Optimization of ECG classification by means of feature selection

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electrocardiogram (ECG) classification is crucial for diagnosing cardiac conditions.
• Existing methods often require significant computational resources, limiting their use in ambulatory settings.
• There is a need for efficient ECG analysis techniques suitable for real-time monitoring.

Purpose of the Study:

• To develop a computationally efficient methodology for ECG classification.
• To enhance the performance of ECG arrhythmia classification.
• To adapt ECG analysis for improved ambulatory monitoring.

Main Methods:

• Application of the sequential forward floating search (SFFS) algorithm.
• Development of a novel criterion function index based on linear discriminants for feature selection.
• Evaluation of selected features using a multilayer perceptron (MLP) classifier.
• Adherence to Association for the Advancement of Medical Instrumentation (AAMI) standard EC57:1998 for performance estimation.

Main Results:

• The proposed method successfully identified a highly suitable subset of ECG features.
• The selected features, when evaluated with an MLP, demonstrated robust classification performance.
• The methodology achieved superior performance compared to similar studies under identical constraints.
• Computational resource requirements were significantly reduced, suitable for ambulatory applications.

Conclusions:

• The developed methodology offers an effective and efficient approach to ECG arrhythmia classification.
• The SFFS algorithm with the new criterion index is a valuable tool for feature selection in ECG analysis.
• This approach holds significant potential for improving real-world ambulatory cardiac monitoring.