ECG R-peak Detection Using CNN Segmentation and Regression

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing CNN-based semantic segmentation models for ECG signals often fail to accurately detect a unique R-peak in a QRS waveform, as they may detect multiple R-peaks or incorrect peaks outside the QRS waveform.

Innovation Solution

A method and system that utilize a waveform detection model for identifying QRS waveforms and a regression model to detect a unique R-peak, combining convolutional neural networks with residual and squeeze-and-excitation blocks for improved performance, and a regression analysis to refine the output and locate the R-peak accurately.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If only CNN-based semantic segmentation model is used to detect R-peak, then the model can identify characteristic sections (P-wave, QRS complex, T-wave), but multiple R-peaks may be detected in a single QRS waveform or R-peak may be detected outside of the QRS waveform

Engineering Contradiction:

ImproveR-peak detection accuracyVSAvoidUnique R-peak identification reliability

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The detection process is segmented into two distinct stages: first, a CNN-based semantic segmentation model identifies characteristic ECG sections (P-wave, QRS complex, T-wave) to locate potential R-peak regions; second, a regression model processes these segmented regions to identify the unique R-peak within each QRS waveform. This segmentation approach resolves the contradiction by separating initial detection from precise identification.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The regression model serves as an intermediary between the CNN-based segmentation model and the final R-peak detection. It takes the segmented QRS waveform regions as input and refines the detection to identify the unique R-peak location, acting as a mediator that corrects the limitations of the segmentation model alone.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If additional processing is added to CNN-based model output to detect unique R-peak, then unique R-peak detection is achieved, but system complexity increases

Engineering Contradiction:

ImproveUnique R-peak identification reliabilityVSAvoidDetection system complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system merges two different modeling approaches (CNN-based semantic segmentation model and regression model) into a unified detection pipeline. The segmentation model provides regional information while the regression model provides precise localization, and their outputs are combined to achieve reliable unique R-peak detection without requiring a completely complex new system.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The regression model is designed to work universally with the CNN-based segmentation model output, processing the segmented ECG signal regions to identify R-peaks across different ECG patterns and conditions. This multi-functional approach allows the system to handle various ECG variations without requiring separate specialized processing for each case.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20240016434A1Method, system and non-transitory computer-readable recording medium for detecting and classifying beat in electrocardiogram signal

Publication Date: 2024.01.18 HUINNO

AI summary

A method for detecting and classifying a beat in an electrocardiogram (ECG) signal includes detecting a QRS waveform in an ECG signal using a waveform detection model, and detecting a class of a cardiac event capable of being derived from the QRS waveform; and detecting a unique R-peak in the QRS waveform using a regression model.