Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems

Patent metadata

Field	Value
Publication number	US-11617528-B2
Application number	US-201916596640-A
Country	US
Kind code	B2
Filing date	Oct 8, 2019
Priority date	Oct 8, 2019
Publication date	Apr 4, 2023
Grant date	Apr 4, 2023

Abstract

Methods and systems are provided for automatically diagnosing a patient based on a reduced lead electrocardiogram (ECG), using one or more deep neural networks. In one embodiment, a method for automatically diagnosing a patient using a reduced lead ECG comprises, acquiring reduced lead ECG data, wherein the reduced lead ECG data comprises less than twelve lead signals, determining a type of each of the less than twelve lead signals, selecting a deep neural network based on the type of each of the less than twelve lead signals, and mapping the less than twelve lead signals to a diagnosis using the deep neural network. In this way, reduced lead ECG data may be mapped to a diagnosis using an intelligently selected deep neural network, wherein the deep neural network was trained on reduced lead ECG data comprising a same set of ECG lead types as the acquired reduced lead ECG data.

First claim

The invention claimed is:

1. A method comprising:
acquiring reduced lead electrocardiographic data (reduced lead ECG data), wherein the reduced lead ECG data comprises less than twelve lead signals;
determining a type of each of the less than twelve lead signals;
identifying at least one missing type of the less than twelve lead signals based on the determined type of each lead signal;
selecting a deep neural network based on the type of each of the less than twelve lead signals and the identified at least one missing type; and
mapping the less than twelve lead signals to a diagnosis using the deep neural network.

2. The method of claim 1 , wherein the type of each of the less than twelve lead signals is one of I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5, and V6, wherein the type of each of the less than twelve lead signals is distinct from each other type of the less than twelve lead signals, and wherein selecting the deep neural network comprises selecting the deep neural network from among a plurality of possible deep neural networks based on the type of each of the less than twelve lead signals.

3. The method of claim 1 , wherein the deep neural network comprises a convolutional neural network and a decision network, and wherein mapping the less than twelve lead signals to a diagnosis using the deep neural network comprises entering the less than twelve lead signals as input to the convolutional neural network.

4. The method of claim 3 , wherein mapping the less than twelve lead signals to the diagnosis using the deep neural network comprises:
mapping the less than twelve lead signals to a plurality of features using the convolutional neural network; and
mapping the plurality of features from the convolutional neural network to the diagnosis using the decision network.

5. The method of claim 4 , the method further comprising:
determining a preliminary diagnosis based on the less than twelve lead signals using a rule-based system, and wherein mapping the plurality of features from the convolutional neural network to the diagnosis using the decision network further comprises:
inputting the preliminary diagnosis into an input layer of the decision network; and
mapping both the preliminary diagnosis and the plurality of features to the diagnosis using the decision network.

6. The method of claim 1 , wherein the deep neural network is stored in a deep neural network library and indexed based on a training data set used to train the deep neural network.

7. The method of claim 6 , wherein selecting the deep neural network based on the type of each of the less than twelve lead signals comprises matching the type of each of the less than twelve lead signals with an index of the deep neural network and selecting the deep neural network based on the type of each of the less than twelve lead signals matching the index, and wherein the index indicates lead signal types used to train the deep neural network.

8. The method of claim 1 , wherein acquiring reduced lead ECG data comprises measuring the less than twelve lead signals using less than 10 electrodes.

9. A method for training a deep neural network to automatically diagnose reduced lead electrocardiograms (ECGs), the method comprising:
selecting an ECG training data pair comprising a lead signal set and a ground truth diagnosis corresponding to the lead signal set;
selectively removing one or more lead signals from the lead signal set to produce a reduced lead signal set;
determining a lead type of each lead signal in the lead signal set;
identifying at least one missing lead type based on the determined lead type of each lead signal;
feeding the reduced lead signal set to the deep neural network, the deep neural network configured to output a diagnosis based on the reduced lead signal set and the identified at least one missing lead type;
calculating a difference between the diagnosis and the ground truth diagnosis; and
adjusting one or more parameters of the deep neural network based on the calculated difference.

10. The method of claim 9 , the method further comprising:
determining a validation error of the deep neural network; and
responding to the validation error being below a pre-determined threshold by:
storing the deep neural network in a deep neural network library; and
indexing the deep neural network in the deep neural network library based on the removed one or more lead signals.

11. The method of claim 9 , wherein the lead signal set is a twelve lead electrocardiogram of a patient, and wherein the ground truth diagnosis corresponding to the lead signal set comprises an expert generated diagnosis determined for the patient.

12. The method of claim 9 , wherein adjusting one or more parameters of the deep neural network based on the calculated difference comprises performing backpropagation using the calculated difference to adjust a plurality of weights and/or a plurality of biases of the deep neural network.

13. The method of claim 9 , wherein the deep neural network comprises a convolutional neural network coupled in series to a decision network, and wherein mapping the reduced lead signal set to the diagnosis comprises:
mapping the reduced lead signal set to a plurality of features using the convolutional neural network; and
mapping the plurality of features from the convolutional neural network to the diagnosis using the decision network.

14. The method of claim 13 , the method further comprising:
determining a preliminary diagnosis based on the reduced lead signal set using a rule-based system, and wherein mapping the plurality of features from the convolutional neural network to the diagnosis using the decision network further comprises:
inputting the preliminary diagnosis into an input layer of the decision network; and
mapping the preliminary diagnosis and the plurality of features to the diagnosis using the decision network.

15. An electrocardiogram (ECG) processing system comprising:
a display device;
a memory storing a deep neural network library comprising a plurality of trained deep neural networks, and instructions; and
a processor communicably coupled to the memory and the display device and when executing the instructions, configured to:
acquire reduced lead ECG data, wherein the reduced lead ECG data comprises less than twelve lead signals, each lead signal of the less than twelve lead signals comprising measured electrical potential across time;
select a deep neural network from the deep neural network library based on the less than twelve lead signals;
map the less than twelve lead signals to a diagnosis using the deep neural network; and
display the diagnosis via the display device, wherein selecting the deep neural network from the deep neural network library based on the less than twelve lead signals comprises:
determining a lead type of each lead signal of the less than twelve lead signals;
identifying at least one missing lead type based on the determined lead type of each lead signal; and
selecting the deep neural network based on the identified at least one missing lead type.

16. The system of claim 15 , the system further comprising less than 10 electrodes, and wherein the system is configured to acquire reduced lead ECG data using the less than 10 electrodes.

17. The system of claim 15 , wherein the deep neural network comprises a convolutional neural network and a decision network.

18. The system of claim 15 , wherein the processor is configured to map the less than twelve lead signals to the diagnosis using the dee

Assignees

• Ge Prec Healthcare Llc

Inventors

• Yu Long
• Xue Joel Qiuzhen
• Rowlandson Gordon Ian

Classifications

• A61B5/349

  Detecting specific parameters of the electrocardiograph cycle · CPC title

• A61B5/339

  Displays specially adapted therefor · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G16H50/20 Primary

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