Classification of Continuous ECG Segments - Performance Analysis of a Deep Learning Model

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) signal classification is crucial for diagnosing heart diseases like arrhythmias.
• Current intelligent systems often analyze ECG signals segment by segment, which may miss complex pathological patterns.
• Diagnosing ECG pathologies requires analyzing successive cardiac cycles, not just individual segments.

Purpose of the Study:

• To evaluate a deep learning method (CNN 1D+MLP) for diagnosing heart conditions using individual and successive ECG signal segments.
• To assess the robustness of the CNN 1D+MLP model in analyzing multiple simultaneous ECG segments.

Main Methods:

• Utilized two ECG databases: MIT-BIH Arrhythmia and European ST-T.
• Created datasets with consecutive segments (2, 3, and 5) for training and testing.
• Employed a Convolutional Neural Network 1D paired with a Multilayer Perceptron (CNN 1D+MLP) for classification.

Main Results:

• The CNN 1D+MLP method demonstrated high robustness in analyzing successive ECG signal segments.
• The model showed potential for detecting anomalous patterns within the ECG signal.
• Classification metrics including precision, recall, f1-score, accuracy, and confusion matrices were used for evaluation.

Conclusions:

• The evaluated deep learning method is robust for analyzing sequential ECG segments.
• This approach holds significant potential for improved ECG-based pathology detection.
• Future work includes real-time ECG analysis from wearable devices via cloud systems.