Ventricular Fibrillation and Tachycardia detection from surface ECG using time-frequency representation images as...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Distinguishing Ventricular Fibrillation (VF) from Ventricular Tachycardia (VT) and other arrhythmias is critical for appropriate patient therapy.
• Incorrect arrhythmia classification can lead to severe patient harm or induce VF.
• Traditional methods involve feature extraction, which can result in information loss.

Purpose of the Study:

• To introduce a novel method for arrhythmia classification using time-frequency (t-f) representation images as direct input.
• To improve the accuracy and safety of VF and VT detection.
• To eliminate the need for manual feature selection and extraction stages.

Main Methods:

• Utilized standard AHA and MIT-BIH databases for evaluation.
• Applied basic preprocessing including denoising and signal alignment.
• Calculated time-frequency Pseudo Wigner-Ville (PWV) representations.
• Tested four distinct classifiers: L2 Regularized Logistic Regression (L2 RLR), Adaptive Neural Network Classifier (ANNC), Support Vector Machine (SSVM), and Bagging classifier (BAGG).

Main Results:

• Achieved 95.56% sensitivity and 98.8% specificity for Ventricular Fibrillation (VF) detection.
• Obtained 88.80% sensitivity and 99.5% specificity for Ventricular Tachycardia (VT) detection.
• Demonstrated high performance for normal sinus rhythm (98.98% sensitivity, 97.7% specificity) and other rhythms (96.87% sensitivity, 99.55% specificity).

Conclusions:

• Time-frequency (t-f) data representations as direct classifier input yield superior performance compared to traditional feature selection methods.
• This approach enhances the accuracy of cardiac arrhythmia detection.
• The method shows potential for application in various other diagnostic detection systems.