US-12622626-B2 - Reducing noise of intracardiac electrocardiograms using an autoencoder and utilizing and refining intracardiac and body surface electrocardiograms using deep learning training loss functions

Abstract

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.

Inventors

• Yariv Avraham Amos
• Matityahu Amit
• Liat Tsoref
• Stanislav Goldberg

Assignees

• BIOSENSE WEBSTER (ISRAEL) LTD.

Dates

Publication Date

20260512

Application Date

20210602

Claims (17)

1. 1 . A method comprising: receiving raw signal data comprising signal noise, the signal noise at least including far field artifacts; encoding, by a denoised autoencoder, the raw signal data by performing a denoising autoencoder operation that reduces dimensionality of the received raw signal data to produce a latent representation with at least the far field artifacts removed, the latent representation being generated by an element-wise activation function that applies a weight matrix to the received raw signal data; and decoding, by the denoised autoencoder, the latent representation to produce clean signal data reconstructed without the signal noise.
2. 2 . The method of claim 1 , wherein the raw signal data includes at least one of power line noise, contact noise, deflection noise, Fluro noise, and ventricular far field.
3. 3 . The method of claim 1 , wherein the denoised autoencoder executes in a mode to build a dedicated graphic user interface comprising one or more filters.
4. 4 . The method of claim 1 , wherein the denoised autoencoder executes in a mode to record noise in a control environment and add the noise to the clean signal data.
5. 5 . The method of claim 1 , wherein the raw signal data comprises electrical signals of a heart from N channels of intracardiac electrocardiograms, intracardiac electrocardiograms and body surface electrocardiograms, or intracardiac electrocardiograms with a position of each electrode and anatomy information.
6. 6 . The method of claim 1 , wherein the denoised autoencoder is trained based on a loss function that emphasizes atrial activity.
7. 7 . The method of claim 6 , wherein the loss function utilizes a neural network where a last layer reconstructs a clean signal data.
8. 8 . The method of claim 6 , wherein the loss function comprises a weighted mean square error loss function.
9. 9 . The method of claim 1 , wherein a patient biometric sensor of a monitoring and processing apparatus records the raw signal data.
10. 10 . The method of claim 1 , wherein the denoising autoencoder operation comprises passing the raw signal data through a deep neural network to reduce a dimensionality of the raw signal data and to retain important information.
11. 11 . The method of claim 1 , wherein the latent representation comprises a reduced dimensionality and important information from the raw signal data.
12. 12 . The method of claim 1 , wherein the clean signal data comprises input intracardiac signals reconstructed minimizing the signal noise.
13. 13 . The method of claim 1 , wherein the denoised autoencoder learns the clean signal data to denoise subsequent raw intracardiac signals during the denoising autoencoder operation.
14. 14 . The method of claim 13 , wherein the denoised autoencoder detects noise type and quality of the subsequent raw intracardiac signals.
15. 15 . The method of claim 14 , wherein the denoised autoencoder performs the denoising autoencoder operation on the subsequent raw intracardiac signals based on whether the quality is above a threshold.
16. 16 . The method of claim 1 , wherein the method further comprises generating an electrocardiogram from the clean signal data, the electrocardiogram being substantially free from the signal noise.
17. 17 . A system comprising: a memory storing processor executable code for a denoised autoencoder; and one or more processors coupled to the memory, the one or more processors configured to execute the processor executable code to cause: receiving raw signal data comprising signal noise, the signal noise at least including far field artifacts; encoding, by the denoised autoencoder, the raw signal data by performing a denoising autoencoder operation that reduces dimensionality of the received raw signal data to produce a latent representation with at least the far field artifacts removed, the latent representation being generated by an element-wise activation function that applies a weight matrix to the received raw signal data; and decoding, by the denoised autoencoder, the latent representation to produce clean signal data reconstructed without the signal noise.

Description

CROSS REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Provisional Patent Application No. 63/034,694 (JNJBIO-6332USPSP1) filed on Jun. 4, 2020 and U.S. Provisional Patent Application No. 63/061,929 (JNJBIO-6368USPSP1) filed on Aug. 6, 2020, which are incorporated by reference as if fully set forth. FIELD OF INVENTION The present invention is related to signal processing, artificial intelligence, and machine learning. More particularly, the present invention is related to a system and method for reducing noise of intracardiac electrocardiograms using an autoencoder and utilizing and refining intracardiac and body surface electrocardiograms using one or more deep learning training loss functions. BACKGROUND Treatments for cardiac conditions, such as cardiac arrhythmia, often require heart mapping (i.e., mapping cardiac tissue, chambers, veins, arteries and/or pathways, which is also known as cardiac mapping). Electrocardiograms or electrocardiographs (ECGs) are examples of heart mappings. ECGs are generated from electrical signals from a heart that describe heart activity. ECGs are utilized during cardiac procedures to identify potential origination locations of cardiac conditions. An autoencoder may be utilized to refine the electrical signals of the heart through encoding and decoding operations. The refined electrical signals of the heart may then be used to generate the ECGs. The autoencoder utilizes artificial intelligence and machine learning to build and train the encoding and decoding operations therein. For example, a denoising autoencoder trains the autoencoder to discover more robust features/representations within the electrical signals and prevents the autoencoder from learning a particular identity of the electrical signals. When trying to extract or learn important features/representations during autoencoder training, present mean square error (MSE) functions and/or other regression loss functions fail to emphasize zones or events of clinical importance (e.g., potential origination locations of cardiac conditions). SUMMARY According to an embodiment, a method is provided. The method includes receiving raw signal data including signal noise and encoding, by a denoised autoencoder, the raw signal data. The encoding includes performing a denoising autoencoder operation on the raw signal data to produce a latent representation. The method also includes decoding, by the denoised autoencoder, the latent representation to produce clean signal data reconstructed without the signal noise. According to an embodiment, another method is provided. The method is implemented by a training algorithm executed by a processor coupled to a memory. The training algorithm applies a first filter to a signal to emphasize activity within the signal and to produce a first modified signal. The training algorithm applies a rectifier and a second filter to the first modified signal to smooth areas of the first modified signal with clinical importance and to produce a second modified signal. The training algorithm automatically detects high frequency energy zones of the second modified signal using an energy threshold to produce a weights vector. According to one or more embodiments, the method embodiments above can be implemented as an apparatus, a system, and/or a computer program product. BRIEF DESCRIPTION OF THE DRAWINGS A more detailed understanding may be had from the following description, given by way of example in conjunction with the accompanying drawings, wherein like reference numerals in the figures indicate like elements, and wherein: FIG. 1 illustrates a diagram of an exemplary system in which one or more features of the disclosure subject matter can be implemented according to one or more embodiments; FIG. 2 illustrates a block diagram of an example system for remotely monitoring and communicating patient biometrics according to one or more embodiments; FIG. 3 illustrates a graphical depiction of an artificial intelligence system according to one or more embodiments; FIG. 4 illustrates a block diagram of a method performed in the artificial intelligence system of FIG. 3 according to one or more embodiments; FIG. 5 illustrates a block diagram of a method according to one or more embodiments; FIG. 6A illustrates an example of a neural network according to one or more embodiments; FIG. 6B illustrates a block diagram of a method performed in the neural network of FIG. 6A according to one or more embodiments; FIG. 7 illustrates contact noise examples according to one or more embodiments; FIG. 8A illustrates deflection noise examples according to one or more embodiments; FIG. 8B illustrates deflection noise examples of FIG. 8A with an increased x-axis to zoom in on features according to one or more embodiments; FIG. 9 illustrates a graphical depiction of a signal according to one or more embodiments; FIG. 10 illustrates a graphical depiction of a signal according to one or more embodim