Interpretable morphological features for efficient single-lead automatic ventricular ectopy detection.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Ventricular ectopic beats (VEBs) are common arrhythmias.
• Accurate detection of VEBs is crucial for assessing cardiovascular health.
• Existing methods for VEB detection may lack efficiency or interpretability.

Purpose of the Study:

• To develop an automated, efficient, and interpretable algorithm for detecting VEBs.
• To utilize novel cardiac cycle features for improved detection accuracy.
• To enable large-scale analysis of VEBs and their association with patient outcomes.

Main Methods:

• Engineered five interpretable features per cardiac cycle, including a novel morphological feature.
• Applied unsupervised, subject-specific normalization.
• Trained an ensemble binary classifier using the AdaBoost algorithm on the MIT-BIH Arrhythmia database.

Main Results:

• Achieved an F1 score of 94.35% on an unseen subset of the MIT-BIH database.
• Attained F1 scores of 92.06% and 91.40% on the INCART and MIT-BIH Long-term databases, respectively.
• Demonstrated high precision on previously unseen subjects and databases.

Conclusions:

• The developed algorithm effectively detects ventricular ectopy with high precision.
• The novel features and methodology contribute to an interpretable and efficient VEB detection system.
• This detector is poised to advance research on the relationship between premature ventricular contractions and adverse patient outcomes.