Heartbeat classification using disease-specific feature selection.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Automatic heartbeat classification is crucial for diagnosing arrhythmias from Holter recordings.
• Existing methods may lack specificity in identifying various ectopic heartbeats.
• Effective feature selection is key to improving classification performance.

Purpose of the Study:

• To introduce a novel disease-specific feature selection method for enhanced heartbeat classification.
• To evaluate the proposed method's performance using the MIT-BIH arrhythmia database.
• To compare the method against traditional approaches and state-of-the-art techniques.

Main Methods:

• A one-versus-one (OvO) feature ranking and search strategy was employed.
• Support Vector Machine (SVM) binary classifiers were utilized within the OvO framework.
• Electrocardiogram (ECG) features including intervals and morphology were analyzed.
• Data was classified into four types: Normal (N), Supraventricular ectopic (S), Ventricular ectopic (V), and Fusion (F).

Main Results:

• The proposed feature selection method achieved an average classification accuracy of 86.66%.
• Performance surpassed methods without feature selection, demonstrating superior accuracy.
• High sensitivities were reported for Normal (88.94%) and Fusion (93.81%) classes.
• Geometric means of sensitivity and positive predictivity indicated better performance than other methods.

Conclusions:

• The novel OvO disease-specific feature selection method significantly improves automatic heartbeat classification.
• This technique offers a more effective approach for identifying ectopic heartbeats in clinical settings.
• The method shows promise for enhancing diagnostic capabilities in long-term cardiac monitoring.