Atrial fibrillation signal recognition method, apparatus and device

Patent metadata

Field	Value
Publication number	US-11538588-B2
Application number	US-202016901033-A
Country	US
Kind code	B2
Filing date	Jun 15, 2020
Priority date	Dec 19, 2017
Publication date	Dec 27, 2022
Grant date	Dec 27, 2022

Abstract

The present disclosure provides an atrial fibrillation signal recognition method, apparatus and device. The method comprises: obtaining an electrocardiogram signal to be recognized; inputting the electrocardiogram signal to be recognized to a pre-established atrial fibrillation signal recognition model, and outputting an atrial fibrillation signal recognition result, where the atrial fibrillation signal recognition model is established in the following way: obtaining a specified number of electrocardiogram sample signals and corresponding identifier information; balancing, according to the number of normal signals, atrial fibrillation signals by means of SMOTE; establishing a network structure of multiple convolutional neural networks, each of the convolutional neural networks being provided with a specific receptive field for recognizing the atrial fibrillation signals of a corresponding granularity; and inputting the normal signals and the balanced atrial fibrillation signals to the network structure for training to generate an atrial fibrillation signal recognition model.

First claim

What is claimed is:

1. An atrial fibrillation signal recognition method, wherein the method comprises:
obtaining an electrocardiogram signal to be recognized; and
inputting the electrocardiogram signal to be recognized to a pre-established atrial fibrillation signal recognition model, and outputting an atrial fibrillation signal recognition result, wherein
the atrial fibrillation signal recognition model is established in the following way:
obtaining a specified number of electrocardiogram sample signals and corresponding identifier information, wherein the identifier information comprises identifier information of normal signals and of atrial fibrillation signals;
balancing, according to the number of normal signals, the atrial fibrillation signals by means of synthetic minority oversampling technique (SMOTE);
establishing a network structure of multiple convolutional neural networks, each of the convolutional neural networks being provided with a specific receptive field for recognizing the atrial fibrillation signals of a corresponding granularity; and
inputting the normal signals and the balanced atrial fibrillation signals to the network structure for training, to generate the atrial fibrillation signal recognition model;
wherein the network structure of multiple convolutional neural networks comprises a first network and a second network;
both the first network and the second network are VGG-16 convolutional neural networks; each of the first network and the second network comprises multiple convolutional lavers and multiple max-pooling layers; a receptive field of a convolution kernel in the first network is less than or equal to a receptive field of a convolution kernel in the second network;
max-pooling lavers of the last layers of the first network and the second network are connected to each other; the interconnected max-pooling layers are further successively connected to multiple fully connected layers and a Softmax layer.

2. The method according to claim 1 , wherein the step of obtaining an electrocardiogram signal to be recognized comprises:
obtaining an original electrocardiogram signal; and
preprocessing the original electrocardiogram signal to generate an electrocardiogram signal to be recognized, wherein the preprocessing comprises filtering processing and regularization processing.

3. The method according to claim 1 , wherein the step of inputting the normal signals and the balanced atrial fibrillation signals to the network structure for training, to generate the atrial fibrillation signal recognition model comprises:
inputting the normal signals and the balanced atrial fibrillation signals to the network structure for training, to generate an initial model;
calculating sensitivity of the initial model:

Sen
=

#
⁢

(
TP
)

#
⁢

(
TP
)

+

#
⁢

(
FN
)

;

calculating specificity of the initial model:

Spe
=

#
⁢

(
TN
)

#
⁢

(
TN
)

+

#
⁢

(
FP
)

;

calculating precision of the initial model:

Pre
=

#
⁢

(
TP
)

#
⁢

(
TP
)

+

#
⁢

(
FP
)

;

calculating accuracy of the initial model:

Acc
=

#
⁢

(
TP
)

+

#
⁢

(
TN
)

#
⁢

(

TP
+
TN
+
FN
+
FP

)

;

wherein IP represents a correctly recognized atrial fibrillation signal; FP represents an incorrectly recognized atrial fibrillation signal; TN represents a correctly recognized normal signal;

FN represents an incorrectly recognized normal signal;
determining whether the sensitivity, the specificity, the precision, and the accuracy meet corresponding thresholds respectively; if no, adjusting configuration parameters in the network structure until the sensitivity, the specificity, the precision, and the accuracy meet the corresponding thresholds; and
determining the initial model as the atrial fibrillation signal recognition model.

4. An atrial fibrillation signal recognition device, comprising a processor and a non-transitory ma.chine-readable storage medium, wherein the non-transitory machine-readable storage medium stores a machine-executable instruction that is executable by the processor, and the pr

Assignees

• Shenzhen Inst Of Adv Tech Cas

Inventors

• Li Ye
• Fan Xiaomao
• Yao Qihang
• Yin Liyan

Classifications

• G16H50/20 Primary

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• A61B5/352

  Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval · CPC title

• A61B5/7267

  involving training the classification device · CPC title

• G06N3/045

  Combinations of networks · CPC title

• A61B5/318

  Heart-related electrical modalities, e.g. electrocardiography [ECG] · CPC title