CNN and SVM-Based Models for the Detection of Heart Failure Using Electrocardiogram Signals

• Published in: Sensors (Basel, Switzerland) Vol. 22; no. 23; p. 9190
• Main Authors: Botros, Jad, Mourad-Chehade, Farah, Laplanche, David
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 26.11.2022; MDPI
• Subjects: Accuracy; Algorithms; Analysis; Arrhythmias, Cardiac - diagnosis; Artificial intelligence; binary classification; Cardiology; Classification; convolutional neural network; Deep learning; Electric properties; Electrical conduction; Electrocardiogram; Electrocardiography; Engineering Sciences; Heart attack; Heart failure; Heart Failure - diagnosis; Humans; Ischemia; Machine learning; Neural networks; Neural Networks, Computer; Nutrients; Patients; Portable equipment; Signal Processing, Computer-Assisted; Support Vector Machine; Support vector machines; deep learning; heart failure; electrocardiogram; support vector machine; convolutional neural network; binary classification; machine learning
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s22239190

Heart failure (HF) is a serious condition in which the heart fails to supply the body with enough oxygen and nutrients to function normally. Early and accurate detection of heart failure is critical for impeding disease progression. An electrocardiogram (ECG) is a test that records the rhythm and electrical activity of the heart and is used to detect HF. It is used to look for irregularities in the heart's rhythm or electrical conduction, as well as a history of heart attacks, ischemia, and other conditions that may initiate HF. However, sometimes, it becomes difficult and time-consuming to interpret the ECG signal, even for a cardiac expert. This paper proposes two models to automatically detect HF from ECG signals: the first one introduces a Convolutional Neural Network (CNN), while the second one suggests an extension of it by integrating a Support Vector Machine (SVM) layer for the classification at the end of the network. The proposed models provide a more accurate automatic HF detection using 2-s ECG fragments. Both models are smaller than previously proposed models in the literature when the architecture is taken into account, reducing both training time and memory consumption. The MIT-BIH and the BIDMC databases are used for training and testing the adopted models. The experimental results demonstrate the effectiveness of the proposed framework by achieving an accuracy, sensitivity, and specificity of over 99% with blindfold cross-validation. The models proposed in this study can provide doctors with reliable references and can be used in portable devices to enable the real-time monitoring of patients.