Reconstructed State Space Features for Classification of ECG Signals

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Informatics

Background:

• Cardiac arrhythmias are critical health conditions requiring rapid diagnosis from electrocardiogram (ECG) signals.
• Distinguishing between different arrhythmias and normal ECG signals is challenging for cardiologists due to signal similarities.
• Accurate ECG interpretation is vital for effective patient management and treatment.

Purpose of the Study:

• To develop and evaluate an image-based machine learning method for classifying cardiac arrhythmias.
• To differentiate between Ventricular Fibrillation (VF), Ventricular Tachycardia (VT), Supraventricular Tachycardia (SVT), and normal ECG signals.
• To enhance the diagnostic accuracy of ECG analysis for arrhythmias.

Main Methods:

• ECG data from three databases (Boston Beth University, Creighton University, MIT-BIH) were utilized.
• An algorithm was developed using MATLAB, transforming ECG signals into a state-space representation with optimal time delay determined by particle swarm optimization.
• Features were extracted from binary images of the signals and processed using a multilayer perceptron neural network for classification.

Main Results:

• The proposed method achieved high classification accuracies for various arrhythmia combinations.
• Specific accuracies included 99.5% for N-VF, 100% for VT-SVT, 94.98% for N-SVT, and 100% for VF-VT.
• The system demonstrated excellent performance in distinguishing between normal and abnormal ECG signals, including complex combinations like VT-VF-SVT-N (95% accuracy).

Conclusions:

• A novel approach for classifying ECG signals of VT, VF, and SVT against normal signals was successfully developed.
• The proposed image-based machine learning system demonstrated superior performance compared to other related studies.
• This method offers a promising tool for improving the accuracy and efficiency of cardiac arrhythmia diagnosis.