Machine learning detection of Atrial Fibrillation using wearable technology.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Atrial Fibrillation (AF) is a prevalent arrhythmia globally, increasing stroke risk.
• Anticoagulation significantly reduces AF-related stroke and mortality.
• Early detection of AF is crucial for stroke prevention, with a need for accessible diagnostic tools.

Purpose of the Study:

• To develop and validate a novel machine learning algorithm for accurate Atrial Fibrillation detection.
• To assess the efficacy of using inexpensive consumer heart rate monitors for AF screening.
• To explore the potential of data compression techniques for efficient AF detection algorithms.

Main Methods:

• A Support Vector Machine (SVM) classifier was trained using de-correlated Lorenz plots of RR intervals, compressed via wavelet transformation (JPEG200).
• Training data utilized the MIT-Beth Israel Hospital (BIH) Atrial Fibrillation and Arrhythmia databases.
• Algorithm performance was validated using RR intervals from a consumer heart rate monitor (Polar-H7) in a case-control study.

Main Results:

• The SVM algorithm achieved high accuracy in training data: 99.2% sensitivity and 99.5% specificity for AF detection.
• Validation data showed excellent performance: 100% sensitivity and 97.6% specificity for identifying AF.
• The algorithm correctly identified all 79 AF cases in the validation set.

Conclusions:

• An inexpensive wearable heart rate monitor and machine learning algorithm can detect AF with high accuracy.
• This approach enables potential intermittent or continuous screening for paroxysmal AF.
• Further development could lead to cost-effective AF burden estimation and improved risk stratification.