A model to enhance the atrial fibrillations’ risk detection using deep learning

Abstract

Atrial fibrillation (AF) is a complex arrhythmia linked to a variety of common cardiovascular illnesses and conventional cardiovascular risk factors. Although awareness and improved detection of AF have improved over the last decade as the incidence and prevalence of AF has increased, current trends in using machine learning approaches to diagnose AF are still lacking in precision. To determine the true nature of the Electrocardiography (ECG) signal segments, a Convolutional Neural Network (CNN) model was employed to discover hidden information. Fully Connected (FC) layers were then utilized to categorize the ECG data segments as normal or abnormal. The suggested algorithm's findings were compared to state-of-the-art arrhythmia identification algorithms in the literature for the MIT-BIH ECG database. The methodology proved not only to yield high classification performance (98.5%) but also low processing computational advantage where the CNN was the most accurate algorithm used for atrial fibrillation detection hence. To conclude the findings of the research, a model was prepared to test the accuracy of the most common ML algorithms used for AF detection. After comparing the results of the experiment, it was clear that CNN algorithm is the best approach compared to Support Vector Machine (SVM) and K-Nearest Neighbor (KNN).

Study of Arrhythmia Classification Algorithms on Electrocardiogram Using Deep Learning

“…Based on the comparison results from Table 6, the results of the LSTM deep learning algorithm in our study also managed to outperform the results of other machine learning algorithms. And for comparison with CNN algorithms from (Almazrouei & Al-Rajab, 2022), their research managed to outperform 0.03% of our research accuracy, while in specificity and sensitivity, our research managed to outperform with results of 6.24% and 19.67%, respectively. This proves that deep learning algorithms are better used than machine learning in terms of arrhythmia classification to get more optimal results.…”

“…According to Almazrouei & Al-Rajab (2022), the current trend of utilizing machine learning approaches for diagnosing arrhythmias still lacks precision. Therefore, deep learning algorithms were chosen for classification.…”

“…However, Almazrouei & Al-Rajab (2022) research exclusively employed the Convolutional Neural Network (CNN) deep learning algorithm. Hence, the objective of this study is to utilize the Long Short-Term Memory (LSTM) deep learning algorithm instead.…”