Multi-Distance Dispersion Entropy for ECG Signal Classification

Abstract

Automatic detection of heartbeat is critical for early cardiovascular disease prevention and diagnosis. Traditional feature methodologies based on expert knowledge cannot abstract and represent multidimensional and multi-view information. Hence traditional research on heartbeat detection pattern recognition cannot produce adequate results. The proposed method in this research used Dispersion Entropy (DisEn) on Multidistance Signal Level Difference (MSLD) for feature extraction and Support Vector Machine (SVM) method for classifying the ECG signals. DisEn generates 20 DE values as feature vectors for each MSLD signal with a distance D of 1 to 20. The datasets used in this research were obtained from the MIT-BIH Arrhythmia database of ECG signals that consist of forty-five patients was captured using 200 [adu/mV] amplification and a sampling frequency of 360 Hz. The experiments result using 5-fold cross-validation revealed that at distance D= 1-15 had the highest accuracy of 91% to classify the ECG data into Normal Sinus Rhythm (NSR), Left Bundle Branch Block (LBBB), and Atrial Fibrillation (AFIB) from the MIT-BIH Arrhythmias database.
 

Empowering AI-Diagnosis: Deep Learning Abilities for Accurate Atrial Fibrillation Classification

“…Unfortunately, the automatic detection of AF is a complex problem, and state-of-theart performance is typically achieved through the use of machine learning (ML). Lack of ML still requires human intervention for feature representation [6][7][8][9].…”

“…These networks are applied in one-dimensional (1D) AF classification and have shown promising results [10][11][12][13][14]. CNN models used for AF classification can perform both feature extraction and classification without the need for manual feature extraction [9]. Such a network consists of multiple back-to-back layers connected in a feed-forward manner, including convolutional, normalization, pooling, and fully connected layers [15].…”

“…Such a network consists of multiple back-to-back layers connected in a feed-forward manner, including convolutional, normalization, pooling, and fully connected layers [15]. The 1D-CNN technique is useful for extracting features from time sequence data for robust AF classification [9,10]. However, the intermittent nature of AF poses a challenge when it comes to screening for this condition.…”