Robust Peak Detection for Holter ECGs by Self-Organized Operational Neural Networks.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• R-peak detection in electrocardiogram (ECG) signals is crucial for cardiac monitoring.
• Existing R-peak detectors struggle with low-quality and noisy ECG signals from mobile sensors.
• Deep 1-D convolutional neural networks (CNNs) show promise but require high computational complexity.

Purpose of the Study:

• To develop a novel R-peak detection method with high performance and computational efficiency.
• To address the limitations of homogeneous CNNs by introducing heterogeneous network configurations.
• To propose 1-D Self-Organized Operational Neural Networks (Self-ONNs) with generative neurons.

Main Methods:

• Proposed 1-D Self-Organized Operational Neural Networks (Self-ONNs) utilizing generative neurons.
• Leveraged the self-organization capability of generative neurons to create optimal operators during training.
• Evaluated performance on the China Physiological Signal Challenge-2020 (CPSC) dataset, comprising over one million ECG beats.

Main Results:

• The proposed 1-D Self-ONNs significantly outperformed state-of-the-art deep CNNs.
• Achieved a 99.10% F1-score, 99.79% sensitivity, and 98.42% positive predictivity on the CPSC dataset.
• Demonstrated superior performance with less computational complexity compared to existing methods.

Conclusions:

• 1-D Self-ONNs represent a significant advancement in R-peak detection for noisy ECG signals.
• The self-organization capability of generative neurons enhances both performance and efficiency.
• This method sets a new benchmark for R-peak detection accuracy and computational elegance.