Clinical Assessment of a Lightweight CNN Model for Real-Time Atrial Fibrillation Prediction in Continuous Wearable...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Atrial Fibrillation (AFib) is a common cardiac arrhythmia with significant health risks.
• Wearable devices and predictive models enable non-invasive, real-time health monitoring.
• Early detection of AFib is crucial for managing patient outcomes.

Purpose of the Study:

• To assess a lightweight Convolutional Neural Network (CNN) model for predicting Atrial Fibrillation (AFib).
• To enable continuous, non-invasive AFib monitoring using wearable devices.
• To evaluate the model's performance and clinical relevance for early arrhythmia detection.

Main Methods:

• Developed and evaluated a lightweight CNN model for AFib prediction.
• Utilized data from wearable devices for continuous monitoring.
• Conducted a clinical assessment with 24-hour ECG monitoring on 56 subjects (28 with arrhythmia, 28 without).

Main Results:

• The model achieved high performance on the test set: F1 score of 0.96, AUC of 0.99, and Accuracy (Acc) of 0.96.
• In clinical assessment, the model achieved F1 score of 0.94, AUC of 0.99, and Acc of 0.97 compared to physician labels.
• Demonstrated robust outcomes in both initial testing and real-world clinical scenarios.

Conclusions:

• The lightweight CNN model shows significant potential for real-world AFib monitoring.
• The model offers a non-invasive and effective approach for early detection of arrhythmia.
• Integration with wearable devices facilitates continuous patient monitoring and risk assessment.