The EfficientNet B3 + BiLSTM + RanA framework achieved an accuracy of 99.48% in classifying atrial fibrillation versus normal sinus rhythm and 99.32% for congestive heart failure.
Does the EfficientNet B3 + BiLSTM + RanA framework improve the accuracy of cardiac arrhythmia detection from ECG signals compared to other methods?
A novel deep learning framework combining EfficientNet B3, BiLSTM, and RanA optimization demonstrates high accuracy (>99%) in detecting atrial fibrillation and congestive heart failure from ECG signals.
Cardiovascular diseases (CVDs) are leading causes of mortality all over the world, and arrhythmias are a leading cause of cardiac mortality. Arrhythmia detection in a timely and accurate manner is essential for medical intervention and plays a very important role in modern healthcare. In this work, the proposed automated arrhythmia identification process involves data collection from the PhysioNet repository such as AF, CHF and NSR-related ECG signals. Arrhythmia features are extracted through a CapsNet by capturing spatial hierarchies and relations in the ECG signals in an efficient manner. Features are filtered and optimized through feature selection by Deep CNNs such as EfficientNet B3, ResNet152, DenseNet201, and VGG19 to consider only the most significant features for the purpose of classification. These features are classified through BiLSTM by employing its sequential learning capability to learn temporal dependencies in the ECG signal and enhance the accuracy of the classifier. RanA hyperparameter optimization is also employed to optimize the model parameters further to achieve improved performance. The purpose of this work is to classify different types of ECG signals by formulating classification problems in which the abnormal heart arrhythmias are compared with NSR. Especially, the classification problems involve the discrimination of AF vs. NSR and CHF vs. NSR. The findings reveal EfficientNet B3 + BiLSTM + RanA performing outstandingly in AF vs. NSR and CHF vs. NSR classification with the use of multiple deep learning techniques in an optimal architecture. The model achieves an impressive accuracy of 99.48% in AF vs. NSR and 99.32% in CHF vs. NSR and outperforms other methods in the literature. With computational efficiency and accuracy, the proposed approach provides an optimal and reliable solution for real time cardiac disease diagnosis.
Manivannan et al. (Tue,) conducted a other in Cardiac Arrhythmia (Atrial Fibrillation and Congestive Heart Failure) (n=56). EfficientNet B3 + BiLSTM + RanA framework vs. Other deep learning methods and unoptimized models was evaluated on Classification accuracy of Atrial Fibrillation vs. Normal Sinus Rhythm. The EfficientNet B3 + BiLSTM + RanA framework achieved an accuracy of 99.48% in classifying atrial fibrillation versus normal sinus rhythm and 99.32% for congestive heart failure.