Model-based feature extraction of electrocardiogram using mean shift.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Medicine

Background:

• Accurate feature extraction from electrocardiogram (ECG) signals is crucial for automated medical diagnosis.
• ECG signals are often corrupted by various types of noise (white, pink, brown), complicating feature extraction.
• Existing methods may struggle with noise robustness and precise feature identification.

Purpose of the Study:

• To develop an accurate and robust ECG feature extraction method.
• To address the challenge of noise interference in ECG signals.
• To improve the reliability of automated ECG analysis.

Main Methods:

• Proposed an ECG feature extraction method based on the mean shift algorithm.
• Utilized the algorithm's embedded Gaussian filter for noise reduction.
• Employed gradient optimization with self-adaptive search steps to locate signal extremes.
• Conducted experiments using synthesized ECG signals (ECGSyn) with controlled noise levels (5-15 dB SNR).

Main Results:

• The proposed method demonstrated effectiveness in handling white, pink, and brown noise.
• Quantitative evaluation showed satisfactory feature extraction performance.
• The mean shift algorithm successfully removed noise while preserving essential ECG features.
• Experimental results validated the method's accuracy and robustness.

Conclusions:

• The mean shift-based ECG feature extraction method is accurate and robust against various noise types.
• This approach offers a reliable solution for preprocessing ECG signals for automated diagnostic systems.
• The method shows significant potential for enhancing the performance of computer-aided ECG analysis.