KR20220033083A - System for diagnosing and explaining cardiac disorder based on explainable artificial intelligence deep learning - Google Patents

Abstract

Disclosed is an explainable artificial intelligence (XAI) deep learning-based heart disease diagnosis and analysis system, which comprises: an XAI deep learning model unit (110) configured to extract a diagnosis name of heart disease and characteristics of the diagnosed heart disease by including one or more corresponding analysis modules using heart disease and electrocardiogram data, which determine heart disease, as learning data and analyze one or more characteristics for determining the diagnosis of heart disease; an input unit (120) receiving electrocardiogram data to provide the received data the XAI deep learning model unit (110);, and an output unit (130) using the electrocardiogram data from the input unit (120) by the XAI deep learning model unit (110) as input to output the diagnosis name of heart disease and the diagnosis characteristics of the diagnosed heart disease. Accordingly, the system can provide diagnosis of heart disease and diagnosis reasons by an explainable neural network model.

Description

XAI deep learning-based heart disease diagnosis and interpretation system {SYSTEM FOR DIAGNOSING AND EXPLAINING CARDIAC DISORDER BASED ON EXPLAINABLE ARTIFICIAL INTELLIGENCE DEEP LEARNING}

The present invention relates to an XAI deep learning-based heart disease diagnosis and interpretation system that can be applied to clinical practice by increasing the accuracy and interpretability of diagnosis by providing a diagnosis and diagnosis reason for heart disease by an explanatory neural network model.

As is well known, since the development of the electrocardiogram, electrocardiogram-related knowledge has expanded exponentially, and it is possible to obtain information on the electrical function of the heart from electrocardiography and to diagnose various heart diseases such as arrhythmias, coronary artery disease, and myocardial disease.

On the other hand, research on AI algorithms for electrocardiography is being actively conducted, and the AI algorithm detects heart failure, predicts atrial fibrillation among arrhythmic rhythms, or determines gender.

In this way, by overcoming the limitations of humans, it is possible to detect subtle changes in the ECG waveform by the AI algorithm, and further improve the ECG interpretation.

Atrial fibrillation is a major cause of increased morbidity and mortality, and timely diagnosis and treatment are important. AI algorithms are used to improve the performance of diagnostic algorithms using biosignals such as ECG and oxygen saturation curves, but AI algorithms cannot be understood by humans. As a black box that does not explain predictions in a possible way, there is a problem with low reliability due to poor transparency and interpretation.

In addition, in the conventional deep learning structure, since the neural network is a unit, there is a limitation in that only the contents of judgment such as the presence or absence of a heart disease, the type of heart disease, or the degree of heart disease can be confirmed.

Therefore, the deep learning model cannot be trusted due to the lack of transparency and interpretability, so a technology is required to build an explanatory artificial intelligence model with improved accuracy and interpretability that can present the diagnosis name and reason for heart disease together. .

Korean Patent Publication No. 10-2156087 (Public atrial fibrillation prediction method in normal sinus rhythm electrocardiogram using deep learning, 2020.09.15) Korean Patent Publication No. 10-2021-0066322 (Atrial fibrillation prediction system and prediction method, 2021.06.07)

The technical task to be achieved by the spirit of the present invention is XAI deep learning-based heart disease diagnosis, which can be applied to clinical practice by increasing the accuracy and interpretability of diagnosis by providing a diagnosis and diagnosis reason for heart disease by an explanatory neural network model and to provide an analysis system.

In order to achieve the above object, an embodiment of the present invention uses one or more corresponding analysis modules that use heart disease and electrocardiogram data for determining heart disease as learning data, and interpret one or more characteristics for determining a diagnosis name of heart disease. The XAI deep learning model unit, which is provided to derive the diagnosis name of heart disease and the judgment characteristics of the diagnosed heart disease; an input unit that receives electrocardiogram data and provides it to the XAI deep learning model unit; and an output unit for outputting, by the XAI deep learning model unit, the electrocardiogram data from the input unit and outputting a diagnosis name of a heart disease and a judgment characteristic of the diagnosed heart disease; XAI deep learning-based heart disease diagnosis and An analysis system is provided.

Here, the heart disease is atrial fibrillation, and the analysis module may interpret the determination characteristic of the atrial fibrillation.

In addition, the analysis module includes an irregularity analysis module for analyzing the irregularity of the heartbeat from the electrocardiogram data, a P wave analysis module for analyzing the absence of a P wave, and combining the output values of the irregularity analysis module and the P wave analysis module to It may consist of an ensemble module that determines atrial fibrillation.

In addition, the irregularity analysis module or the P-wave analysis module includes a first residual block to a fourth residual block, and a first FC layer and a second FC layer, and the first residual block to the fourth residual block includes two convolutions. A layer and two batch normalization layers are each repeated, and the fourth residual block is connected to a platen layer, the platen layer is connected to the first FC layer, and the second FC layer is connected to a single output node. and the single output node outputs, by a sigmoid activation function, a first probability of irregularity of heartbeat and a second probability of absence of a P wave, respectively, and the ensemble module outputs the first probability and the second probability, respectively. By combining the probabilities and outputting the probability of the presence of atrial fibrillation, the presence of the atrial fibrillation may be predicted, and the predicted judgment characteristics of the atrial fibrillation may be interpreted.

Meanwhile, the electrocardiogram data may be single-guided, 6-lead, or 12-lead electrocardiogram data.

In addition, the input unit may include a pre-processing module for normalizing and removing noise of the electrocardiogram data.

In addition, the verification data of the XAI deep learning model unit may consist of internal verification data extracted at a certain rate from the same diagnostic group as the training data, and external verification data extracted from a diagnostic group different from the training data.

In addition, the output unit may provide a treatment method corresponding to the judgment characteristic of the diagnosed heart disease.

In addition, the XAI deep learning model unit may be an NBDT model.

According to the present invention, it is possible to increase the accuracy and interpretability of diagnosis by providing the diagnosis and diagnosis reason for heart disease by means of an explanatory neural network model, so that it can be applied to clinical practice, and to increase the accuracy of the model by utilizing internal and external verification data. There is an effect that can provide a corresponding treatment according to the diagnosis reason, and can diagnose various heart diseases other than atrial fibrillation using electrocardiogram data.

1 shows a schematic configuration diagram of an XAI deep learning-based heart disease diagnosis and analysis system according to an embodiment of the present invention.
FIG. 2 illustrates an XAI deep learning model unit of the XAI deep learning-based heart disease diagnosis and interpretation system of FIG. 1 .
FIG. 3 illustrates a dataset table of electrocardiogram data for building the XAI deep learning model unit of FIG. 2 .
4 is a flowchart illustrating a diagnosis and interpretation process by the XAI deep learning-based heart disease diagnosis and interpretation system of FIG. 1 .

Hereinafter, embodiments of the present invention having the above-described characteristics with reference to the accompanying drawings will be described in more detail.

XAI deep learning-based heart disease diagnosis and interpretation system according to an embodiment of the present invention uses electrocardiogram data to determine heart disease and heart disease as learning data, and interprets one or more characteristics for determining a diagnosis name of heart disease XAI deep learning model unit 110, which is built to derive the diagnosis name of heart disease and the judgment characteristics of the diagnosed heart disease by having the above analysis module, receives the electrocardiogram data and provides it to the XAI deep learning model unit 110 By the input unit 120 and the XAI deep learning model unit 110, the electrocardiogram data from the input unit 120 is input to the output unit 130 for outputting the diagnosis name of heart disease and the judgment characteristics of the diagnosed heart disease. It aims to provide a diagnosis and diagnostic reason for heart disease by an explanatory neural network model, including

Hereinafter, with reference to FIGS. 1 to 4 , the XAI deep learning-based heart disease diagnosis and analysis system of the above configuration will be described in detail as follows.

First, the XAI deep learning model unit 110 uses, as training data, electrocardiogram (ECG) data stored digitally among the biosignals for determining heart disease and heart disease as training data, and the diagnosis name of heart disease. It has one or more corresponding analysis modules that interpret one or more characteristics to be determined, and is constructed to derive a diagnosis name of a heart disease and a judgment characteristic of a diagnosed heart disease, You may be asked to provide evidence together.

Here, the XAI deep learning model unit 110 is used as training data for model building or validation data for model validation. ,VF) or 12-lead ECG data, various types of ECG data can be used.

In addition, the verification data of the XAI deep learning model unit 110 is composed of internal verification data extracted at a certain rate from the same diagnostic group as the training data, and external verification data extracted from a diagnostic group different from the training data. The data can be used to verify the accuracy of the model by solving the overfitting problem.

For example, referring to <Table 1> of FIG. 3, the PTB-XL ECG data set including 21,837 12-lead ECGs with a sampling rate of 400 Hz and 10,605 12-lead ECGs with a sampling rate of 500 Hz are included. The Chapman ECG data set is exemplified, the PhysioNet ECG data set containing 8,528 single-lead ECGs with a sampling rate of 300 Hz, and the Sejong ECG data set containing 128,399 12-lead ECGs with a sampling rate of 500 Hz are illustrated. When building the model in the example, training data and internal validation data are selected by randomly dividing the population of Sejong ECG data set at a ratio of 9: 1, and PTB-XL ECG data set and Chapman ECG data are selected as external validation data. Sets and PhysioNet ECG data sets can be selected.

In this embodiment, the XAI deep learning model unit 110 is a neural network and a decision tree combined NBDT ( It may be a Neural-Backed Decision Tree) model.

On the other hand, the heart disease is atrial fibrillation, and the analysis module is a module for classifying the characteristics of atrial fibrillation, and may interpret the judgment characteristics of the criterion, reason, or basis for diagnosing atrial fibrillation.

For reference, atrial fibrillation is an arrhythmic disease that causes irregular heartbeat in the form of a fast pulse due to the loss of regular contraction of the atrium and irregular tremor. Normal P-waves do not appear, only irregular lines.

Accordingly, in order to identify the irregularity of the heartbeat and the absence of the P wave, which are the judgment characteristics of atrial fibrillation, the analysis module, referring to FIG. 2 , the irregularity analysis module 111 that analyzes the irregularity of the heartbeat from the electrocardiogram data, the P wave A P-wave analysis module 112 that analyzes the absence of , and an ensemble module 113 that combines the output values of the irregularity analysis module 111 and the P-wave analysis module 112 to determine the presence probability of atrial fibrillation can be composed of

Specifically, the irregularity analysis module 111 or the P-wave analysis module 112 includes a first residual block to a fourth residual block that makes learning easier, respectively, and convolution and max pooling ( Max Pooling) may include a first FC layer (Fully Connected layer) and a second FC layer for classifying the ECG data image into a defined label by taking the result of the process.

Here, as shown enlarged in FIG. 2 , the first to fourth residual blocks have two convolution layers (CONV) and the possibility of performing learning faster or falling into a local optimum problem. Two batch normalization layers (BN) for reducing ?

In addition, the fourth residual block is connected to a platen layer that transforms into a single vector, the platen layer is connected to the first FC layer, the second FC layer is connected to a single output node, and a single output node is, The first probability of irregular heartbeat and the second probability of the absence of P wave may be respectively output by a sigmoid activation function that determines whether or not activation is performed according to the magnitude of the weighted sum of the previous layer. .

Finally, the ensemble module 113 may combine the first probability and the second probability to output the presence probability of atrial fibrillation, predict the presence of atrial fibrillation, and interpret the predicted atrial fibrillation judgment characteristics.

On the other hand, the XAI deep learning model unit 110 is not limited to the aforementioned atrial fibrillation, but is applied to various heart diseases such as cardiomyopathy, congenital heart disease, valve disease, heart failure, pericardial disease, hypertension, arteriosclerosis or coronary artery disease. Thus, by extracting and analyzing the shape of the electrocardiogram data that provides a sensory signal such as 2D diagnostic data, 3D diagnostic data, or waveform, N characteristics for diagnosing heart disease are extracted by an auto encoder, and the characteristics of By learning the corresponding diagnosis name of heart disease according to the combination, it is also possible to derive a diagnosis name which is a prediction result according to electrocardiogram data, which is a predictive variable.

Next, the input unit 120 receives the electrocardiogram data for predicting the diagnosis name and judgment characteristics of heart disease and provides it to the XAI deep learning model unit 110 .

Meanwhile, the input unit 120 may include a pre-processing module 121 for normalizing and removing noise from the electrocardiogram data.

For example, the pre-processing module 121 removes artifacts by removing the start and end portions of each of the 12-guide ECG data digitally stored for 10 seconds by 1 second to secure better quality ECG data.

Next, the output unit 130 receives the electrocardiogram data, which is a predictive variable, from the input unit 120 by the XAI deep learning model unit 110 as an input, and determines the diagnosis name of heart disease and the diagnosed heart disease as the prediction result. By outputting the characteristics, the possibility of the diagnosis name and the criterion, reason or basis for judging the diagnosis name are confirmed.

That is, the output unit 130 presents whether the heart beat is regular or the P wave is present together with the result of diagnosing that there is no atrial fibrillation, or whether the P wave is irregular with the result of diagnosing that there is atrial fibrillation. It can be presented because it is absent.

In addition, the output unit 130 may provide a treatment corresponding to the judgment characteristics of the diagnosed heart disease, for example, irregular heartbeat or absence of a P wave.

4 is a flowchart illustrating a diagnosis and interpretation process by the XAI deep learning-based heart disease diagnosis and interpretation system of FIG.

First, in the XAI deep learning model unit construction step (S110), digitally stored electrocardiogram data among biosignals determining heart disease and heart disease are used as learning data, and one or more characteristics for determining the diagnosis name of heart disease are analyzed By constructing the XAI deep learning model unit 110 to derive the diagnosis name of heart disease and the judgment characteristics of the diagnosed heart disease by having one or more corresponding analysis modules that Alternatively, the basis for judgment may be presented together.

Here, as training data for model building or validation data for model validation, various formats such as single-lead (lead I), 6-lead (I, II, III, VL, VR, VF) or 12-lead ECG data are used. ECG data may be used.

In addition, the verification data consists of internal verification data extracted at a certain rate from the same diagnostic group as the training data, and external verification data extracted from a different diagnostic group than the training data, so that the overfitting problem is solved through the external verification data. It can be used to verify the accuracy of the model.

On the other hand, the heart disease is atrial fibrillation, and the analysis module is a module for classifying the characteristics of atrial fibrillation, and may interpret the judgment characteristics of the criterion, reason, or basis for diagnosing atrial fibrillation.

For reference, atrial fibrillation is an arrhythmic disease that causes irregular heartbeat in the form of a fast pulse due to loss of regular contraction of the atrium and irregular tremor. Normal P-waves do not appear, only irregular lines.

Accordingly, in order to identify the irregularity of the heartbeat and the absence of the P wave, which are the judgment characteristics of atrial fibrillation, the analysis module, referring to FIG. 2 , the irregularity analysis module 111 that analyzes the irregularity of the heartbeat from the electrocardiogram data, the P wave It can be composed of a P-wave analysis module 112 that analyzes the absence of there is.

Specifically, the irregularity analysis module 111 or the P-wave analysis module 112 takes the results of the first residual block to the fourth residual block, convolution and max pooling process to make learning easier, respectively, to obtain an electrocardiogram data image It may include a first FC layer and a second FC layer for classifying as a defined label.

Here, as shown enlarged in FIG. 2 , in the first residual block to the fourth residual block, two convolution layers and two batch normalization layers to perform learning faster or to reduce the possibility of falling into a micro-optimal solution problem are each It can be constructed repeatedly.

In addition, the fourth residual block is connected to the platen layer that transforms into a single vector, the platen layer is connected to the first FC layer, the second FC layer is connected to a single output node, and the single output node is for the previous layer. The first probability of irregularity of the heartbeat and the second probability of the absence of the P wave may be respectively output by the sigmoid activation function that determines whether the activity is active according to the magnitude of the weighted sum.

Finally, the ensemble module 113 may combine the first probability and the second probability to output the presence probability of atrial fibrillation, predict the presence of atrial fibrillation, and interpret the predicted atrial fibrillation judgment characteristics.

On the other hand, the XAI deep learning model unit 110 is not limited to the aforementioned atrial fibrillation, but is applied to various heart diseases such as cardiomyopathy, congenital heart disease, valve disease, heart failure, pericardial disease, hypertension, arteriosclerosis or coronary artery disease. Therefore, by analyzing the electrocardiogram data, extracting N characteristics for diagnosing heart disease, learning the corresponding diagnosis name of heart disease according to the combination of characteristics, and deriving a diagnosis name that is a prediction result according to the electrocardiogram data, which is a predictive variable there is.

Thereafter, in the test data input step ( S120 ), ECG data, which is a predictive variable for predicting the diagnosis name and judgment characteristics of heart disease, is received and provided to the XAI deep learning model unit 110 .

Meanwhile, the test data input step ( S120 ) may further include a pre-processing step ( S121 ) of normalizing and removing noise of the electrocardiogram data by the pre-processing module 121 .

For example, in the pre-processing step S121, each of the start and end portions of the 12-guide electrocardiogram data stored digitally for 10 seconds are removed by 1 second to remove artifacts to secure better quality ECG data.

Then, in the prediction result output step (S130), the XAI deep learning model unit 110 receives the ECG data, which is the predictive variable, from the input unit 120 as input, and determines the irregularity of the heartbeat and the absence of the P wave. , by outputting the diagnosis name of the heart disease and the judgment characteristics of the diagnosed heart disease, which are the prediction results, to confirm the possibility of the diagnosis name, and the criterion, reason, or basis for judging by the diagnosis name.

That is, in the prediction result output step ( S130 ), it is suggested whether the heartbeat is regular or the P wave is present together with the diagnosis result of the absence of atrial fibrillation, or whether the heartbeat is irregular with the diagnosis result of atrial fibrillation. It can be suggested that the P wave is absent.

Thereafter, in the treatment providing step (S140), through the output unit 130, it is possible to provide a treatment corresponding to the judgment characteristics of a diagnosed heart disease such as atrial fibrillation, for example, irregular heartbeat or absence of P wave. .

In addition, it is possible to provide computer software stored in a computer-readable recording medium for performing the diagnosis and interpretation method by the aforementioned XAI deep learning-based heart disease diagnosis and interpretation system.

Therefore, by the configuration of the XAI deep learning-based heart disease diagnosis and interpretation system as described above, it is possible to apply to clinical practice by providing the diagnosis and diagnosis reason for heart disease by an explanatory neural network model to increase the accuracy and interpretability of the diagnosis. It is possible to increase the accuracy of the model by using internal and external verification data, to provide a corresponding treatment according to the diagnosis reason, and to diagnose various heart diseases other than atrial fibrillation using electrocardiogram data.

Configurations shown in the embodiments and drawings described in this specification are only the most preferred embodiment of the present invention, and do not represent all of the technical spirit of the present invention, so various equivalents that can be substituted for them at the time of the present application It should be understood that there may be water and variations.

110: XAI deep learning model unit 111: irregularity analysis module
112: P wave analysis module 113: ensemble module
120: input unit 121: pre-processing module
130: output unit
S110: XAI deep learning model part construction stage
S120: Test data input step
S121: pre-processing step
S130: Prediction result output step
S140: Treatment provision stage

Claims (9)

Using the electrocardiogram data for determining heart disease and heart disease as learning data, and having one or more corresponding analysis modules for interpreting one or more characteristics for determining the diagnosis name of heart disease, determining the diagnosis name of the heart disease and the diagnosed heart disease Built to derive characteristics, XAI deep learning model unit;
an input unit that receives electrocardiogram data and provides it to the XAI deep learning model unit; and
An output unit for outputting, by the XAI deep learning model unit, the electrocardiogram data from the input unit, and outputting a diagnosis name of a heart disease and a judgment characteristic of the diagnosed heart disease;
XAI deep learning-based heart disease diagnosis and interpretation system.
According to claim 1,
The heart disease is atrial fibrillation, characterized in that the analysis module interprets the judgment characteristics of the atrial fibrillation,
XAI deep learning-based heart disease diagnosis and interpretation system.
3. The method of claim 2,
The analysis module includes an irregularity analysis module for analyzing the irregularity of the heartbeat from the electrocardiogram data, a P wave analysis module for analyzing the absence of a P wave, and combining the output values of the irregularity analysis module and the P wave analysis module to obtain the atrial fibrillation. characterized in that it consists of an ensemble module to determine
XAI deep learning-based heart disease diagnosis and interpretation system.
4. The method of claim 3,
The irregularity analysis module or the P wave analysis module includes a first residual block to a fourth residual block, and a first FC layer and a second FC layer,
In the first to fourth residual blocks, two convolutional layers and two batch normalization layers are each repeatedly configured, the fourth residual block is connected to a platen layer, and the platen layer is the first FC layer and the second FC layer is connected to a single output node, and the single output node outputs, by a sigmoid activation function, a first probability of heartbeat irregularity and a second probability of absence of a P wave, respectively. and,
The ensemble module outputs the presence probability of atrial fibrillation by combining the first probability and the second probability,
Predicting the presence of the atrial fibrillation and interpreting the predicted judgment characteristics of the atrial fibrillation,
XAI deep learning-based heart disease diagnosis and interpretation system.
According to claim 1,
The electrocardiogram data is characterized in that it is single-guided, 6-lead or 12-lead electrocardiogram data,
XAI deep learning-based heart disease diagnosis and interpretation system.
According to claim 1,
The input unit, characterized in that it comprises a pre-processing module for normalizing and removing noise of the electrocardiogram data,
XAI deep learning-based heart disease diagnosis and interpretation system.
According to claim 1,
The verification data of the XAI deep learning model unit is characterized in that it consists of internal verification data extracted at a certain rate from the same diagnostic group as the training data, and external verification data extracted from a different diagnostic group than the training data,
XAI deep learning-based heart disease diagnosis and interpretation system.
According to claim 1,
The output unit is characterized in that it provides a treatment corresponding to the judgment characteristics of the diagnosed heart disease,
XAI deep learning-based heart disease diagnosis and interpretation system.
According to claim 1,
The XAI deep learning model unit, characterized in that the NBDT model,
XAI deep learning-based heart disease diagnosis and interpretation system.

Images