Detection of cardiac conditions from reduced lead set ECG
Embodiments of the present disclosure utilize DNNs for ECG interpretation, where original ECG waveforms are directly ingested by the DNNs using paired interpretation labels for training, without the need for explicatory feature extraction or rule-based criteria.
1 . A method comprising:
training a first deep neural network (DNN) to classify electrocardiogram (ECG) rhythm characteristics, wherein a training weight matrix is used to balance a difference in prevalence among classifications during training;
training a second DNN to classify ECG morphology characteristics;
training a third DNN to identify global ECG parameters, wherein both the first and second DNNs comprise a classification layer and the third DNN comprises a regression layer; and
implementing the first, second, and third DNNs as an ensemble DNN for ECG interpretation.
2 . The method of claim 1 , further comprising:
in response to receiving an ECG:
classifying rhythm characteristics of the ECG using the first DNN;
classifying morphology characteristics of the ECG using the second DNN; and
identifying global parameters of the ECG using the third DNN; and
applying a set of decision logic to the ECG rhythm characteristics and the ECG morphology characteristics to diagnose one or more conditions indicated by the ECG based at least in part on the ECG global parameters.
3 . The method of claim 2 , wherein the set of decision logic comprises:
a first set of criteria for validating the classification of the rhythm characteristics of the ECG based at least in part on the global parameters of the ECG; and
a second set of criteria for validating the classification of the morphology characteristics of the ECG based at least in part on the global parameters and the validated classification of the rhythm characteristics of the ECG.
4 . The method of claim 1 , wherein:
the classification of the rhythm characteristics of the ECG comprises a probability for each of a set of rhythms; and
the classification of the morphology characteristics of the ECG comprises a probability for each of a set of morphologies.
5 . The method of claim 4 , wherein one or more of the first, second and third DNN comprise convolution neural networks (CNNs).
6 . The method of claim 5 , further comprising:
using a random data shifting technique to further generalize each of the first, second and third DNNs.
7 . The method of claim 1 , wherein each of the first, second and third DNNs comprises a convolution block having a residual block structure that combines a set of convolution layers and a set of skip layers.
8 . A system comprising:
an electrocardiogram (ECG) monitor configured to perform an ECG of a person and transmit the ECG; and
a computing device configured to:
in response to receiving the ECG from the ECG monitor:
classify rhythm characteristics of the ECG using a first deep neural network (DNN), wherein the first DNN is structured to classify rhythm characteristics of ECG data and wherein a training weight matrix is used to balance a difference in prevalence among classifications during training of the first DNN;
classify morphology characteristics of the ECG using a second DNN, wherein the second DNN is structured to classify morphology characteristics of ECG data; and
identify global parameters of the ECG using a third DNN, wherein the third DNN is structured to identify global parameters of ECG data; and
apply a set of decision logic to the ECG rhythm characteristics and the ECG morphology characteristics to diagnose one or more conditions indicated by the ECG based at least in part on the ECG global parameters.
9 . The system of claim 8 , wherein the processing device is further to:
train the first deep neural network (DNN) to classify electrocardiogram (ECG) rhythm characteristics;
training the second DNN to classify ECG morphology characteristics;
training the third DNN to identify global ECG parameters, wherein both the first and second DNNs comprise a classification layer and the third DNN comprises a regression layer.
10 . The system of claim 8 , wherein:
the classification of the rhythm characteristics of the ECG comprises a probability for each of a set of rhythms; and
the classification of the morphology characteristics of the ECG comprises a probability for each of a set of morphologies.
11 . The system of claim 10 , wherein one or more of the first, second and third DNN comprise convolution neural networks (CNNs).
12 . The system of claim 11 , wherein the processing device is further to:
use a random data shifting technique to further generalize each of the first, second and third DNNs.
13 . The system of claim 8 , wherein the set of decision logic comprises:
a first set of criteria for validating the classification of the rhythm characteristics of the ECG based at least in part on the global parameters of the ECG; and
a second set of criteria for validating the classification of the morphology characteristics of the ECG based at least in part on the global parameters and the validated classification of the rhythm characteristics of the ECG.
14 . The system of claim 8 , wherein each of the first, second and third DNNs comprises a convolution block having a residual block structure that combines a set of convolution layers and a set of skip layers.
15 . A non-transitory computer-readable medium having instructions stored thereon which, when executed by a processing device, cause the processing device to:
train a first deep neural network (DNN) to classify electrocardiogram (ECG) rhythm characteristics, wherein a training weight matrix is used to balance a difference in prevalence among classifications during the training of the first DNN;
train a second DNN to classify ECG morphology characteristics;
train a third DNN to identify global ECG parameters;
in response to receiving an ECG:
classify rhythm characteristics of the ECG using the first DNN, wherein the first DNN is structured to classify rhythm characteristics of ECG data;
classify morphology characteristics of the ECG using the second DNN, wherein the second DNN is structured to classify morphology characteristics of ECG data; and
identify global parameters of the ECG using the third DNN, wherein the third DNN is structured to identify global parameters of ECG data; and
apply a set of decision logic to the ECG rhythm characteristics and the ECG morphology characteristics to diagnose one or more conditions indicated by the ECG based at least in part on the ECG global parameters.
16 . The non-transitory computer-readable medium of claim 15 , wherein:
the classification of the rhythm characteristics of the ECG comprises a probability for each of a set of rhythms; and
the classification of the morphology characteristics of the ECG comprises a probability for each of a set of morphologies.
17 . The non-transitory computer-readable medium of claim 15 , wherein the set of decision logic comprises:
a first set of criteria for validating the classification of the rhythm characteristics of the ECG based at least in part on the global parameters of the ECG; and
a second set of criteria for validating the classification of the morphology characteristics of the ECG based at least in part on the global parameters and the validated classification of the rhythm characteristics of the ECG.
18 . The non-transitory computer-readable medium of claim 17 , wherein one or more of the first, second and third DNN comprise convolution neural networks (CNNs).
19 . The non-transitory computer-readable medium of claim 17 , wherein the processing device is further to:
use a random data shifting technique to further generalize each of the first, second and third DNNs.
20 . The non-transitory computer-readable medium of claim 15 , wherein each of the first, second and third DNNs comprises a convolution block having a residual block structure that combines a set of convolution layers and a set of skip layers.