A genetic segmentation of ECG signals

Area of Science:

• Biomedical Signal Processing
• Computational Intelligence
• Data Compression

Background:

• Electrocardiogram (ECG) signal analysis is crucial for diagnosing cardiac conditions.
• Existing ECG segmentation methods often lack efficiency for lossy compression.
• Signal linearization and monotonicity are key challenges in ECG data processing.

Purpose of the Study:

• To develop an advanced segmentation technique for ECG signals.
• To enable efficient lossy compression of ECG data using geometric constructs.
• To identify optimal signal segments with high monotonicity for improved analysis.

Main Methods:

• Utilized genetic algorithms (GAs) for ECG signal segmentation.
• Developed a chromosome structure representing segment endpoints.
• Designed a fitness function maximizing signal monotonicity and minimizing derivative variability.

Main Results:

• Successfully segmented ECG signals into regions of high monotonicity.
• Demonstrated the effectiveness of GA-based segmentation for normal and abnormal heartbeats (left/right bundle branch block).
• Established a strong correlation between fitness function values and segment approximation accuracy (sum of squared errors).

Conclusions:

• The proposed GA-based segmentation technique effectively identifies monotonic ECG signal segments.
• This approach facilitates efficient lossy compression and enhances ECG data analysis.
• The method shows promise for various ECG signal classes, including those with abnormalities.