Intracardiac ECG Denoising With Weighted Autoencoder Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating electrocardiograms (ECGs) during cardiac procedures fail to effectively emphasize zones or events of clinical importance, such as potential origination locations of cardiac conditions, due to the use of mean square error (MSE) functions and other regression loss functions that do not adequately highlight these critical areas.
Innovation Solution
A training algorithm utilizing deep learning techniques applies filters and energy thresholds to intracardiac and body surface electrocardiograms, refining them using a denoising autoencoder to produce improved ECGs by emphasizing clinical significance and removing noise, thereby enhancing the accuracy of cardiac condition identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional mean square error loss functions are used for autoencoder training, then the model can learn general signal representations, but it fails to emphasize zones of clinical importance and cannot adequately reduce noise in critical areas
Solution Approach 1:
The patent applies local quality by creating a spatially varying loss function where different regions of the ECG signal are weighted differently based on their clinical importance. The loss function assigns higher weights to zones of clinical importance (such as ST segments, QRS complexes) and lower weights to less critical regions, enabling the autoencoder to prioritize noise reduction and feature preservation in clinically significant areas while maintaining overall signal reconstruction capability.
Solution Approach 2:
The patent changes the parameter of the loss function from a uniform mean square error to a spatially varying weighted loss function. This parameter change introduces a weight map that varies across the signal timeline, transforming the optimization objective to emphasize clinically important regions. The weight map can be derived from signal characteristics, clinical annotations, or automated detection of critical zones, fundamentally altering how the autoencoder learns to reconstruct the signal.
2Object-affected harmful factors
If noise reduction is aggressively applied to intracardiac ECGs, then signal clarity improves, but important clinical features and signal nuances may be lost or distorted
Solution Approach 1:
The patent applies local quality by implementing region-specific noise reduction strategies through the weighted loss function. Different segments of the ECG signal receive different levels of denoising pressure based on their clinical importance and noise characteristics. Critical zones like the ST segment and QRS complex are protected from aggressive denoising that could distort diagnostic features, while less critical regions undergo more aggressive noise suppression.
Solution Approach 2:
The patent incorporates feedback mechanisms where the autoencoder's reconstruction output is compared against the original signal using the weighted loss function. This feedback loop continuously adjusts the encoding and decoding operations to minimize weighted error, providing real-time correction that preserves clinically important features while removing noise. The feedback ensures that denoising operations do not inadvertently remove or distort diagnostic information.
Data Source
AI summary
A system and method include a memory storing processor executable code for a denoised autoencoder, and one or more processors coupled to the memory to execute the processor executable code to receive raw signal data comprising signal noise, encode, by the denoised autoencoder, the raw signal data by performing a denoising autoencoder operation to produce a latent representation, and decode, by the denoised autoencoder, the latent representation to produce clean signal data reconstructed without the signal noise. A first filter is applied to a signal to emphasize activity within the signal and to produce a first modified signal, a rectifier and a second filter are applied to the first modified signal to smooth areas of the first modified signal with clinical importance and to produce a second modified signal, and high frequency energy zones of the second modified signal are automatically detected using an energy threshold to produce a weights vector.