FADE: Forecasting for anomaly detection on ECG.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiology

Background:

• Cardiovascular diseases are a leading cause of death, necessitating early detection.
• Current ECG anomaly detection relies on time-consuming manual interpretation.
• Advances in machine learning offer new avenues for ECG analysis.

Purpose of the Study:

• To propose FADE, a deep learning system for normal ECG forecasting and anomaly detection.
• To reduce reliance on extensive labeled datasets and manual ECG interpretation.
• To enhance the accuracy and efficiency of detecting cardiac anomalies.

Main Methods:

• Developed FADE, a deep learning system trained in a self-supervised manner.
• Employed a novel morphological-inspired loss function for ECG forecasting.
• Utilized a unique distance function for comparing forecasted and actual ECG data to identify anomalies.
• Incorporated domain adaptation techniques for contextual flexibility.

Main Results:

• Achieved 83.84% average accuracy in anomaly detection.
• Attained 85.46% accuracy in classifying normal ECG signals.
• Demonstrated superior performance in detecting a wider range of cardiac anomalies compared to previous methods.

Conclusions:

• FADE offers superior performance in early cardiac anomaly detection.
• The system effectively identifies abnormal heartbeats and arrhythmias.
• FADE presents advantages in cost reduction, remote monitoring, and large-scale ECG data processing.