Cycle-GAN ECG Signal Restoration for Motion Artifact Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ECG signal processing technologies face challenges in accurately denoising biomedical signals corrupted by motion-related artifacts, such as baseline wander, signal cuts, and varying noise levels, which can render ECG signals undiagnosable by machines or doctors.
Innovation Solution
The use of operational cycle-consistent adversarial networks (Cycle-GANs) and self-organized operational neural networks (Self-ONNs) for blind ECG signal restoration, which transforms corrupted ECG segments into clinically clean signals without prior assumptions about artifact types or severity, preserving major signal characteristics like R-peak intervals and QRS waveforms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional denoising methods are used on ECG signals corrupted by motion artifacts, then processing time and computational resources are reduced, but signal quality and diagnostic accuracy deteriorate significantly
Solution Approach 1:
The patent transforms the ECG denoising problem from the time domain to the frequency domain using Fourier transforms, and further to the wavelet domain using continuous wavelet transforms. By changing the representation parameters of the signal, the method can selectively filter artifacts while preserving diagnostically important features like QRS complexes and P waves, thereby improving signal quality without proportionally increasing processing complexity
Solution Approach 2:
The patent introduces wavelet coefficients as an intermediary representation between the raw ECG signal and the final denoised output. These coefficients act as a mediator that separates signal components by frequency and time localization, allowing selective manipulation of artifact-containing coefficients while preserving clean signal coefficients, thus achieving high-quality denoising with controlled computational complexity
2Object-affected harmful factors
If aggressive filtering is applied to remove motion artifacts, then artifact reduction improves, but important diagnostic features like QRS waveforms and R-peak intervals are distorted or lost
Solution Approach 1:
The patent applies different processing strategies to different portions of the ECG signal based on their local characteristics. In the wavelet domain, coefficients corresponding to high-frequency artifact regions are filtered differently from those representing diagnostically important low-frequency cardiac features. This localized approach allows aggressive artifact removal in contaminated regions while preserving signal integrity in clean regions, resolving the contradiction between artifact reduction and feature preservation
Solution Approach 2:
The patent employs dynamic thresholding in the wavelet domain where the filtering strength adapts locally based on the estimated artifact level in each signal segment. Rather than applying a static filter across the entire signal, the method dynamically adjusts filtering parameters to match local artifact characteristics, thereby removing motion artifacts where present while preserving diagnostic features where intact, achieving both artifact reduction and feature preservation
3Reliability
If manual preprocessing and postprocessing are performed to ensure signal quality, then diagnostic accuracy improves, but processing time and operational complexity increase
Solution Approach 1:
The patent performs preliminary denoising and artifact removal directly at the point of data acquisition using the wavelet transform method. By addressing signal quality issues immediately when they occur rather than requiring subsequent manual intervention, the system maintains high diagnostic reliability while eliminating time-consuming manual preprocessing and postprocessing steps, thus resolving the contradiction between reliability and processing time
Data Source
AI summary
Systems, methods, apparatuses, and computer program products for real-time, personalized cardiac monitoring for early detection of heart-beat anomalies. One method may include a device selecting at least one set of clean ECG segments, and at least one set of corrupted ECG segments; transforming at least one of a one-dimensional or two-dimensional version cycle-CANs trained to transform ECG signals from at least one different dataset; and restoring the at least one set of corrupted ECG segments based upon a one- or two-dimensional operational cycle-GAN trained over the batches.