Automated differentiation of wide QRS complex tachycardia using QRS complex polarity

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Differentiating wide QRS complex tachycardia (WCT) into ventricular tachycardia (VT) and supraventricular wide complex tachycardia (SWCT) is clinically challenging.
• Existing 12-lead electrocardiogram (ECG) criteria and algorithms have limitations in accuracy.
• Computerized ECG interpretation (CEI) with engineered features offers a path to improve diagnostic accuracy.

Purpose of the Study:

• To develop and validate automated algorithms for WCT differentiation.
• To utilize novel engineered features: WCT QRS polarity Code (WCT-PC) and QRS Polarity Shift (QRS-PS).
• To compare the performance of machine learning models using these engineered features.

Main Methods:

• A three-part study employing machine learning (ML) models including logistic regression, artificial neural network, Random Forests, support vector machine, and ensemble learning.
• Engineered features (WCT-PC, QRS-PS) and established WCT differentiation features were used.
• Models were trained and validated using WCT ECG measurements alone, paired WCT and baseline ECG features, and combined features.

Main Results:

• Models using WCT ECG features alone achieved Area Under the Curve (AUC) of 0.86-0.88.
• Pairing WCT and baseline ECG features improved accuracy, with AUCs ranging from 0.90-0.93.
• Random Forests and Support Vector Machine models demonstrated the best performance in Part 3, with AUCs up to 0.93.

Conclusions:

• Engineered parameters focusing on QRS polarity direction and shifts are effective for WCT differentiation.
• These novel features show promise for enhancing automated CEI algorithms.
• The proposed automated algorithms represent a practical advancement in WCT diagnosis.