Visual saliency detection approach for long-term ECG analysis.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Electrocardiogram (ECG) signal processing for QRS-complex detection has been researched for decades.
• Developing superior ECG processing algorithms remains a global research priority.
• Existing methods face challenges in filtering complex noise types like motion artifacts (MA) and sudden changes in baseline (SCB).

Purpose of the Study:

• To propose a novel saliency detection-based technique for processing one-dimensional biomedical signals, specifically ECG.
• To enhance the accuracy of QRS-complex detection by effectively addressing MA and SCB noise.
• To introduce a new approach for identifying abnormal QRS-complexes in ECG signals.

Main Methods:

• A trigonometric threshold technique identifies initial QRS-complexes.
• Feature extraction and normalization create grayscale images for MA/SCB noise detection via saliency analysis.
• Further feature extraction and normalization create RGB images for abnormal QRS-complex identification using saliency analysis.

Main Results:

• The technique successfully identifies regions corrupted by MA and SCB noise.
• Abnormal QRS-complexes are detected from noise-free regions using a combined feature approach.
• Performance evaluation on long-term ECG data (17.54 days) demonstrates effectiveness.

Conclusions:

• The proposed saliency detection technique offers an advantageous approach to ECG signal processing.
• The method shows potential for real-time implementation in clinical settings.
• Clinicians and cardiologists provided positive feedback on the technique's applicability and advantages.