System and method for machine-learning-based atrial fibrillation detection

• US 10,463,269 B2
• Filed: 11/26/2018
• Issued: 11/05/2019
• Est. Priority Date: 09/25/2013
• Status: Active Grant

• Current Assignee: Bardy Diagnostics, Inc. (Baxter International Inc.)
• Original Assignee: Bardy Diagnostics, Inc. (Baxter International Inc.)
• Inventors: Boleyn, Rodney, Dreisbach, Ezra M., Dulken, Chuck, Bardy, Gust H.
• Primary Examiner(s): Gedeon, Brian T
• Application Number:
  US16/200,089
• Publication Number:
  US 20190090769A1
• Time in Patent Office: 344 Days
• Field of Search: None
• CPC Class Codes: A61B 2560/0487 Special user inputs or interfaces
• CPC Class Codes: A61B 2562/08 Sensors provided with means for identification, e.g. barcodes or memory chips
• CPC Class Codes: A61B 5/0006 ECG or EEG signals
• CPC Class Codes: A61B 5/0022 Monitoring a patient using a global network, e.g. telephone networks, internet
• CPC Class Codes: A61B 5/257 using adhesive means, e.g. adhesive pads or tapes
• CPC Class Codes: A61B 5/259 using conductive adhesive means, e.g. gels
• CPC Class Codes: A61B 5/28 for electrocardiography [ECG]
• CPC Class Codes: A61B 5/282 Holders for multiple electrodes for introduction into the body A61B5/287
• CPC Class Codes: A61B 5/316 Modalities, i.e. specific diagnostic methods
• CPC Class Codes: A61B 5/318 Heart-related electrical modalities, e.g. electrocardiography [ECG]
• CPC Class Codes: A61B 5/33 specially adapted for cooperation with other devices
• CPC Class Codes: A61B 5/332 Portable devices specially adapted therefor
• CPC Class Codes: A61B 5/333 Recording apparatus specially adapted therefor
• CPC Class Codes: A61B 5/339 Displays specially adapted therefor
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/353 Detecting P-waves
• CPC Class Codes: A61B 5/361 Detecting fibrillation
• CPC Class Codes: A61B 5/6823 Trunk, e.g., chest, back, abdomen, hip
• CPC Class Codes: A61B 5/6832 using adhesives A61B5/259 takes precedence
• CPC Class Codes: A61B 5/6833 Adhesive patches
• CPC Class Codes: A61B 5/7203 : for noise prevention, reduction or removal
• CPC Class Codes: A61B 5/7264 : Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7267 : involving training the classification device
• CPC Class Codes: A61B 5/742 : using visual displays A61B5/339 takes precedence
• CPC Class Codes: A61B 5/748 : Selection of a region of interest, e.g. using a graphics tablet
• CPC Class Codes: G06N 20/00 : Machine learning
• CPC Class Codes: G06N 3/045 : Combinations of networks
• CPC Class Codes: G06N 3/08 : Learning methods
• CPC Class Codes: G16H 40/67 : for remote operation
• CPC Class Codes: G16H 50/70 : for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A system and method for machine-learning based atrial fibrillation detection are provided. A database is maintained that is operable to maintain a plurality of ECG features and annotated patterns of the features. At least one server is configured to: train a classifier based on the annotated patterns in the database; receive a representation of an ECG signal recorded by an ambulatory monitor recorder during a plurality of temporal windows; detect a plurality of the ECG features in at least some of the portions of the representation falling within each of the temporal windows; use the trained classifier to identify patterns of the ECG features within one or more of the portions of the ECG signal; for each of the portions, calculate a score indicative of whether the portion of the representation within that ECG signal is associated the patient experiencing atrial fibrillation; and take an action based on the score.

Claims

1. A system for machine-learning-based atrial fibrillation detection with the aid of a digital computer, comprising:

a database operable to maintain a plurality of electrocardiography (ECG) features and annotated patterns of the features, at least some of the patterns associated with atrial fibrillation;

at least one server interconnected to the database, the at least one server configured to;

train a classifier based on the annotated patterns in the database;

receive a representation of an ECG signal recorded by an ambulatory monitor recorder during a plurality of temporal windows;

detect a plurality of the ECG features in at least some of the portions of the representation falling within each of the temporal windows;

use the trained classifier to identify patterns of the ECG features within one or more of the portions of the ECG signal;

for each of the portions, calculate a value indicative of whether the portion of the representation within that ECG signal is associated the patient experiencing atrial fibrillation;

calculate a further value indicative of whether the portion of the representation within that ECG signal is associated with the patient not experiencing atrial fibrillation;

compare the further value to the value;

determine that the portion of the ECG signal is associated with the patient experiencing atrial fibrillation based on the comparison; and

take an action based on the determination that the portion of the ECG signal is associated with the patient experiencing atrial fibrillation.

2. A system for machine-learning-based atrial fibrillation detection according to claim 1, the at least one server further configured to:

obtain training data comprising a plurality of the ECG features and a plurality of patterns of the ECG features; and

obtain annotations of patterns of the ECG features in the training data, wherein the training of the classifier is based on the annotations.

3. A system for machine-learning-based atrial fibrillation detection according to claim 1, the at least one server further configured to:

test an accuracy of the trained classifier and perform further training based on a result of the test.

4. A system for machine-learning-based atrial fibrillation detection according to claim 1, wherein the determination is made upon the value exceeding the further value.

5. A system for machine-learning-based atrial fibrillation detection according to claim 1, wherein the action comprises sending an alert regarding the determination.

6. A system for machine-learning-based atrial fibrillation detection according to claim 1, the at least one server further configured to:

generate a matrix with the identified features and the patterns; and

generate at least one matrix with weights for the identified features and patterns, wherein the value and the further value are calculated using the weight matrix.

7. A system for machine-learning-based atrial fibrillation detection according to claim 1, wherein each of the temporal windows is between 2 and 60 seconds.

8. A system for machine-learning-based atrial fibrillation detection according to claim 1, wherein the database comprises 32 of the ECG features.

9. A system for machine-learning-based atrial fibrillation detection according to claim 1, the at least one server further configured to:

perform a noise filtering of at least some of the portions of the ECG signal prior to identification of the ECG features.

10. A method for machine-learning-based atrial fibrillation detection with the aid of a digital computer, comprising:

maintaining in a database a plurality of electrocardiography (ECG) features and annotated patterns of the features, at least some of the patterns associated with atrial fibrillation;

training by an at least one server connected to the database a classifier based on the annotated patterns in the database;

receiving by the at least one server a representation of an ECG signal recorded by an ambulatory monitor recorder during a plurality of temporal windows;

detecting by the at least one server a plurality of the ECG features in at least some of the portions of the representation falling within each of the temporal windows;

using by the at least one server the trained classifier to identify patterns of the ECG features within one or more of the portions of the ECG signal;

for each of the portions, calculating by the at least one server a value indicative of whether the portion of the representation within that ECG signal is associated the patient experiencing atrial fibrillation;

calculating by the at least one server a further value indicative of whether the portion of the representation within that ECG signal is associated with the patient not experiencing atrial fibrillation;

comparing the further value to the score;

determining that the portion of the ECG signal is associated with the patient experiencing atrial fibrillation based on the comparison;

taking by the at least one server an action based on the determination that the portion of the ECG signal is associated with the patient experiencing atrial fibrillation.

11. A method for machine-learning-based atrial fibrillation detection according to claim 10, further comprising:

obtaining training data comprising a plurality of the ECG features and a plurality of patterns of the ECG features; and

obtaining annotations of patterns of the ECG features in the training data, wherein the training of the classifier is based on the annotations.

12. A method for machine-learning-based atrial fibrillation detection according to claim 10, further comprising:

test an accuracy of the trained classifier and performing further training based on a result of the test.

13. A method for machine-learning-based atrial fibrillation detection according to claim 10, wherein the determination is made upon the value exceeding the further value.

14. A method for machine-learning-based atrial fibrillation detection according to claim 10, wherein the action comprises sending an alert of the regarding the determination.

15. A method for machine-learning-based atrial fibrillation detection according to claim 10, further comprising:

generating a matrix with the identified features and the patterns; and

generating at least one matrix with weights for the identified features and patterns, wherein the value and the further value are calculated using the weight matrix.

16. A method for machine-learning-based atrial fibrillation detection according to claim 10, wherein each of the temporal windows is between 2 and 60 seconds.

17. A method for machine-learning-based atrial fibrillation detection according to claim 10, wherein the database comprises 32 of the ECG features.

18. A method for machine-learning-based atrial fibrillation detection according to claim 10, further comprising:

performing a noise filtering of at least some of the portions of the ECG signal prior to identification of the ECG features.