural network (CRNN) architecture for detecting atrial fibrillation (AFib) using smartphone-based mechanocardiography (MCG), which combines seismocardiography (SCG) and gyrocardiography (GCG) signals. The model employs attention-based residual blocks, bidirectional LSTM layers, and sensor fusion techniques to classify AFib, sinus rhythm (SR), and noise with high accuracy, achieving a measurement-level macro-F1 score of 0.95. The findings demonstrate the feasibility of smartphone-based MCG as a cost-effective alternative to traditional ECG-based AFib detection methods, with potential implications for scalable cardiac monitoring.
Answer Machine Insights
Q: What is the accuracy of the model for AFib detection?
The model achieved a measurement-level macro-F1 score of 0.95 for AFib detection.
On the unseen test set, the model delivered average micro- and macro-F1-score of 0.95 (0.94–0.96; 95% CI) and 0.95 (0.94–0.96; 95% CI) for the measurement-wise classification, respectively.
Q: What is the significance of the noise class in the study?
The noise class helps improve the reliability of MCG signal analysis by identifying episodes where physiological properties cannot be extracted due to sensor placement failure or external disturbances.
Detecting the noisy episodes of a measurement is, therefore, a crucial step toward improving the reliability of the MCG signal analysis.
Key Results
Measurement-level macro-F1 score of 0.95 (95% CI: 0.94–0.96).
Segment-level macro-F1 score of 0.83 (95% CI: 0.83–0.84).
Clinical Snapshot
Evidence Rating
Relevance
high Priority