KR102483149B1 - Carotid artery stenosis diagnosis method using machine learning - Google Patents

Abstract

The present invention relates to a method for providing information for determining carotid artery stenosis using machine learning, wherein auscultation sound data of an electronic stethoscope is divided into a plurality of segment data, and the divided data is input to a machine learning neural network to generate noise of each data. (bruit) signals are determined, and the determined data are calculated as quantitative index values to be used for diagnosing carotid artery stenosis, thereby solving the existing limitation of relying only on the auditory judgment of experts and at low cost. It also has the effect of enabling precise and rapid diagnosis of carotid artery stenosis.

Description

Information provision method for carotid artery stenosis diagnosis using machine learning {Carotid artery stenosis diagnosis method using machine learning}

The present invention relates to a method for providing information for determining carotid artery stenosis using machine learning, and more particularly, overcomes the limitation of analyzing the sound of blood flow in the carotid artery through auscultation by an expert, and converts the auditory signal into data-based quantitative data. It relates to a method for providing information for determining carotid artery stenosis using improved machine learning so that the presence or absence of carotid artery stenosis can be automatically diagnosed through an analysis method.

Stroke incidence is closely related to carotid artery stenosis. To diagnose this, techniques such as magnetic resonance vascular imaging (MRA), computed tomography angiography (CTA), or medical ultrasound (ultrasonography) are used to confirm stenosis by visualization through images. In addition, by taking advantage of the fact that blood flow sounds caused by carotid artery stenosis are different from normal blood flow sounds, experts diagnose stenosis through auscultation.

Methods for visually diagnosing stenosis through imaging are expensive and cumbersome because they require periodic visits to hospitals for treatment. In addition, in the case of magnetic resonance vascular imaging and computed tomography angiography, side effects such as vomiting and dizziness occur when a contrast agent that enhances image visibility is administered, and radiation is exposed during imaging, which adversely affects the human body. .

In contrast, auscultation can diagnose stenotic symptoms in a relatively inexpensive and safe manner through a simple method. However, since there is no clear criterion related to the diagnosis of stenosis, it is determined by the expert's subjective judgment based on auditory stimulation. In addition, there is great difficulty in distinguishing between normal and abnormal blood flow sounds because there is a large variation in blood flow sound for each person.

Korean Registered Patent Publication No. 10-1178867 Korean Registered Patent Publication No. 10-2123147 Japanese Patent Registration No. 6073799

The present invention has been made by the above necessity, and an object of the present invention is to determine carotid artery stenosis using machine learning, which enables accurate and rapid diagnosis of carotid artery stenosis without a doctor's examination or expensive tests. It is intended to provide a method of providing information for

In order to achieve the above object, a method for providing information for determining carotid artery stenosis using machine learning according to the present invention is used for diagnosing carotid artery stenosis using a machine learning neural network learned through deep learning. obtaining stethoscope sound data from the subject's carotid artery using a stethoscope; The data processor converts the stethoscope sound data into signal values in the time-frequency domain by applying a short-time Fourier transform, divides the time data of the signal value into a plurality of pieces, and converts the signal values in a specific frequency band corresponding to the divided time domain. extracting to form a plurality of segment data; determining, by a determination unit, the presence or absence of bruit by inputting the plurality of segment data to the machine learning neural network; calculating, by a calculation unit, an index value that is a ratio of the number of segment data with noise to the total number of segment data; and a diagnosis unit comparing the index value and the reference value and utilizing the result for diagnosing carotid artery stenosis.

The machine learning neural network is preferably a support vector machine (SVM) neural network.

The reference value includes a first reference value greater than 0 and less than 1 and a second reference value greater than the first reference value and less than 1, and when the calculated index value is less than or equal to the first reference value, it is diagnosed that there is no carotid artery stenosis. And if it is more than the second reference value, it is diagnosed as having carotid artery stenosis, and if it exceeds the first reference value and is less than the second reference value, it is preferable to diagnose carotid artery stenosis through a separate precise diagnosis.

Preferably, each set of data overlaps a part of the divided time domain.

Preferably, each segment data is information about signals corresponding to a frequency band of 50-300 Hz for 2 seconds, and an offset between segment data adjacent to each other in the time domain is 0.3 seconds.

In the method for providing information for determining carotid artery stenosis using machine learning according to the present invention having the configuration as described above, the stethoscope sound data of the electronic stethoscope is divided into a plurality of segment data, and the divided data is applied to the machine learning neural network. It is configured to determine whether each data has a bruit signal, calculate the determined data as a quantitative index value, and use it to diagnose the presence or absence of carotid artery stenosis. It can solve the limitations and has the effect of enabling precise and rapid diagnosis of carotid artery stenosis at low cost.

1 is a block diagram illustrating the logic flow of a method for providing information for determining carotid artery stenosis using machine learning according to an embodiment of the present invention.
Figure 2 is a block diagram for explaining the device configuration for the implementation of one embodiment of the present invention.
3 is a diagram showing a signal converted into a time-frequency domain through a short-time Fourier transform of a sound wave in a stethoscope sound data acquisition step employed in an embodiment of the present invention.
4 is a diagram for explaining a process of obtaining a spectrogram segment from a signal acquired through short-time Fourier transform in a segment data forming step employed in an embodiment of the present invention;
5 is a diagram showing results determined through a noise presence determination step employed in an embodiment of the present invention.
6 is a view for explaining a process of diagnosing carotid artery stenosis using an index value calculation step and a diagnosis utilization step employed in an embodiment of the present invention.

In order to clarify the understanding of the present invention in the following description, descriptions of known techniques for the features of the present invention will be omitted. The following examples are detailed descriptions to aid understanding of the present invention, and it will be natural that they do not limit the scope of the present invention. Therefore, equivalent inventions that perform the same functions as the present invention will also fall within the scope of the present invention.

And, in the following description, the same identification symbol means the same configuration, and unnecessary redundant descriptions and descriptions of known technologies will be omitted. In addition, the description of each embodiment of the present invention below, which overlaps with the description of the background technology of the present invention, will also be omitted.

Hereinafter, a method for providing information for determining carotid artery stenosis using machine learning according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram for explaining the logic flow of a method for providing information for determining carotid artery stenosis using machine learning according to an embodiment of the present invention, and FIG. 2 describes a device configuration for implementation of an embodiment of the present invention. Figure 3 is a diagram showing a signal converted into a time-frequency domain through a short-time Fourier transform of a sound wave in the stethoscope sound data acquisition step employed in one embodiment of the present invention, and Figure 4 is a diagram showing one embodiment of the present invention It is a diagram for explaining the process of obtaining a spectrogram segment from a signal obtained through short-time Fourier transform in the segment data formation step employed in the example, and FIG. 6 is a diagram illustrating a process of diagnosing carotid artery stenosis using the indicator value calculation step and diagnosis utilization step employed in an embodiment of the present invention.

As well shown in FIGS. 1 and 2, a method for providing information for determining carotid artery stenosis using machine learning according to an embodiment of the present invention uses a machine learning neural network learned through deep learning to detect carotid artery stenosis. It is used for diagnosis, and includes a stethoscope sound data acquisition step (S1), segment data formation step (S2), bruit presence/absence determination step (S3), index value calculation step (S4), and diagnosis utilization step (S5). It is done by

In the stethoscope sound data acquisition step (S1), a process of obtaining auscultation sound data from the subject's carotid artery by the data acquisition unit 1 including the electronic stethoscope is performed.

When measuring a stethoscope sound using the electronic stethoscope, since the patient's breathing sound affects the analysis, it is preferable to recommend that the person being measured to hold his breath, and the accuracy of the analysis can be improved as the ambient noise is reduced. An electronic stethoscope with a sampling rate of 4000 Hz that stores 4000 data can be used.

In the segment data forming step (S2), the data processing unit 2 applies a short-time Fourier transform to the stethoscope sound data to convert it into a signal value in the time-frequency domain as shown in FIG. 3, and in FIG. 4 As shown, a process of forming a plurality of segment data by dividing the time data of the signal value into a plurality of pieces and then extracting the signal value of a specific frequency band corresponding to the divided time domain is performed.

Here, the data processing unit 2 includes a known program for short-time Fourier transform, and utilizes the program to enable formation of the segment data as well as signal values in the time-frequency domain.

Of course, the parameters for the short-time Fourier transform can be set to various values, but in this embodiment, the window size is set to 0.5s and the offset is set to 0.025s, and the frequency resolution of the signal generated through this transformation is set to 2Hz and time resolution. was set to 0.025 s.

In addition, each segment data may be formed to have various time values and frequency bands by the data processing unit, but in this embodiment, it is formed to have a signal value corresponding to a frequency band of 50-300 Hz for 2 seconds. It became. In addition, it is preferable that offset times between segment data adjacent to each other in the time domain are 0.2 seconds corresponding to 10% of the 2 seconds and 0.4 seconds corresponding to 20%. 0.3 seconds is employed in this embodiment.

In the noise determination step (S3), a process of determining the presence or absence of noise by inputting the plurality of segment data to the machine learning neural network is performed. That is, the determination unit 4 including the machine learning neural network determines whether or not bruit signal information is included in each segment data based on the control signal of the control unit 3 through a pre-learned machine learning neural network. will do

Here, the machine learning neural network is one of various well-known machine learning devices, and in this embodiment, a support vector machine (SVM) is employed. The support vector machine is a classifier that is trained by setting segment data for other objects for constructing a neural network that are not a diagnosis object as learning data.

In the indicator value calculation step (S4), the calculation unit 5 calculates an indicator value that is the ratio of the number of segment data with noise to the total number of segment data based on the control signal of the control unit 3. process is performed.

For example, as shown in FIG. 5, when it is determined that 78 of the 80 pieces of segment data, excluding 2, contain bruit signal information through the pre-learned machine learning neural network in the determination unit, Based on the number of nonbruits, the index value corresponds to 2÷80, 0.025, and based on the number of bruit, the index value corresponds to 0.975.

In the diagnostic utilization step (S5), the diagnosis unit 6 compares the index value with the reference value and uses it to diagnose the presence or absence of carotid artery stenosis.

That is, the reference value is a value derived through learning of the machine learning neural network, and is a preset value for diagnosing carotid artery stenosis according to the presence or absence of noise compared to an index value of a subject to be diagnosed.

In the method for providing information for determining carotid artery stenosis using machine learning according to an embodiment of the present invention having such a configuration, a stethoscope sound data of an electronic stethoscope is divided into a plurality of segment data, and the divided data is divided into a machine learning neural network. It is configured to determine whether each data has a bruit signal, calculate the determined data as a quantitative index value, and use it to diagnose the presence or absence of carotid artery stenosis. It can solve the limitations of the method and has the effect of enabling precise and rapid diagnosis of carotid artery stenosis at low cost.

Preferably, the reference value includes a first reference value (β) greater than 0 and less than 1 and a second reference value (α) greater than the first reference value (β) and less than 1. The first reference value means a minimum value for diagnosing stenosis when a bruit is detected, and the second reference value means a maximum value for normal diagnosis when a bruit is not detected.

For example, as shown in FIG. 6, if the calculated index value is less than the first reference value, it is diagnosed as having no carotid artery stenosis, and if it is greater than or equal to the second reference value, it is diagnosed as having carotid artery stenosis. Of course, it can be configured to diagnose carotid artery stenosis through a separate precise diagnosis when the reference value is greater than the second reference value.

Preferably, each set of data overlaps a portion of the divided time domains so that data of all time domains can be secured without omission.

Although various embodiments of the present invention have been described above, this embodiment and the accompanying drawings only clearly show some of the technical ideas included in the present invention, and are included in the specification and drawings of the present invention. It will be apparent that all modified examples and specific embodiments that can be easily inferred by those skilled in the art within the scope of the technical idea are included in the scope of the present invention.

1: data acquisition unit
2: Data Processing Department

Claims (5)

It is used to determine carotid artery stenosis using a machine learning neural network learned through deep learning.
obtaining, by a data acquisition unit, stethoscope sound data from the carotid artery of the subject using an electronic stethoscope;
The data processing unit converts the stethoscope sound data into signal values in the time-frequency domain by applying short-time Fourier transform, divides the time data of the signal value into a plurality of pieces, and signals in a specific frequency band corresponding to the divided time domain. extracting values to form a plurality of segment data;
determining, by a determination unit, the presence or absence of bruit by inputting the plurality of segment data to the machine learning neural network;
calculating, by a calculator, an index value that is a ratio of the number of segment data with bruit to the total number of segment data; and
A diagnostic unit comparing the index value and the reference value and utilizing them to diagnose the presence or absence of carotid artery stenosis;
The reference value includes a first reference value greater than 0 and less than 1 and a second reference value greater than the first reference value and less than 1,
The first reference value is a minimum value for diagnosis of stenosis when noise is detected, and the second reference value is a maximum value for normal diagnosis when noise is not detected.
The diagnostic unit diagnoses that there is no carotid artery stenosis when the calculated index value is less than or equal to the first reference value, and diagnoses that there is carotid artery stenosis if it is greater than or equal to the second reference value, and exceeds the first reference value and is less than the second reference value. A method for providing information for determining carotid artery stenosis using machine learning, characterized in that it enables diagnosis of carotid artery stenosis through a separate precise diagnosis. According to claim 1,
The machine learning neural network is a support vector machine (SVM) neural network. delete According to claim 1,
The information providing method for determining carotid artery stenosis using machine learning, characterized in that each set data overlaps a part of the divided time domain. According to claim 4,
Each of the segment data is information about signals corresponding to a frequency band of 50-300 Hz for 2 seconds,
An information providing method for determining carotid artery stenosis using machine learning, characterized in that the offset between segment data adjacent to each other in the time domain is 0.025 seconds.

Images