Training framework for multi-group electrocardiography (MG-ECG) analysis

Patent metadata

Field	Value
Publication number	US-11450424-B2
Application number	US-201916556491-A
Country	US
Kind code	B2
Filing date	Aug 30, 2019
Priority date	Aug 30, 2019
Publication date	Sep 20, 2022
Grant date	Sep 20, 2022

Abstract

A method of performing electrocardiography (ECG) analysis by at least one processor, the method including receiving ECG data that is from multiple leads; grouping the ECG data into groups of data; generating, from each group of the groups of data, a feature vector using a respective machine learning model; and performing ECG analysis using the feature vectors generated from each of the groups of data.

First claim

What is claimed is:

1. A method of performing electrocardiography (ECG) analysis training by at least one processor, the method comprising:
receiving a plurality of pieces of ECG data, each of the plurality of pieces of ECG data including data from each of multiple leads;
identifying common features in the plurality of pieces of ECG data;
grouping individual pieces of the plurality of pieces of the ECG data into groups of data based on the identified common features, wherein the identified common features are based on a combination of a contiguity of the multiple leads and a random grouping strategy;
generating, from each group of the groups of data, a feature vector using a respective machine learning model, wherein the respective machine learning model is used to generate the feature vector for a respective group of the groups of data, and wherein each of the respective machine learning models used has a group-specific extraction strategy based on the each of the respective machine learning models being a model trained separately; and
performing ECG analysis using the feature vectors generated from each of the groups of data.

2. The method of claim 1 , wherein the grouping further includes grouping the individual pieces of the plurality of pieces of the ECG data into the groups of data by using a grouping criteria pool or rules stored in memory.

3. The method of claim 2 , wherein
the grouping includes grouping the individual pieces of the plurality of pieces of the ECG data into the groups of data using the grouping criteria pool, and the identified common features include geometry properties of each of the multiple leads.

4. The method of claim 1 , wherein the respective machine learning models used for each group of the groups of data have a same subset of parameters.

5. The method of claim 1 , wherein the performing ECG analysis comprises:
selecting a model from a plurality of models stored in memory; and
generating an output using the feature vectors generated from each of the groups of data with the model.

6. A device for performing electrocardiography (ECG) analysis training, the device comprising:
at least one memory configured to store computer program code;
at least one processor configured to access said computer program code and operate as instructed by said computer program code, said computer program code including:
receiving code configured to cause the at least one processor to receive a plurality of pieces of ECG data, each of the plurality of pieces of ECG data including data from each of multiple leads;
identifying code configured to identify common features in the plurality of pieces of ECG data;
grouping code configured to cause the at least one processor to group individual pieces of the plurality of pieces of the ECG data into groups of data based on the identified common features, wherein the identified common features are based on a combination of a contiguity of the multiple leads and a random grouping strategy;
generating code configured to cause the at least one processor to generate, from each group of the groups of data, a feature vector using a respective machine learning model stored in the at least one memory, wherein the respective machine learning model is used to generate the feature vector for a respective group of the groups of data, and wherein each of the respective machine learning models used has a group-specific extraction strategy based on the each of the respective machine learning models being a model trained separately; and
performing code configured to cause the at least one processor to perform ECG analysis using the feature vectors generated from each of the groups of data.

7. The device of claim 6 , wherein the grouping code is configured to cause the at least one processor to group the individual pieces of the plurality of pieces of the ECG data into the groups of data by using a grouping criteria pool or rules stored in the at least one memory.

8. The device of claim 7 , wherein the grouping code is configured to cause the at least one processor to group the individual pieces of the plurality of pieces of the ECG data into the groups of data by using the grouping criteria pool, and the identified common features include geometry properties of each of the multiple leads.

9. The device of claim 6 , wherein the respective machine learning models used for each group of the groups of data have a same subset of parameters.

10. The device of claim 6 , wherein the performing code comprises:
model selecting code configured to cause the at least one processor to select a model from a plurality of models stored in the at least one memory; and
output generating code configured to cause the at least one processor to generate an output using the feature vectors generated from each of the groups of data with the model.

11. A non-transitory computer-readable medium storing computer instructions that, when executed by at least one processor of a device, cause the at least one processor to:
receive a plurality of pieces of ECG data, each of the plurality of pieces of ECG data including data from each of multiple leads;
identifying common features in the plurality of pieces of ECG data;
group individual pieces of the plurality of pieces of the ECG data into groups of data based on the identified common features, wherein the identified common features are based on a combination of a contiguity of the multiple leads and a random grouping strategy;
generate, from each group of the groups of data, a feature vector using a respective machine learning model stored in a memory, wherein the respective machine learning model is used to generate the feature vector for a respective group of the groups of data, and wherein each of the respective machine learning models used has a group-specific extraction strategy based on the each of the respective machine learning models being a model trained separately; and
perform ECG analysis using the feature vectors generated from each of the groups of data.

Assignees

• Tencent America LLC

Inventors

• Wang Kun
• Yang Tao
• Tu min
• Li Yaliang
• Tang Hui
• Fan Wei

Classifications

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title