Method, program and device for diagnosing thyroid dysfunction based on electrocardiogram

Abstract

According to one embodiment of the present disclosure, there may be provided a method for diagnosing thyroid dysfunction based on an electrocardiogram, the method being executed by a computing device including at least one processor, the method including the steps of acquiring electrocardiogram data and estimating a probability of onset of thyroid dysfunction for a subject measured in the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model, the neural network model being trained based on a correlation between thyroid function and changes in electrocardiogram characteristics.

Description

[Technical field]

[0001] The present disclosure relates to a method for diagnosing thyroid dysfunction, and more particularly to a method for diagnosing thyroid dysfunction based on an electrocardiogram using a neural network model. [Background technology]

[0002] An electrocardiogram (ECG) is a signal that measures the electrical signals generated in the heart and can be used to determine the presence or absence of disease by checking for abnormalities in the conduction system from the heart to the electrodes.

[0003] Heartbeats, which are the cause of electrocardiograms, begin with an impulse that starts in the sinus node in the right atrium, depolarizing the right and left atria, and then after a short delay in the atrioventricular node, activates the ventricles.

[0004] The septum is activated first, and the right ventricle, which has a thinner wall, activates before the left ventricle, which has a thicker wall. The depolarization wave that is transmitted to the Purkinje fibers propagates like a wavefront from the endocardium to the epicardium in the myocardium, causing ventricular contraction. As electrical impulses are normally conducted through the heart, the heart contracts approximately 60 to 100 times per minute. Each contraction is represented by one heart beat.

[0005] Such an electrocardiogram can be detected by bipolar leads, which record the potential difference between two sites, and unipolar leads, which record the potential at the site where an electrode is attached. Methods for measuring an electrocardiogram include standard limb leads, which are bipolar leads, unipolar limb leads, and precordial leads, which are unipolar leads.

[0006] The electrical activity stages of the heart are roughly divided into the periods of atrial depolarization, ventricular depolarization, and ventricular repolarization, and each of these stages is reflected as several waveforms, namely, P, Q, R, S, and T waves, as shown in Figure 1.

[0007] When such waves have a standard morphology, the electrical activity of the heart can be considered normal. To determine whether the wave has a standard morphology, it is necessary to check whether characteristics such as the duration of each wave, the interval between each wave, the amplitude of each wave, and kurtosis are within the normal range.

[0008] Such electrocardiograms are measured using expensive measuring equipment and are used as an auxiliary tool to measure the health condition of patients. Generally, the measuring equipment only displays the measurement results, and diagnosis is entirely the responsibility of the doctor.

[0009] Currently, in order to reduce dependency on doctors, research is being continuously conducted on rapid and accurate disease diagnosis using artificial intelligence based on electrocardiograms. In addition, with the development of wearable and lifestyle electrocardiogram measuring devices, it is becoming increasingly possible to diagnose and monitor not only heart diseases but also various other diseases based on electrocardiograms.

[0010] In particular, in the case of thyroid function-related diseases, early detection is difficult because routine checkups are not performed and the symptoms are not obvious. However, there is an increasing possibility that thyroid dysfunction can be diagnosed early by detecting subtle changes in the electrocardiogram. Summary of the Invention [Problem to be solved by the invention]

[0011] The present disclosure has been devised in response to the above-mentioned background technology, and aims to provide a method for diagnosing thyroid dysfunction according to one embodiment of the present disclosure, which uses a neural network model to estimate the probability of thyroid dysfunction in a subject whose electrocardiogram data is measured, based on the electrocardiogram data.

[0012] However, the problems to be solved by the present disclosure are not limited to those mentioned above, and other problems not mentioned will be clearly understood from the description below. [Means for solving the problem]

[0013] In order to achieve the above-mentioned object, one embodiment of the present disclosure provides a method for diagnosing thyroid dysfunction based on an electrocardiogram, executed by a computing device including at least one processor, the method including: acquiring electrocardiogram data; and estimating a probability of onset of thyroid dysfunction for a subject measured in the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model, wherein the neural network model is trained based on a correlation between thyroid function and changes in electrocardiogram characteristics.

[0014] Alternatively, a method can be provided in which the neural network model includes a first sub-neural network model trained on electrocardiogram data measured on 12 multiple leads.

[0015] Alternatively, a method can be provided in which the neural network model includes a second sub-neural network model trained based on at least one of six limb leads or six precordial leads.

[0016] Alternatively, a method may be provided in which the neural network model includes a third sub-neural network model trained on electrocardiogram data measured in a single lead.

[0017] Alternatively, the neural network model may include a neural network composed of a plurality of residual blocks, and the neural network composed of the residual blocks may provide a method for receiving the electrocardiogram data and outputting the probability of developing overt hyperthyroidism.

[0018] Alternatively, the method can provide that the overt hyperthyroidism is when the free thyroxine level is higher than a pre-determined reference range or the thyroid stimulating hormone level is lower than a pre-determined reference range.

[0019] Alternatively, the neural network model may include a neural network corresponding to each of multiple leads of electrocardiogram data, and the outputs of the neural networks may be concatenated together to derive the probability of occurrence of thyroid dysfunction.

[0020] Alternatively, the correlation between thyroid function and changes in electrocardiogram characteristics can be based on electrocardiogram characteristics including at least one of the following: frequency of tachycardia, length of QT interval, deflection direction of P, R and T waves, or QRS duration.

[0021] Alternatively, a method may be provided in which the probability of developing a thyroid disorder increases with increasing frequency of the tachycardia.

[0022] Alternatively, a method may be provided in which the probability of developing a thyroid dysfunction increases with increasing length of the QT interval.

[0023] Alternatively, a method may be provided in which the probability of developing a thyroid dysfunction increases as the deflection directions of the P, R and T waves move to the right.

[0024] Alternatively, a method may be provided in which the probability of developing a thyroid disorder is higher the shorter the QRS duration.

[0025] Alternatively, a method can be provided in which the step of estimating the probability of thyroid dysfunction for the subject of the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model includes a step of inputting biological data including at least one of age and gender together with the electrocardiogram data into the neural network model, and estimating the probability of thyroid dysfunction for the subject of the electrocardiogram data.

[0026] According to another embodiment of the present disclosure, there may be provided a computer program stored on a computer-readable storage medium, the computer program, when executed by one or more processors, performing operations for diagnosing thyroid dysfunction based on an electrocardiogram, the operations including: acquiring electrocardiogram data; and estimating a probability of thyroid dysfunction for a subject measured based on the electrocardiogram data using a pre-trained neural network model, the neural network model being trained based on a correlation between thyroid function and changes in electrocardiogram characteristics.

[0027] According to yet another embodiment of the present disclosure, there is provided a computing device for diagnosing thyroid dysfunction based on an electrocardiogram, the device including: a processor including at least one core; and a memory including program code executable by the processor, the processor acquiring electrocardiogram data by executing the program code, and estimating a probability of thyroid dysfunction for a subject measured in the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model, the neural network model being trained based on a correlation between thyroid function and changes in electrocardiogram characteristics. Effect of the Invention

[0028] A method for diagnosing thyroid dysfunction according to one embodiment of the present disclosure can provide a method for estimating the probability of thyroid dysfunction for a subject based on electrocardiogram data using a neural network model. [Brief description of the drawings]

[0029] [Figure 1] FIG. 2 illustrates an electrocardiogram signal according to the present disclosure.

[0030] [Diagram 2] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0031] [Diagram 3] 1 is a flow chart illustrating a method for diagnosing thyroid dysfunction based on an electrocardiogram according to one embodiment of the present disclosure.

[0032] [Figure 4] FIG. 2 is a diagram illustrating the structure of a neural network model according to an embodiment of the present disclosure.

[0033] [Diagram 5] FIG. 2 is a diagram illustrating a verification research process of a neural network model according to an embodiment of the present disclosure.

[0034] [Figure 6] FIG. 13 illustrates performance test results of a neural network model according to one embodiment of the present disclosure.

[0035] [Figure 7] FIG. 13 illustrates subgroup electrocardiogram analysis results categorized by gender and age according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0036] Hereinafter, the embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that a person having ordinary skill in the art can easily carry out the present disclosure. The embodiments presented in the present disclosure are provided to enable a person skilled in the art to use or practice the contents of the present disclosure. Thus, various modifications to the embodiments of the present disclosure will be apparent to a person skilled in the art. That is, the present disclosure may be embodied in various different forms and is not limited to the following embodiments.

[0037] Throughout the specification of the present disclosure, the same or similar reference numerals refer to the same or similar components. In addition, in order to clearly describe the present disclosure, reference numerals of parts that are not related to the description of the present disclosure may be omitted from the drawings.

[0038] The term "or" as used in this disclosure is intended to mean inclusive "or" rather than exclusive "or." That is, unless otherwise specified in this disclosure or the context makes the meaning clear, "x uses a or b" should be understood to mean one of the natural inclusive permutations. For example, unless otherwise specified in this disclosure or the context makes the meaning clear, "x uses a or b" can be interpreted as either x uses a, x uses b, or x uses both a and b.

[0039] The term "and / or" as used in this disclosure must be understood to indicate and include all possible combinations of one or more of the associated listed concepts.

[0040] The terms "comprise" and / or "comprising" as used in this disclosure should be understood to mean the presence of the specified features and / or components. However, the terms "comprise" and / or "comprising" should be understood not to exclude the presence or addition of one or more other features, other components and / or combinations thereof.

[0041] In this disclosure, unless otherwise specified or clear from the context as indicating the singular form, the singular should generally be construed as including "one or more."

[0042] The term "nth (n is a natural number)" used in this disclosure can be understood as an expression used to distinguish components of the present disclosure from each other based on a certain criterion such as a functional viewpoint, a structural viewpoint, or convenience of explanation. For example, in this disclosure, components performing different functional roles can be distinguished as a first component or a second component. However, components that are substantially the same within the technical idea of ​​the present disclosure but must be distinguished for convenience of explanation can also be distinguished as a first component or a second component.

[0043] The term "acquire" as used in this disclosure may be understood to mean not only receiving data via a wired or wireless communication network for an external device or system, but also generating data in an on-device form.

[0044] Meanwhile, the term "module" or "unit" used in the present disclosure may be understood as a term indicating an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a part thereof, hardware or a part thereof, or a combination of software and hardware. Here, a "module" or "unit" may be a unit composed of a single element, or a unit expressed as a combination or set of multiple elements. For example, as a concept of agreement, a "module" or "unit" may indicate a hardware element or a set of hardware elements of a computing device, an application program that performs a specific function of software, a process implemented by the execution of software, or a set of commands for the execution of a program. In addition, as a broad concept, a "module" or "unit" may indicate a computing device itself that constitutes a system, or an application executed by a computing device. However, the above concept is merely an example, and the concept of a "module" or "unit" may be variously defined within a scope that can be understood by a person skilled in the art based on the contents of the present disclosure.

[0045] The term "model" as used in this disclosure may be understood as a system implemented using mathematical concepts and language to solve a particular problem, a set of software units for solving a particular problem, or an abstract model of a process for solving a particular problem. For example, a neural network "model" may refer to a system implemented as a neural network having a problem-solving ability through learning. Here, a neural network may have a problem-solving ability by optimizing parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network, or may include a neural network set in which multiple neural networks are combined.

[0046] "Data" as used in this disclosure may include "image", signals, etc. The term "image" as used in this disclosure may refer to multi-dimensional data made up of discrete image elements. In other words, "image" may be understood as a term that refers to a digital representation of an object that can be viewed by the human eye. For example, "image" may refer to multi-dimensional data made up of elements that correspond to pixels in a two-dimensional image. "Image" may refer to multi-dimensional data made up of elements that correspond to voxels in a three-dimensional image.

[0047] The term "block" used in this disclosure may be understood as a collection of configurations classified according to various criteria such as type, function, etc. Thus, the configurations classified into one "block" may vary depending on the criteria. For example, a neural network "block" may be understood as a collection of neural networks including at least one neural network. Here, it may be assumed that the neural networks included in a neural network "block" perform a particular operation equally.

[0048] The above explanations of the terms are intended to facilitate understanding of the present disclosure. Therefore, unless the above terms are explicitly described as matters limiting the contents of the present disclosure, care should be taken to ensure that they are not used to limit the technical ideas of the contents of the present disclosure.

[0049] FIG. 2 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0050] The computing device 100 according to an embodiment of the present disclosure may be a hardware device or a part of a hardware device that performs comprehensive data processing and calculations, or may be a software-based computing environment connected via a communication network. For example, the computing device 100 may be a server that performs intensive data processing functions and shares resources, or a client that shares resources by interacting with the server. The computing device 100 may also be a cloud system in which multiple servers and clients interact with each other to comprehensively process data. The above description is merely an example related to the type of the computing device 100, and the type of the computing device 100 may be variously configured within a scope that can be understood by a person skilled in the art based on the contents of the present disclosure.

[0051] Referring to Fig. 2, a computing device 100 according to an embodiment of the present disclosure may include a processor 110, a memory 120, and a network unit 130. However, Fig. 2 is merely an example, and the computing device 100 may include other components for implementing a computer environment. Also, only some of the disclosed components may be included in the computing device 100.

[0052] The processor 110 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for performing computing operations. For example, the processor 110 may read a computer program to perform data processing for machine learning. The processor 110 may process operations such as input data processing for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor 110 for performing such data processing may include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application specific integrated circuit (ASICc), or a field programmable gate array (FPGA). The above-mentioned types of the processor 110 are merely examples, and the types of the processor 110 may be variously configured within a scope that can be understood by a person skilled in the art based on the contents of the present disclosure.

[0053] The processor 110 can train a neural network model for diagnosing thyroid dysfunction based on medical data. For example, the processor 110 can train the neural network model to estimate the presence or absence of hyperthyroidism and the degree of progression based on biological data including information such as gender and age together with electrocardiogram data. Specifically, the processor 110 can train the neural network model by inputting electrocardiogram data and various biological data to the neural network model so that the neural network model detects changes in the electrocardiogram due to the onset of hyperthyroidism. Here, the neural network model can perform training based on the correlation between thyroid function and electrocardiogram changes. The correlation between thyroid function and electrocardiogram changes can be understood as information on the association between changes in thyroid function and morphological changes in the electrocardiogram signal. The processor 110 can execute a calculation to represent at least one neural network block included in the neural network model during the training process of the neural network model.

[0054] The processor 110 can estimate the occurrence of thyroid dysfunction based on medical data using the neural network model generated by the above-mentioned learning process. The processor 110 can generate inference data indicating the result of estimating the occurrence probability of thyroid dysfunction in humans by inputting electrocardiogram data and biological data including information on age and sex to the neural network model trained by the above-mentioned process. For example, the processor 110 can input electrocardiogram data to the neural network model that has completed learning to predict the occurrence of hyperthyroidism and the degree of progression. The processor 110 can accurately predict the occurrence of thyroid dysfunction by effectively grasping subtle electrocardiogram changes that are difficult for humans to interpret through such a neural network model for diagnosing thyroid dysfunction.

[0055] In addition to the above examples, the types of medical data and the output of the neural network model can be configured in a variety of ways within the scope of what one skilled in the art can understand based on the contents of this disclosure.

[0056] The memory 120 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for storing and managing data processed by the computing device 100. That is, the memory 120 may store any type of data generated or determined by the processor 110 and any type of data received by the network unit 130. For example, the memory 120 may include at least one type of storage medium among a flash memory type, a hard disk type, a multimedia card micro type, a card type memory, a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk. The memory 120 may also include a database system that controls and manages data in a predetermined system. The above-mentioned types of the memory 120 are merely examples, and the types of the memory 120 may be variously configured within a range that can be understood by a person skilled in the art based on the contents of the present disclosure.

[0057] The memory 120 may structure and organize and manage data, a combination of data, and program code executable by the processor 110, which are necessary for the processor 110 to perform calculations. For example, the memory 120 may store medical data received via the network unit 130, which will be described later. The memory 120 may store program code for operating a neural network model to receive medical data and perform learning, program code for operating the neural network model to receive medical data and perform inference according to the purpose of use of the computing device 100, and processed data generated by executing the program code.

[0058] The network unit 130 according to an embodiment of the present disclosure may be understood as a component that transmits and receives data via any type of known wired or wireless communication system. For example, the network unit 130 may transmit and receive data using wired or wireless communication systems such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), 5G, ultrawide-band, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity, near field communication (NFC), or Bluetooth. The above-mentioned communication systems are merely examples, and wired or wireless communication systems for transmitting and receiving data by the network unit 130 may be applied in various ways other than the above examples.

[0059] The network unit 130 may receive data necessary for the processor 110 to perform the calculation through wired or wireless communication with any system or any client. The network unit 130 may also transmit data generated by the calculation of the processor 110 through wired or wireless communication with any system or any client. For example, the network unit 130 may receive medical data through communication with a database in a hospital environment, a cloud server performing work such as standardization of medical data, or a computing device. The network unit 130 may transmit output data of the neural network model, and intermediate data and processed data derived in the calculation process of the processor 110 through communication with the above-mentioned database, server, computing device, etc.

[0060] FIG. 3 is a flow chart illustrating a method for diagnosing thyroid dysfunction based on an electrocardiogram according to one embodiment of the present disclosure.

[0061] Referring to FIG. 3, a method for diagnosing thyroid dysfunction based on an electrocardiogram, executed by a computing device 100 including at least one processor 110, may first perform a step (S100) of acquiring electrocardiogram data.

[0062] The electrocardiogram data can be directly acquired by measuring the electrocardiogram using an electrocardiogram measuring device, or can be acquired from the electrocardiogram measuring device via network communication.

[0063] Thereafter, a step (S110) of estimating the probability of onset of thyroid dysfunction for the subject of measurement of the electrocardiogram data based on the electrocardiogram data using the pre-trained neural network model can be performed.

[0064] The estimation step (S110) may include a step of inputting biological data including at least one of age and sex together with the electrocardiogram data into the neural network model, and estimating the probability of developing thyroid dysfunction for the subject of the electrocardiogram data measurement.

[0065] Here, the neural network model may be trained based on the correlation between thyroid function and changes in electrocardiogram characteristics. The neural network model may also be trained based on the correlation between thyroid function and changes in characteristics such as electrocardiogram, gender, and age. Specifically, the neural network model may be trained based on the correlation between the onset and progression of hyperthyroidism and changes in electrocardiogram and other characteristics. The neural network model can be used to diagnose various thyroid dysfunctions, such as not only hyperthyroidism but also hypothyroidism.

[0066] The neural network model may be trained based on a 12-lead electrocardiogram obtained from electrodes of an electrocardiogram measuring device connected to the human body. As an example, the electrocardiogram may be measured over 12 leads for 10 seconds and stored at 500 points per second. Additionally, the neural network model may be trained based on partial information extracted from only 6 limb lead electrocardiograms and a lead I electrocardiogram from the 12-lead electrocardiogram.

[0067] Referring to FIG. 4, a diagram showing the structure of a neural network model according to one embodiment of the present disclosure.

[0068] 4, a neural network model according to an embodiment of the present disclosure may include a neural network composed of a plurality of residual blocks. The neural network composed of the residual blocks may receive electrocardiogram data and output the probability of developing overt hyperthyroidism.

[0069] Here, overt hyperthyroidism can be diagnosed when the free thyroxine level is higher than a predetermined reference range or the thyroid stimulating hormone level is lower than a predetermined reference range.

[0070] Specifically, the neural network model may have a ResNet neural network structure using six residual blocks. Each residual block may be composed of a convolutional neural network (CNN), batch normalization, a rectified linear unit (ReLU) function, and a dropout layer. The convolutional neural network may be set to one dimension, and the filter size may be set to 21. The input length may be reduced by half each time it passes through three residual blocks out of a total of six residual blocks. A different neural network may be applied to each lead of the electrocardiogram. Average pooling may be applied per channel at the end of the residual block. The outputs of the neural network may be concatenated together to derive the probability of thyroid dysfunction.

[0071] The neural network model may include neural networks corresponding to each of a plurality of leads of electrocardiogram data, i.e., the neural network model may include separate neural networks input with respective electrocardiograms measured in separate leads.

[0072] For example, the neural network model may include a first sub-neural network model trained based on electrocardiogram data measured in 12 multiple leads. The neural network model may further include a second sub-neural network model trained based on at least one of six limb leads or six precordial leads. The neural network model may further include a third sub-neural network model trained based on electrocardiogram data measured in a single lead. The neural network model may selectively use at least one of the first sub-neural network model, the second sub-neural network model, and the third sub-neural network model depending on the number of leads. Thus, the neural network model may effectively predict the onset of thyroid dysfunction regardless of the number of leads. In addition, when 12-lead electrocardiogram data is input, the neural network model may use all of the first sub-neural network model, the second sub-neural network model, and the third sub-neural network model, and combine the outputs of the respective sub-models to output the onset probability of thyroid dysfunction. Through such combinations, the neural network model can improve the accuracy of predicting the onset of thyroid dysfunction.

[0073] The following describes the statistical analysis methods carried out to verify the neural network model with the above structure. To confirm basic characteristics, continuous variables were presented as mean and standard deviation. Verification results were compared using the unpaired student's t-test or the Mann-Whitney u-test. Categorical variables were expressed as percentages, and the chi-square test was used.

[0074] The performance of the neural network model was verified by comparing the probability calculated by the model and the presence or absence of hyperthyroidism in the internal / external validation dataset. Validation was performed by referring to the area under the receiver operating characteristic curve (AUC). In the training dataset, the cutoff point was identified using Youden J statistic. The cutoff point was applied to calculate the sensitivity, specificity, positive predictive value, and negative predictive value in the internal / external validation dataset. The 95% confidence interval of the AUC was calculated using Sun&Su's optimization of the De-Long method.

[0075] In the following, a validation study of a neural network model according to one embodiment of the present disclosure is described.

[0076] To verify the robustness of the neural network model, we performed sensitivity analyses by creating subgroups according to age and gender: gender was categorized as male and female, and age was categorized as <40, 40–50, 50–60, 60–70, and 70 or older.

[0077] We hypothesized that subtle changes in electrocardiograms may occur during the pre-hyperthyroidism period, and that the neural network model could detect these small changes and predict the onset of the disease. To confirm this, subgroup analysis was performed. The external validation dataset was extracted from patients who had a normal initial thyroid function test (TFT) and subsequently underwent a subsequent thyroid function test. The time interval between the initial and subsequent thyroid function tests was ≥4 weeks. Based on the probability of onset of hyperthyroidism estimated by the neural network model, the study subjects were classified into high-risk and low-risk groups. The cutoff point was determined using the Juden J statistic on the training dataset. The Kaplan-Meier method was used to analyze the 36-month outcomes.

[0078] FIG. 5 is a diagram illustrating a verification research process of a neural network model according to an embodiment of the present disclosure.

[0079] Referring to FIG. 5, the subjects of the validation study of the neural network model according to one embodiment of the present disclosure were 113,215 patients from Hospital A and 33,485 patients from Hospital B. 21 patients from Hospital A and 7 patients from Hospital B whose clinical information or electrocardiogram data was leaked were excluded. A total of 2,164 hyperthyroid patients were included. For neural network training, 139,521 electrocardiogram data measured in 90,554 patients from Hospital A were used. For internal validation, 34,810 electrocardiogram data measured in 518 patients from Hospital A were used. For external validation, 48,684 electrocardiogram data measured in 33,478 patients from Hospital B were used. The basic characteristics of the neural network model training cohort (Hospital A, n=113,175) and the external validation cohort (Hospital B, n=33,478) are shown in Table 1 below.

[0080] [Table 1]

[0081] In Table 1, the alternative hypothesis for p denoted with † is that there is a difference between hyperthyroidism and overt hyperthyroidism. The alternative hypothesis for p-values ​​denoted with ‡ is that there is a difference between Hospital A (model development and internal validation data group) and Hospital B (external validation group) for each variable. Gender, age, and incidence of hyperthyroidism showed statistically significant differences between hospitals. Hyperthyroid patients had relatively more tachycardia and longer QT intervals. Hyperthyroid patients showed right-sided deflection of P, R, and T wave axes and shorter QRS duration.

[0082] In summary, the correlation between thyroid function and changes in electrocardiogram characteristics described above may be based on electrocardiogram characteristics including at least one of the frequency of tachycardia, the length of the QT interval, the deflection direction of P waves, R waves and T waves, or the QRS duration.

[0083] The probability of thyroid dysfunction estimated by the neural network model may be higher with a higher frequency of tachycardia. The probability of thyroid dysfunction may be higher with a longer QT interval.

[0084] The probability of developing thyroid dysfunction estimated by the neural network model may be higher as the deviation direction of P waves, R waves, and T waves moves toward the right.

[0085] The probability of developing thyroid dysfunction estimated by the neural network model may be higher when the QRS duration is shorter.

[0086] FIG. 6 illustrates a performance test result of a neural network model according to one embodiment of the present disclosure.

[0087] Referring to Figure 6, in the internal validation and external validation, AUC means the area under the receiver operating characteristic curve of the neural network model, DLM means the neural network model, ECG means the electrocardiogram, NPV means the negative predictive value, PPV means the positive predictive value, SEN means the sensitivity, and SPE means the specificity.

[0088] In internal and external validation, the AUC of the neural network model using 12-lead ECG was 0.918 (0.909-0.927) and 0.897 (0.879-0.916), respectively. Sensitivity analysis confirmed the robustness of the neural network model with respect to gender and age. Those identified by the neural network model as high-risk patients showed a significant change in the onset of hyperthyroidism compared to those identified as low-risk patients (p<0.01). The performance of the neural network model using 6-lead ECG and single-lead ECG can also be confirmed. Referring to Table 2, in the sensitivity analysis for gender and age, the performance of all models was AUC value 0.830 or higher.

[0089] [Table 2]

[0090] 7 is a diagram showing the results of subgroup electrocardiogram analysis classified by gender and age according to an embodiment of the present disclosure. Referring to FIG. 7, the subgroup analysis was performed using the subsequent thyroid function test results of 6,762 patients who were determined to be normal in the thyroid function test. Of these, 24 patients developed hyperthyroidism. The subjects of the subgroup analysis were divided into a high-risk group of 4,749 patients and a low-risk group of 2,013 patients according to the probability of developing hyperthyroidism output by the neural network model. It was confirmed that the high-risk group had a significantly higher risk of developing hyperthyroidism than the low-risk group (0.48% vs. 0.05%, p<0.01).

[0091] Meanwhile, thyroid function is closely related to cardiovascular disease, and it can affect not only cardiac function, vascular resistance, and cardiovascular autonomic control, but also the cardiovascular system. In particular, thyroid hormone-mediated changes can affect cardiac function through enhanced relaxation function, inotropy, and heart rate chronotropy. Signs and symptoms of thyroid dysfunction can be determined to be the result of thyroid hormone affecting the heart and cardiovascular relationship. Thyroid dysfunction can be associated with increased cardiovascular disease morbidity and mortality. Furthermore, untreated hyperthyroidism was associated with a higher risk of cardiovascular disease than treated hyperthyroidism. In both treated and untreated hyperthyroidism, cardiovascular morbidity increased with the duration of decreased thyroid-stimulating hormone levels.

[0092] Therefore, in order to detect and predict such risky thyroid dysfunction early, a thyroid dysfunction diagnosis method according to one embodiment of the present disclosure uses a neural network model to estimate the probability of developing thyroid dysfunction for a subject whose electrocardiogram data is measured based on the electrocardiogram data.

[0093] Furthermore, the method for diagnosing thyroid dysfunction according to an embodiment of the present disclosure has the effect of being able to diagnose hyperthyroidism using a neural network model based on information such as an electrocardiogram, gender, and age.

[0094] The various embodiments of the present disclosure described above can be combined with additional embodiments and can be modified within the scope that a person skilled in the art can understand from the above detailed description. The embodiments of the present disclosure are illustrative in all respects and should be understood as not being limiting. For example, each component described as a single type can be implemented in a distributed form, and similarly, components described as distributed can be implemented in a combined form. Therefore, all modifications or variations derived from the meaning, scope, and equivalent concept of the claims of the present disclosure should be interpreted as being included in the scope of the present disclosure. [Explanation of symbols]

[0095] 100: Computing equipment 110: Processor 120: Memory 130: Network Department

Claims

1. 1. A method for diagnosing thyroid dysfunction based on an electrocardiogram, the method being performed by a computing device including at least one processor, the method comprising: acquiring electrocardiogram data; and estimating a probability of onset of thyroid dysfunction for the subject of the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model, the neural network model is trained based on a correlation between thyroid function and changes in electrocardiogram characteristics; The neural network model is configured to have a resnet structure including a plurality of residual blocks for inputting the electrocardiogram data and extracting features; The method, wherein the neural network model includes a first sub-neural network model trained based on electrocardiogram data measured in 12 multiple leads.

2. 2. The method of claim 1, wherein the neural network model includes a second sub-neural network model trained based on at least one of six limb leads or six precordial leads.

3. The method of claim 1 , wherein the neural network model includes a third sub-neural network model trained based on electrocardiogram data measured in a single lead.

4. The neural network model includes a neural network composed of a plurality of residual blocks, The method according to claim 1 , wherein the neural network formed by the residual blocks receives the electrocardiogram data and outputs a probability of onset of overt hyperthyroidism.

5. 5. The method of claim 4, wherein the overt hyperthyroidism is when the free thyroxine level is higher than a predetermined reference range or the thyroid stimulating hormone level is lower than a reference range.

6. the neural network model includes a neural network corresponding to each of a plurality of leads of electrocardiogram data; 2. The method of claim 1, wherein the outputs of the neural networks are concatenated together to derive a probability of occurrence of thyroid dysfunction.

7. 2. The method of claim 1, wherein the correlation between thyroid function and changes in electrocardiogram characteristics is based on electrocardiogram characteristics including at least one of frequency of tachycardia, length of QT interval, deflection direction of P waves, R waves and T waves, or QRS duration.

8. The method according to claim 7 , wherein the probability of developing the thyroid dysfunction increases as the frequency of the tachycardia increases.

9. The method according to claim 7, wherein the probability of developing the thyroid dysfunction increases as the length of the QT interval increases.

10. 8. The method according to claim 7, wherein the probability of developing a thyroid dysfunction increases as the deviation directions of the P waves, R waves and T waves move toward the right.

11. The method of claim 7 , wherein the probability of developing a thyroid dysfunction is higher the shorter the QRS duration.

12. The step of estimating a probability of onset of thyroid dysfunction for the subject of the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model includes: inputting biological data including at least one of age and sex together with the electrocardiogram data into the neural network model to estimate a probability of onset of thyroid dysfunction for the subject of the electrocardiogram data; The method of claim 1 , comprising:

13. A computer program stored on a computer-readable storage medium, comprising: The computer program, when executed by one or more processors, performs operations for diagnosing thyroid dysfunction based on an electrocardiogram, The operation includes: acquiring electrocardiogram data; and estimating a probability of onset of thyroid dysfunction for the subject of the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model, the neural network model is trained based on a correlation between thyroid function and changes in electrocardiogram characteristics; The neural network model is configured to have a resnet structure including a plurality of residual blocks for inputting the electrocardiogram data and extracting features; The neural network model includes a first sub-neural network model trained based on electrocardiogram data measured in 12 multiple leads.

14. 1. A computing device for electrocardiogram-based thyroid dysfunction diagnosis, comprising: A processor including at least one core; a memory containing program code executable by the processor; Including, The processor executes the program code to acquire electrocardiogram data, and estimates a probability of thyroid dysfunction for the subject of the electrocardiogram data based on the electrocardiogram data using a pre-trained neural network model; the neural network model is trained based on a correlation between thyroid function and changes in electrocardiogram characteristics; The neural network model is configured to have a resnet structure including a plurality of residual blocks for inputting the electrocardiogram data and extracting features; The apparatus, wherein the neural network model includes a first sub-neural network model trained based on electrocardiogram data measured on 12 multiple leads.

Images