Automated detection of shockable and non-shockable arrhythmia using novel wavelet-based ECG features

• Published in: Computers in Biology and Medicine Vol. 115; p. 103446
• Main Authors: Sharma, Manish, Singh, Swapnil, Kumar, Abhishek, San Tan, Ru, Acharya, U. Rajendra
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.12.2019; Elsevier BV; Elsevier Limited
• Subjects: Accuracy; Algorithms; Arrhythmia; Arrhythmias, Cardiac; Automation; Bradycardia; Cardiac arrhythmia; Classification; Decomposition; Defibrillators; Diagnostic systems; ECG; EKG; Electrocardiography; Emergency medical services; Feature extraction; Fibrillation; Filter banks; Filter design; Heart; Humans; Inspection; Localization; Machine learning; Mechanical properties; Methods; Optimization problem; Semi-definite program; Shockable; Signal classification; Signal Processing, Computer-Assisted; Support vector machines; Tachycardia; Therapy; Ventricle; Ventricular fibrillation; Wavelet analysis; Wavelets; Heart; Wavelets; Shockable; Machine learning; Semi-definite program; Classification; ECG; Filter design; Optimization problem
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2019.103446

Malignant arrhythmia can lead to sudden cardiac death (SCD). Shockable arrhythmia can be terminated with device electrical shock therapies. Ventricular-tachycardia (VT) and ventricular fibrillation (VF) are responsive to electrical anti-tachycardia pacing therapy and defibrillation which help to restore normal electrical and mechanical function of the heart. In contrast, non-shockable arrhythmia like asystole and bradycardia are not responsive to electric shock therapy. Distinguishing between shockable and non-shockable arrhythmia is an important diagnostic challenge that has practical clinical relevance. It is difficult to accurately differentiate between these two types of arrhythmia by manual inspection of electrocardiogram (ECG) segments within the short time duration before triggering the device for electrical therapy. Automated defibrillators are equipped with automatic shockable arrhythmia detection algorithms based on ECG morphological features, which may possess variable diagnostic performance depending on machine models. In our work, we have designed a robust system using wavelet decomposition filter banks for extraction of features from the ECG signal and then classifying the features. We believe this method will improve the accuracy of discriminating between shockable and non-shockable arrhythmia compared with existing conventional algorithms. We used a novel three channel orthogonal wavelet filter bank, which extracted features from ECG epochs of duration 2 s to distinguish between shockable and non-shockable arrhythmia. The fuzzy, Renyi and sample entropies are extracted from the various wavelet coefficients and fed to support vector machine (SVM) classifier for automated classification. We have obtained an accuracy of 98.9%, sensitivity and specificity of 99.08% and 97.11.9%, respectively, using 10-fold cross validation. The area under the receiver operating characteristic has been found to be 0.99 with F1-score of 0.994. The system developed is more accurate than the existing algorithms. Hence, the proposed system can be employed in automated defibrillators inside and outside hospitals for emergency revival of patients suffering from SCD. These automated defibrillators can also be implanted inside the human body for automatic detection of potentially fatal shockable arrhythmia and to deliver an appropriate electric shock to the heart. •A new optimal three-band wavelet-based system for identification of shockable and non-shockable arrhythmia.•High Classification performance.•The system can be employed in automated defibrillators.•The method outperforms existing state-of-the-art methods.