A lightweight deep learning approach for detecting electrocardiographic lead misplacement

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Electrocardiogram (ECG) lead misplacement can distort waveforms, compromising accurate interpretation.
• While infrequent (0.4%-4%), the high volume of ECGs necessitates robust detection methods.
• Existing methods may lack efficiency or accuracy in identifying lead misplacements.

Purpose of the Study:

• To develop and validate novel deep learning models for detecting both limb and chest ECG lead misplacements.
• To provide an effective and reliable algorithm for automated lead misplacement detection in clinical settings.
• To offer an open-source solution for the research and clinical community.

Main Methods:

• Developed two lightweight deep learning models: one for limb lead misplacement (using limb leads and V6) and one for chest lead misplacement (using six chest leads).
• Trained and validated models on the Chapman database (8:2 split), evaluated on PTB-XL, PTB, and LUDB databases.
• Simulated limb lead misplacements with mathematical transformations and chest lead misplacements by interchanging leads; assessed performance using accuracy, precision, sensitivity, specificity, and Macro F1-score.

Main Results:

• The models demonstrated high effectiveness in detecting simulated limb and chest lead misplacements.
• Achieved Macro F1-scores ranging from 93.42% to 99.61% across heterogeneous test databases.
• The models accurately detected lead misplacements across various simulated scenarios and arrhythmias.

Conclusions:

• The proposed deep learning models offer a reliable and accurate solution for detecting ECG lead misplacements.
• The study provides a significant contribution by offering an open-source algorithm for improved ECG analysis.
• The developed algorithm has the potential to enhance the diagnostic accuracy of ECG interpretation in clinical practice.