Data driven feature selection and machine learning to detect misplaced V1 and V2 chest electrodes when recording the...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Electrocardiogram (ECG) lead misplacement, particularly V1 and V2 leads, can lead to incorrect diagnoses and clinical decisions.
• Common misplacement involves positioning leads superior to their correct anatomical locations.

Purpose of the Study:

• To develop and evaluate machine learning models for detecting V1 and V2 lead misplacement.
• To enhance the quality and reliability of ECG data through automated lead placement verification.

Main Methods:

• Utilized ECG data from 453 patients with normal, Left Ventricular Hypertrophy (LVH), and Myocardial Infarction conditions.
• Extracted 16 features across time, statistical, and time-frequency domains using wavelet transform.
• Applied hybrid feature selection and compared five classifiers for V1 and V2 misplacement detection in three intercostal space (ICS) positions.

Main Results:

• Achieved high accuracy for V1 misplacement detection: 93.9% (1st ICS), 89.3% (2nd ICS), and 72.8% (3rd ICS).
• Achieved high accuracy for V2 misplacement detection: 93.6% (1st ICS), 86.6% (2nd ICS), and 68.1% (3rd ICS).
• Observed a decline in detection accuracy for the third intercostal space, as anticipated.

Conclusions:

• Machine learning effectively detects V1 and V2 lead misplacement.
• The developed methods show promise for improving ECG data quality and diagnostic accuracy.
• Accurate lead placement is critical for reliable ECG interpretation.