A new hierarchical method for inter-patient heartbeat classification using random projections and RR intervals

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Automated heartbeat classification systems rely on inter-patient schema and AAMI standards.
• Existing methods often use uniform features and classifiers, leading to suboptimal performance for ventricular ectopic beats (VEB) and supraventricular ectopic beats (SVEB).

Purpose of the Study:

• To develop a novel hierarchical heartbeat classification system to enhance the detection accuracy of VEB and SVEB.
• To leverage distinct features and classification methods tailored to the characteristics of VEB and SVEB.

Main Methods:

• A hierarchical approach was employed, utilizing random projection and support vector machine (SVM) ensemble for VEB detection.
• Supraventricular ectopic beat (SVEB) detection involved comparing the RR interval ratio to a threshold.
• Optimal model parameters were determined on a training set and validated on an independent testing set.

Main Results:

• The hierarchical system demonstrated superior performance compared to existing methods.
• High sensitivity and positive predictive values were achieved for VEB (93.9% sensitivity, 90.9% PPV) and SVEB (91.1% sensitivity, 42.2% PPV).
• The classification process was found to be relatively fast.

Conclusions:

• The proposed hierarchical system effectively classifies VEB and SVEB using inter-patient data division.
• This approach offers improved classification performance over current methodologies.
• The system shows promise for clinical application in identifying VEB and SVEB in new patients.