Method for providing information for the prediction of left ventricular outflow tract obstruction in hypertrophic cardiomyopathy

Abstract

The present disclosure relates to a method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, and in particular, the method may include: a step of acquiring an echocardiogram image of a patient and performing preprocessing; a step of inputting the preprocessed echocardiogram image to an artificial intelligence model that is pre-trained to take a single echocardiogram image as input, extract spatial structural features and temporal motion features of the myocardium, either individually or in combination, and predict whether there is left ventricular outflow tract obstruction based on the extracted features and output a prediction result; and a step of generating prognosis evaluation information for the patient based on the prediction result when a prediction result based on the preprocessed echocardiogram image is output from the artificial intelligence model.

Description

Technology Field

[0001] The present disclosure relates to a method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy. Background Technology

[0002] Hypertrophic cardiomyopathy (HCM) is a relatively common hereditary cardiomyopathy worldwide, reported to occur in approximately 1 in 200 to 500 people. Clinically, HCM manifests with a wide range of symptoms, from asymptomatic cases to dyspnea, chest pain, and syncope, and in some patients, it serves as a major cause of sudden cardiac death. In particular, the presence and severity of left ventricular outflow tract obstruction (LVOTO) are known to be key factors determining a patient's symptoms and prognosis.

[0003] Currently, clinical evaluation is primarily performed using transthoracic echocardiography (TTE), often accompanied by multiple echocardiographic views, Doppler examinations, and provocation tests such as the Valsalva maneuver. However, such evaluations require a high level of expertise, and their accuracy can vary depending on patient cooperation and the examination environment; furthermore, their application is limited in community hospitals or emergency situations.

[0004] In addition, while existing studies have proposed echocardiography analysis techniques using artificial intelligence, most have focused on specific diseases such as aortic valve stenosis and diastolic function evaluation, and techniques that accurately reflect left ventricular outflow tract occlusion dynamics, which are important for predicting the prognosis of patients with hypertrophic cardiomyopathy, in a single image have not been sufficiently presented.

[0005] Therefore, there is a steady demand for assistive technologies that automate the prediction and prognosis assessment of left ventricular outflow tract obstruction, which is critical in the clinical management of patients with hypertrophic cardiomyopathy, and that can be immediately utilized in real clinical settings.

[0006] The background art is provided to facilitate understanding of the present disclosure. It should not be understood as an acknowledgment that the matters described in the background art exist as prior art. The problem to be solved

[0007] The prognosis of patients with hypertrophic cardiomyopathy varies significantly depending on the presence and severity of left ventricular outflow tract obstruction. However, conventional diagnosis relies on multiechocardiography and Doppler measurements, which makes it difficult to apply in environments with low proficiency or limited resources, and has resulted in inconsistency in prognostic assessment.

[0008] Furthermore, existing AI-based echocardiography analysis studies have focused on specific disease classification or single-indicator analysis rather than prognosis prediction based on easily obtainable single images, which has limited the clinical prognosis prediction of patients with hypertrophic cardiomyopathy. Consequently, early screening of high-risk patients and monitoring of treatment responses have been restricted, and personalized prognosis management has also been limited.

[0009] Therefore, there is a need for a new technological means to automatically predict the presence of left ventricular outflow tract obstruction and the resulting patient prognosis using a single echocardiogram image of a patient with hypertrophic cardiomyopathy. In particular, a system is required that can precisely analyze spatial and temporal myocardial motion characteristics through an artificial intelligence model and utilize this for patient-specific prognosis assessment and tracking of treatment response.

[0010] Accordingly, the inventors of the present disclosure recognized the need for an artificial intelligence-based prognosis evaluation framework capable of quantitatively calculating the probability of left ventricular outflow tract obstruction using a single echocardiogram image as input, predicting the clinical prognosis of patients with hypertrophic cardiomyopathy based on the results, and further reflecting changes before and after treatment.

[0011] Accordingly, the problem to be solved by the present disclosure is to provide a method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, which evaluates the presence of left ventricular outflow tract obstruction and the patient's prognosis by acquiring echocardiogram images of a patient with hypertrophic cardiomyopathy and analyzing them with an artificial intelligence model, and enables the results to be utilized for patient management and treatment decisions.

[0012] The problems of the present disclosure are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0013] To solve the problem described above, a method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy is provided according to one embodiment of the present disclosure. The method may include: a step of acquiring an echocardiogram image of a patient and performing preprocessing; a step of inputting the preprocessed echocardiogram image to an artificial intelligence model that is pre-trained to take a single echocardiogram image as input, extract spatial structural features and temporal motion features of the myocardium, either individually or in combination, and predict whether there is left ventricular outflow tract obstruction based on the extracted features and output a prediction result; and a step of generating prognosis evaluation information for the patient based on the prediction result when a prediction result based on the preprocessed echocardiogram image is output from the artificial intelligence model.

[0014] According to the features of the present disclosure, the step of performing the preprocessing may be to perform at least one of noise removal, brightness correction, contrast correction, image resolution normalization, region of interest (ROI) extraction, time interval segmentation, or inter-frame alignment on the echocardiogram image.

[0015] According to the features of the present disclosure, the artificial intelligence model may include a first neural network for extracting spatial structural features from the preprocessed echocardiogram image and a second neural network for extracting temporal motion features from the frame sequence of the preprocessed echocardiogram image.

[0016] According to the features of the present disclosure, the first neural network is implemented as a 3D Convolutional Neural Network (3D-CNN), and the second neural network may be a 2D Convolutional Neural Network (2D-CNN) or a Transformer-based network configured to analyze time-axis changes of the myocardium using automatically generated M-mode representations.

[0017] According to the features of the present disclosure, the second neural network may automatically track the myocardial motion trajectory within the echocardiogram image to generate a multi-level M-mode feature map, and embed the multi-level M-mode feature map to improve the accuracy of left ventricular outflow tract closure prediction.

[0018] According to the features of the present disclosure, the artificial intelligence model may predict whether the left ventricular outflow tract is closed by combining the embedding result output from the second neural network with the output of the first neural network.

[0019] According to the features of the present disclosure, the prediction result may include at least one of a binary classification result classifying the presence or absence of left ventricular outflow tract obstruction, or a continuous value (score) that quantitatively predicts the left ventricular outflow tract pressure gradient.

[0020] According to the features of the present disclosure, the prediction result is calculated as a prediction index (DLI-LVOTO) that quantifies the degree of left ventricular outflow tract obstruction, and the prediction index may be calculated based on at least one of the contractile speed, travel distance, or timing difference of the myocardium in the preprocessed echocardiogram.

[0021] According to the features of the present disclosure, the step of generating the prognosis evaluation information may be to calculate a prognosis grade or risk level by combining the prediction result with at least one clinical information (age, gender, heart rate, drug administration history, etc.) for the patient, and to output a treatment response prediction result according to the calculated prognosis grade or risk level.

[0022] To solve the problem described above, an information providing device for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure is provided. The device comprises a communication interface; a memory; and a processor operably connected to the communication interface and the memory. The processor may be configured to acquire an echocardiogram image of a patient and perform preprocessing, extract spatial structural features and temporal motion features of the myocardium, either individually or in combination, using a single echocardiogram image as input, and input the preprocessed echocardiogram image to an artificial intelligence model that is pre-trained to predict whether there is left ventricular outflow tract obstruction based on the extracted features and output a prediction result, and when a prediction result based on the preprocessed echocardiogram image is output from the artificial intelligence model, generate prognosis evaluation information for the patient based on the prediction result.

[0023] Specific details of other embodiments are included in the description of the invention and the drawings. Effects of the invention

[0024] The present disclosure enables the evaluation of left ventricular outflow tract obstruction and prognosis by acquiring echocardiogram images of a patient with hypertrophic cardiomyopathy and analyzing them with an artificial intelligence model, and allows the results to be utilized for patient management and treatment decisions.

[0025] The effects according to the present disclosure are not limited to those exemplified above, and a wider variety of effects are included within the present disclosure. Brief explanation of the drawing

[0026] FIG. 1 is a schematic diagram illustrating an information providing system for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure. FIG. 2 is a block diagram showing the configuration of a service server that performs information provision for predicting left ventricular outflow tract occlusion in hypertrophic cardiomyopathy according to one embodiment of the present disclosure. FIG. 3 is a block diagram showing the configuration of a user terminal receiving prognosis evaluation information according to one embodiment of the present disclosure. FIG. 4 is a flowchart schematically illustrating a method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure. FIGS. 5A and 5B are schematic diagrams illustrating a pipeline for an artificial intelligence model according to one embodiment of the present disclosure to predict whether there is left ventricular outflow tract obstruction based on a single echocardiogram image. FIGS. 6a to 6c schematically illustrate the learning and evaluation process of an artificial intelligence model for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure. FIGS. 7a and 7b are drawings illustrating an example of a prediction point-based M-mode generation process and a sampling guide application in an artificial intelligence model according to one embodiment of the present disclosure. FIG. 8 is a diagram showing the structure of an artificial intelligence model according to an M-mode sampling method according to one embodiment of the present disclosure. FIGS. 9a and 9b are drawings showing the results of a performance evaluation using internal and external verification datasets of an artificial intelligence model according to one embodiment of the present disclosure. FIG. 10 is a diagram showing the results of a visual comparison based on whether the left ventricular outflow tract is blocked, as predicted by an artificial intelligence model according to one embodiment of the present disclosure. FIG. 11 is a diagram showing an example of interpretation of an M-mode image generated by an artificial intelligence model according to one embodiment of the present disclosure using a prediction point-based sampling guide. FIG. 12 is a diagram showing an example of tracking the response to Mavacamten treatment in a patient with obstructive hypertrophic cardiomyopathy using an artificial intelligence model according to one embodiment of the present disclosure. Specific details for implementing the invention

[0027] The advantages and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the description of the present disclosure is complete and to fully inform those skilled in the art of the scope of the invention, and the present disclosure is defined only by the scope of the claims. In connection with the description of the drawings, similar reference numerals may be used for similar components.

[0028] In this document, expressions such as "have," "can have," "include," or "can include" refer to the existence of the relevant feature (e.g., numerical values, functions, actions, or components, etc.) and do not exclude the existence of additional features.

[0029] In this document, expressions such as “A or B,” “at least one of A or / and B,” or “one or more of A or / and B” may include all possible combinations of items listed together. For example, “A or B,” “at least one of A and B,” or “at least one of A or B” may refer to cases including (1) at least one A, (2) at least one B, or (3) both at least one A and at least one B.

[0030] Expressions such as "first," "second," "first," or "second" used in this document may modify various components regardless of order and / or importance, and are used merely to distinguish one component from another without limiting such components. For example, the first user device and the second user device may represent different user devices regardless of order or importance. For example, without departing from the scope of rights set forth in this document, the first component may be named the second component, and similarly, the second component may be renamed the first component.

[0031] Where it is stated that a certain component (e.g., a first component) is "(operatively or communicatively) coupled with" or "connected to" another component (e.g., a second component), it should be understood that the said certain component may be directly connected to the said other component or connected through another component (e.g., a third component). On the other hand, where it is stated that a certain component (e.g., a first component) is "directly connected" or "directly connected" to another component (e.g., a second component), it may be understood that no other component (e.g., a third component) exists between the said certain component and the said other component.

[0032] As used in this document, the expression “configured to” may be replaced, depending on the context, with, for example, “suitable for,” “having the capacity to,” “designed to,” “adapted to,” “made to,” or “capable of.” The term “configured to” does not necessarily mean only “specifically designed to” in hardware. Instead, in some situations, the expression “device configured to” may mean that the device is “capable of” in conjunction with other devices or components. For example, the phrase “processor configured to perform A, B, and C” may mean a dedicated processor for performing those operations (e.g., an embedded processor) or a generic-purpose processor (e.g., a CPU or application processor) capable of performing those operations by executing one or more software programs stored in a memory device.

[0033] The terms used in this document are used merely to describe specific embodiments and are not intended to limit the scope of other embodiments. Singular expressions may include plural expressions unless the context clearly indicates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as generally understood by those skilled in the art described in this document. Terms used in this document that are defined in general dictionaries may be interpreted as having the same or similar meaning as they have in the context of the relevant technology, and are not to be interpreted in an ideal or overly formal sense unless explicitly defined in this document. In some cases, even terms defined in this document may not be interpreted to exclude the embodiments of this document.

[0034] The features of each of the various embodiments of the present disclosure may be combined or combined with one another, either partially or wholly, and as will be fully understood by those skilled in the art, various technical interlocking and operation are possible, and each embodiment may be implemented independently of one another or together in an interlocking relationship.

[0035] For clarity in the interpretation of this specification, the terms used in this specification are defined below.

[0036] The device referred to as a “service server” below is a device according to the present disclosure and may mean a physically independent server, but is not limited thereto; it may be a virtual machine, and is intended to encompass all modules, programs, or Docker, etc., operating on a virtual or physical machine.

[0037] Hereinafter, the present disclosure will be described in detail by describing preferred embodiments of the present disclosure with reference to the attached drawings.

[0038] FIG. 1 is a schematic diagram illustrating an information providing system for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure.

[0039] Referring to FIG. 1, an information providing system (1000) (hereinafter referred to as the "information providing system") for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure may be a system configured to automatically predict whether there is left ventricular outflow tract obstruction (LVOTO) and the prognosis of the patient based on the evaluative results, by taking an echocardiographic image of a specific patient as input, and to provide prognosis evaluation information based on the evaluation results. Accordingly, the information providing system (1000) may include a service server (100), a user terminal (200), and a medical information providing device (300).

[0040] The service server (100) may be in the form of a web server or a cloud server and refers to a device that performs the reception and preprocessing of echocardiogram images, artificial intelligence analysis, calculation of prediction results, and generation of prognosis evaluation information in the information providing system (1000) of the present disclosure. The service server (100) stores and manages prediction results, prognosis evaluation information, and prediction indicators (e.g., DLI-LVOTO) obtained during the analysis process in a database, and can transmit and receive various data in real time by linking with a user terminal (200) and / or a medical information providing device (300).

[0041] To this end, the service server (100) may be equipped with at least one artificial intelligence model. In this case, the artificial intelligence model may be configured to extract spatial structural features and temporal motion features of the myocardium from a single echocardiogram image, either individually or in combination, and to calculate the presence (binary classification) or degree (continuous value / score) of left ventricular outflow tract obstruction based on the extracted features. In this case, the training data may be obtained from a hospital PACS, electronic medical records (EMR), or a research database, and the predictive performance of the model may be maintained and improved through periodic updates.

[0042] Additionally, the service server (100) can calculate a prediction index (Deep Learning-derived LVOT obstruction index, DLI-LVOTO) by comparing the prediction results with actual clinical data and store it in a database for use in long-term model verification and performance evaluation. Furthermore, the service server (100) can calculate a prognosis grade or risk level for each patient by combining the prediction results with clinical information such as the patient's age, gender, heart rate, and drug administration history, and transmit the results in real time to a user terminal (200) to support the decision-making of medical staff.

[0043] Meanwhile, the service server (100) can improve the prediction accuracy of the artificial intelligence model by performing periodic learning updates, and can maintain the reliability of the prediction model by continuously collecting and correcting new clinical data. This service server (100) can be implemented as a cloud server, a web server for medical data processing, or a local hospital server, and can support medical staff in quickly evaluating the patient's prognosis and establishing a treatment plan.

[0044] In addition, the service server (100) can calculate performance evaluation indicators such as accuracy, sensitivity, specificity, F1 score, and AUC for the output results of the artificial intelligence model. Through this, the reliability of the left ventricular outflow tract obstruction prediction and prognosis evaluation results can be verified, and if the results fall short of the criteria, retraining can be performed by securing additional clinical data or supplementing the model structure. At this time, feedback information collected from the user terminal (200), the accuracy of the prediction results, medical staff evaluations, and follow-up echocardiogram results can be used as retraining data. Therefore, the service server (100) can secure the long-term stability and clinical reliability of the artificial intelligence model through continuous performance evaluation and periodic updates, and can gradually improve the accuracy and explainability of the patient-specific prognosis evaluation information.

[0045] Meanwhile, the user terminal (200) is a general term for a terminal possessed or used by a user who wishes to use the information provision service provided by the service server (100), and may be at least one or more. Here, the user may include various users such as the patient themselves, medical staff, researchers, or administrators, and is not limited thereto.

[0046] This user terminal (200) can upload echocardiogram images to the service server (100) and receive and display prediction results, prognosis evaluation information, and related indicators provided by the server. To this end, the user terminal (200) can be operated in the form of a mobile application, a web application, or dedicated software, and can provide results visualized as probability values, grade displays, score gauges, warning messages, trend graphs, etc. In addition, the user terminal (200) collects clinical confirmation of the prediction results, follow-up test results, or user feedback and transmits them to the service server (100), thereby enabling them to be utilized for checking the performance and retraining of the model.

[0047] The user terminal (200) described above may be a computer, UMPC (Ultra Mobile PC), workstation, netbook, PDA (Personal Digital Assistants), portable computer, web tablet, wireless phone, mobile phone, smartphone, pad, smart watch, wearable terminal, e-book, PMP (portable multimedia player), portable game console, navigation device, black box or digital camera, other mobile communication terminal, etc., capable of installing and running multiple applications (i.e., applications) desired by each user, and is not limited to the terminals mentioned above.

[0048] Meanwhile, the medical information providing device (300) stores and manages echocardiogram images, clinical records, and examination history for at least one patient, and provides necessary medical information to the service server (100). The medical information providing device (300) may be implemented as an in-hospital EMR system, a medical image storage and transmission system (PACS), a cloud-based medical database, or an external medical data platform linked thereto. Additionally, the medical information providing device (300) may update and store prediction results and prognosis evaluation information generated from the service server (100) to maintain consistency in clinical records.

[0049] Meanwhile, the information providing system (1000) is not limited to the configuration shown in FIG. 1, and may be configured to include other devices (terminals, servers, etc.) or implemented by excluding and integrating some configurations.

[0050] Through this information provision system (1000), the user can receive comprehensive early identification of the risk of left ventricular outflow tract obstruction and prognosis evaluation for patients with hypertrophic cardiomyopathy, and based on this, can obtain practical medical effects such as rapid clinical decision-making, patient-tailored management, and improved efficiency of follow-up monitoring.

[0051] FIG. 2 is a block diagram showing the configuration of a service server that performs information provision for predicting left ventricular outflow tract occlusion in hypertrophic cardiomyopathy according to one embodiment of the present disclosure.

[0052] Referring to FIG. 2, the service server (100) may include a communication interface (110), memory (120), I / O interface (130), and a processor (140), and each component may communicate with one or more communication buses or signal lines.

[0053] The communication interface (110) can be connected to the user terminal (200) and the medical information providing device (300), as well as other devices, via a wired / wireless communication network to exchange data.

[0054] Meanwhile, a communication interface (110) that enables the transmission and reception of such data includes a wired communication port (111) and a wireless circuit (112), wherein the wired communication port (111) may include one or more wired interfaces, for example, Ethernet, Universal Serial Bus (USB), FireWire, etc. Additionally, the wireless circuit (112) may transmit and receive data with an external device through an RF signal or an optical signal. Furthermore, wireless communication may use at least one of a plurality of communication standards, protocols, and technologies, such as GSM, EDGE, CDMA, TDMA, Bluetooth, Wi-Fi, VoIP, Wi-MAX, or any other suitable communication protocol.

[0055] The memory (120) may store data for at least one process (algorithm) or a program that reproduces the process for providing information provision services. In addition, the memory (120) may store additional processes for performing other operations, and is not limited thereto.

[0056] Meanwhile, the memory (120) can store various data used in the service server (100) as well as at least one pre-trained artificial intelligence-based model as needed.

[0057] In various embodiments, the memory (120) may include a volatile or non-volatile recording medium capable of storing various data, commands, and information. 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 (e.g., SD or XD memory, etc.), RAM, SRAM, ROM, EEPROM, PROM, network storage, cloud, and blockchain database.

[0058] In various embodiments, the memory (120) may store at least one configuration of an operating system (121), a communication module (122), a user interface module (123), and one or more applications (124).

[0059] An operating system (121) (e.g., embedded operating systems such as LINUX, UNIX, MAC OS, WINDOWS, VxWorks, etc.) may include various software components and drivers for controlling and managing general system operations (e.g., memory management, storage device control, power management, etc.) and may support communication between various hardware, firmware, and software components.

[0060] The communication module (122) can support communication with another device through the communication interface (110). The communication module (122) may include various software components for processing data received by the wired communication port (111) or wireless circuit (112) of the communication interface (110).

[0061] The user interface module (123) can receive user requests or inputs from a keyboard, touch screen, microphone, etc. through the I / O interface (130) and provide a user interface on the display.

[0062] The application (124) may include a program or module configured to be executed by one or more processors (140).

[0063] The I / O interface (130) can connect at least one of an input / output device (not shown) of the service server (100), such as a display, keyboard, touch screen, and microphone, to the user interface module (123). The I / O interface (130) can receive user input (e.g., voice input, keyboard input, touch input, etc.) together with the user interface module (123) and process commands based on the received user input.

[0064] The processor (140) is connected to a communication interface (110), memory (120), and I / O interface (130) to control the overall operation of the service server (100) and to execute various commands through an application or program stored in memory (120).

[0065] The processor (140) may correspond to a computing device such as a CPU (Central Processing Unit) or an AP (Application Processor). Additionally, the processor (140) may be implemented in the form of an Integrated Chip (IC), such as a System on Chip (SoC) that integrates various computing devices. Alternatively, the processor (140) may include a module for computing artificial neural network models, such as a Neural Processing Unit (NPU).

[0066] Specifically, the processor (140) may be configured to acquire an echocardiogram image of a specific patient and perform preprocessing in response to a user's service provision request, input the preprocessed echocardiogram image into a pre-trained artificial intelligence model, and then, when a prediction result regarding whether there is left ventricular outflow tract obstruction (LVOTO) is output from the artificial intelligence model, generate prognosis evaluation information based on this.

[0067] At this time, the processor (140) may perform at least one of noise removal, brightness correction, contrast correction, image resolution normalization, region of interest (ROI) extraction, time interval segmentation, or frame-to-frame alignment on the echocardiogram image when performing a preprocessing operation. However, these are merely examples and are not limited to these.

[0068] For example, the processor (140) can convert the image into a form that can be processed by an artificial intelligence model by normalizing the frame resolution of the input echocardiogram image, detecting regions of interest of the left ventricle and interventricular septum to remove background noise, and dividing the time interval into regular intervals to perform alignment between frames. This preprocessing process is intended to clarify myocardial boundaries and contraction patterns and to minimize noise caused by differences in shooting conditions.

[0069] Meanwhile, the processor (140) can run an artificial intelligence model including a first neural network for extracting spatial structural features and a second neural network for extracting temporal motion features. That is, the artificial intelligence model is configured to extract spatial structural features and temporal motion features of the myocardium from a preprocessed echocardiogram image, either individually or in combination. For example, the first neural network is implemented as a 3D convolutional neural network (3D-CNN), and the second neural network may be a 2D convolutional neural network (2D-CNN) or a Transformer-based network that analyzes temporal axis changes of the myocardium using automatically generated M-mode representations.

[0070] Additionally, the processor (140) can automatically track the myocardial motion trajectory within the echocardiogram image through the second neural network to generate a multi-level M-mode feature map and embed it to improve the accuracy of the left ventricular outflow tract occlusion prediction. Furthermore, the output embedding result of the second neural network can be combined with the output of the first neural network to comprehensively determine whether there is left ventricular outflow tract occlusion and output the result as a prediction result.

[0071] This prediction result may be a binary classification result classifying the presence or absence of left ventricular outflow tract obstruction, or may include a continuous value (score) that quantitatively predicts the left ventricular outflow tract pressure gradient. Based on this prediction result, the processor (140) may calculate a prediction index (DLI-LVOTO) that quantifies the degree of left ventricular outflow tract obstruction, and this prediction index may be calculated based on at least one of the contractile speed, travel distance, or timing difference of the myocardium in the echocardiogram image. Thus, the prediction index can be used to verify the quantitative prediction performance of the model through correlation with the actual clinical pressure gradient (Peak PG).

[0072] Furthermore, the processor (140) may be configured to generate prognosis evaluation information by combining the prediction results output from the artificial intelligence model with the patient's clinical information (e.g., age, gender, heart rate, history of drug administration, etc.). In this case, the prognosis evaluation information may include a prognosis grade or risk level calculated by comprehensively analyzing the prediction results and patient-specific clinical data, and a treatment response prediction result may be generated based on this prognosis grade or risk level. Accordingly, medical staff can determine the direction of treatment or identify patients in a high-risk group early based on the prognosis evaluation information.

[0073] Meanwhile, the processor (140) can calculate performance evaluation indicators such as accuracy, sensitivity, specificity, F1 score, and AUC for the output result of the artificial intelligence model and determine the reliability of the prediction result by comparing it with a reference value. If it falls short of the standard, the processor (140) can control the model to be retrained by obtaining additional clinical data, adjusting model parameters, or supplementing the structure. In this process, feedback information collected from the user terminal (200), whether the prediction result is accurate, evaluations by medical staff, or follow-up echocardiogram results can be used as retraining data, thereby continuously improving the long-term stability and clinical reliability of the artificial intelligence model.

[0074] Ultimately, the processor (140) can be configured to integrally manage the entire pipeline, including preprocessing of echocardiogram images, extraction of spatial and temporal features, prediction of left ventricular outflow tract obstruction, generation of prognosis evaluation information, and performance evaluation and relearning control. Through this, the information providing system (1000) according to the present disclosure can rapidly and consistently evaluate the possibility of left ventricular outflow tract obstruction and the prognosis of a patient with hypertrophic cardiomyopathy using only a single echocardiogram image, and can significantly improve the diagnostic efficiency of medical staff and the accuracy of establishing treatment plans.

[0075] FIG. 3 is a block diagram showing the configuration of a user terminal receiving prognosis evaluation information according to one embodiment of the present disclosure.

[0076] Referring to FIG. 3, the user terminal (200) may include a memory interface (210), one or more processors (220) and a peripheral interface (230). Various components within the user terminal (200) may be connected by one or more communication buses or signal lines.

[0077] The memory interface (210) is connected to the memory (250) and can transmit various data to the processor (220). Here, the memory (250) may include at least one type of storage medium among flash memory type, hard disk type, multimedia card micro type, card type memory (e.g., SD or XD memory, etc.), RAM, SRAM, ROM, EEPROM, PROM, network storage, cloud storage, or blockchain database.

[0078] In various embodiments, the memory (250) may store a web / app application or program built to enable information provision services to be provided by the service server (100). Additionally, the memory (250) may store various information about a user and may store various information obtained through the application or program.

[0079] In various embodiments, memory (250) may store at least one of an operating system (251), a communication module (252), a graphical user interface module (GUI) (253), a sensor processing module (254), a telephone module (255), and an application module (256). Specifically, the operating system (251) may include instructions for processing basic system services and instructions for performing hardware operations. The communication module (252) may communicate with at least one of one or more other devices, computers, and servers. The graphical user interface module (GUI) (253) may process a graphical user interface. The sensor processing module (254) may process sensor-related functions (e.g., processing voice input received through one or more microphones (292)). The telephone module (255) may process telephone-related functions. The application module (256) may perform various functions of a user application, such as electronic messaging, web browsing, media processing, navigation, imaging, and other processing functions. In addition, the user terminal (200) can store one or more software applications (256-1, 256-2) (e.g., applications for the service) associated with any one type of service in the memory (250).

[0080] In various embodiments, the memory (250) may store a digital assistant client module (257) (hereinafter, DA client module) for providing prognosis evaluation information, and accordingly, may store commands for performing client-side functions of the digital assistant (e.g., commands for performing functions for providing prognosis evaluation information) and various user data (258) (e.g., other data such as user-customized vocabulary data, preference data, user's electronic address book, etc.).

[0081] Meanwhile, the DA client module (257) can obtain voice input, text input, touch input and / or gesture input from the user through various user interfaces (e.g., I / O subsystem (240)) provided in the user terminal (200).

[0082] Additionally, the DA client module (257) can output data in the form of audiovisual and tactile signals. For example, the DA client module (257) can output data consisting of at least one combination of voice, sound, notifications, text messages, menus, graphics, videos, animations, and vibrations. Furthermore, the DA client module (257) can communicate with a digital assistant server (not shown) using a communication subsystem (280).

[0083] In various embodiments, the DA client module (257) may collect additional information about the surrounding environment of the user terminal (200) from various sensors, subsystems, and peripheral devices to construct the context associated with the user input. For example, the DA client module (257) may provide context information along with the user input to a digital assistant server to infer the user's intent. Here, the context information that may accompany the user input may include sensor information, e.g., lighting, ambient noise, ambient temperature, images, videos, etc. of the surrounding environment. As another example, the context information may include the physical state of the user terminal (200) (e.g., device orientation, device location, device temperature, power level, speed, acceleration, motion pattern, cellular signal strength, etc.). As yet another example, the context information may include information related to the software state of the user terminal (200) (e.g., processes running on the user terminal (200), installed programs, past and present network activity, background services, error logs, resource usage, etc.).

[0084] In various embodiments, the memory (250) may include additional or deleted instructions. Furthermore, the user terminal (200) may include additional configurations in addition to the configuration shown in FIG. 3, or exclude some configurations.

[0085] The processor (220) can control the overall operation of the user terminal (200) and can execute various commands to use the information provision service provided by the service server (100) by running an application or program stored in memory (250).

[0086] The processor (220) may correspond to a computing device such as a CPU (Central Processing Unit) or an AP (Application Processor). Additionally, the processor (220) may be implemented in the form of an integrated chip (IC), such as a System on Chip (SoC) that integrates various computing devices that perform machine learning, such as a Neural Processing Unit (NPU).

[0087] In various embodiments, the processor (220) may display various information and / or data, etc., for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy based on the platform through a user interface screen.

[0088] The peripheral interface (230) is connected to various sensors, subsystems, and peripheral devices and can provide data to enable the user terminal (200) to perform various functions. Here, it can be understood that the user terminal (200) performing a function is performed by the processor (220).

[0089] The peripheral interface (230) can acquire data from a motion sensor (260), a light sensor (light sensor) (261), and a proximity sensor (262), thereby enabling the user terminal (200) to perform orientation, light, and proximity detection functions. As another example, the peripheral interface (230) can acquire data from other sensors (263) (positioning system—GPS receiver, temperature sensor, biometric sensor), thereby enabling the user terminal (200) to perform functions related to the other sensors (263).

[0090] In various embodiments, the user terminal (200) may include a camera subsystem (270) connected to a peripheral interface (230) and an optical sensor (271) connected thereto, thereby enabling the user terminal (200) to perform functions such as viewing echocardiogram images or uploading reference images.

[0091] In various embodiments, the user terminal (200) may include a communication subsystem (280) connected to a peripheral interface (230). The communication subsystem (280) is composed of one or more wired / wireless networks and may include various communication ports, radio frequency transceivers, and optical transceivers.

[0092] In various embodiments, the user terminal (200) includes an audio subsystem (290) connected to a peripheral interface (230), and the audio subsystem (290) includes one or more speakers (291) and one or more microphones (292), so that the user terminal (200) can perform voice-operable functions, such as voice recognition, voice cloning, digital recording, and telephone functions.

[0093] In various embodiments, the user terminal (200) may include an I / O subsystem (240) connected to a peripheral interface (230). For example, the I / O subsystem (240) may control a touch screen (243) included in the user terminal (200) through a touch screen controller (241).

[0094] For example, the touch screen controller (241) can detect user contact and movement or interruption of contact and movement using any one of a plurality of touch sensing technologies, such as capacitive, resistive, infrared, surface acoustic wave technology, proximity sensor array, etc. As another example, the I / O subsystem (240) can control other input / control devices (244) included in the user terminal (200) through other input controller(s) (242). As an example, other input controller(s) (242) can control one or more pointer devices such as buttons, rocker switches, thumb wheels, infrared ports, USB ports, and styluses.

[0095] FIG. 4 is a flowchart schematically illustrating a method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure.

[0096] Referring to FIG. 4, the processor (140) acquires an echocardiogram image of a patient and performs preprocessing (S110).

[0097] In this S110 step, the processor (140) receives a patient's echocardiogram image from a user terminal (200) or a medical information providing device (300) and can perform at least one preprocessing step on the echocardiogram image, such as noise removal, brightness correction, contrast correction, image resolution normalization, region of interest extraction, time interval division, or frame-to-frame alignment.

[0098] This allows the left ventricular outflow tract and hypertrophic areas within the echocardiogram to be clearly distinguished, thereby converting the data into a form that can be reliably processed by an artificial intelligence model. Here, the echocardiogram image may generally be a B-mode image based on tissue brightness, but is not limited thereto.

[0099] Next, the processor (140) inputs the echocardiogram image preprocessed by step S110 into the artificial intelligence model (S120).

[0100] In this case, the artificial intelligence model may include a convolutional neural network (CNN)-based image analysis network and a recurrent neural network (RNN) or transformer-based time series analysis module.

[0101] In this S120 step, the processor (140) predicts at least one of whether the left ventricular outflow tract is blocked, the pressure gradient (Peak PG), or the degree of left ventricular outflow tract stenosis through this artificial intelligence model, and outputs the prediction result in the form of a numerical indicator (e.g., DLI-LVOTO). Here, the output of the artificial intelligence model, i.e., the prediction result, may include not only a simple normal / abnormal determination, but also quantitative evaluation results such as the trend of change of the indicator, prediction reliability, and risk.

[0102] Next, the processor (140) generates patient prognosis evaluation information based on the prediction results output from the artificial intelligence model (S130).

[0103] In this S130 step, the processor (140) can calculate the likelihood of future deterioration of the patient's condition, whether surgery or drug treatment is necessary, and information on the prediction of treatment response by comprehensively analyzing at least one of the predicted DLI-LVOTO value, pressure gradient, or degree of myocardial hypertrophy. Here, the prognosis evaluation information may be expressed in at least one form among a graph, a numerical indicator, or a grade (e.g., a qualitative grade such as good prognosis / caution / risk of deterioration), and transmitted to a user terminal (200) so that medical staff can check it immediately.

[0104] Meanwhile, although not illustrated in FIG. 4, according to various embodiments of the present disclosure, the processor (140) may calculate performance evaluation indicators such as accuracy, sensitivity, specificity, F1 score, and AUC for the prediction results of the artificial intelligence model, and if the results fall short of the criteria, retraining may be performed by obtaining additional clinical data, adjusting model parameters, or supplementing the structure. In this process, medical staff feedback provided through the user terminal (200), whether the prediction results are accurate, and follow-up test results may be used as retraining data.

[0105] Therefore, the processor (140) is not limited to simply receiving an echocardiogram image and outputting a prediction result, but also supports patient-specific treatment decisions by generating prognosis evaluation information based on the prediction result, and can improve the reliability and clinical utility of the prediction through continuous performance updates.

[0106] FIGS. 5A and 5B are schematic diagrams illustrating a pipeline for an artificial intelligence model according to one embodiment of the present disclosure to predict whether there is left ventricular outflow tract obstruction based on a single echocardiogram image.

[0107] Referring to FIG. 5a, when a patient's echocardiogram image (Parasternal Long-Axis View, PLAX) is input, the processor (140) generates an M-mode image based on the corresponding echocardiogram image, i.e., the B-mode image, and extracts spatial structural features and temporal motion features of the myocardium from the B-mode image and the M-mode image at multiple levels.

[0108] Specifically, the processor (140) uses a 3D convolutional neural network (3D CNN) to analyze spatial structural features, such as the shape and thickness of the myocardium and the opening and closing pattern of the left ventricular outflow tract, from a B-mode image, and simultaneously uses a 2D convolutional neural network (2D CNN) to analyze changes in myocardial contraction and movement patterns (temporal features) along the time axis from an M-mode image. These two features are embedded and processed together and used as input to an artificial intelligence model that predicts the presence and degree of left ventricular outflow tract obstruction.

[0109] The processor (140) calculates a prediction index (DLI-LVOTO) that quantifies the degree of left ventricular outflow tract obstruction based on the prediction results of the artificial intelligence model. This prediction index is expressed as a continuous value in the range of 0 to 100, and a higher value indicates a more severe degree of obstruction.

[0110] For example, as shown on the right side of Fig. 5a, if the DLI-LVOTO value is low, it can be evaluated as 'No obstruction', and if it is high, it can be evaluated as 'Severe LVOTO'.

[0111] For example, this feature extraction and prediction process can be performed by a pre-trained deep learning-based artificial intelligence model. This artificial intelligence model may include, for example, a 3D convolutional neural network (3D-CNN) that extracts spatial structural features using a sequence of frames of B-mode images as input, and a 2D convolutional neural network (2D-CNN) that analyzes the temporal motion patterns of the heart muscle using M-mode representations.

[0112] Furthermore, the processor (140) can fuse the multi-level feature embeddings output from these two neural networks to predict whether the left ventricular outflow tract is closed, and convert the prediction result into a prediction index (DLI-LVOTO) to be used for prognosis evaluation.

[0113] However, the configuration illustrated in FIG. 5a is merely one exemplary embodiment, and the artificial intelligence model may be replaced with other forms of deep learning structures (RNN, Transformer, etc.) that learn the time-series characteristics of myocardial motion.

[0114] Afterwards, the processor (140) can evaluate the accuracy by verifying the performance of the model against an internal validation set and an external validation set, as shown in FIG. 5b, and calculating indicators such as the ROC curve (AUC), specificity, and sensitivity.

[0115] Additionally, the processor (140) can verify whether the predictive indicator reflects the actual change in treatment response by comparing and analyzing patient data after myectomy or drug treatment (Mavacamten) to verify the clinical validity of the predictive indicator.

[0116] Accordingly, according to the configuration illustrated in FIG. 5a and FIG. 5b, the information providing system (1000) according to the present disclosure can not only quantitatively predict the degree of left ventricular outflow tract obstruction with only a single echocardiogram image input, but also continuously monitor changes in the patient's prognosis over the course of treatment.

[0117] FIGS. 6a to 6c are schematic diagrams illustrating the learning and evaluation process of an artificial intelligence model for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy according to one embodiment of the present disclosure, where FIG. 6a illustrates the learning data configuration and classification pipeline, and FIGS. 6b and FIGS. 6c each illustrate the process of evaluating the performance of the artificial intelligence model and analyzing drug response using internal and external validation sets.

[0118] Referring to FIG. 6a, the processor (140) collects echocardiogram images (PLAX) of non-obstructive HCMP (Obstructive HCMP(-)) and obstructive HCMP (Obstructive HCMP(+)) patients and builds them into a training dataset.

[0119] The processor (140) applies a 3D convolutional neural network (3D CNN) to extract features (B-mode features) from an input echocardiogram image, which is a B-mode image, and uses a point predictor to estimate representative points of the left ventricular outflow tract region from the features.

[0120] Next, based on the predicted coordinates, an M-mode Sampler automatically generates an M-mode image in the time axis direction, and the generated M-mode image extracts temporal motion features of the heart muscle through a 2D convolutional neural network (2D CNN).

[0121] The processor (140) has B-mode features ( ) and M-mode characteristics( It classifies whether there is left ventricular outflow tract obstruction (LVOT obstruction) by integrating the results, and distinguishes the results as non-obstructive HCMP (Obstructive HCMP(-)) or obstructive HCMP (Obstructive HCMP(+)).

[0122] These classification results can be used to assess the severity of left ventricular outflow tract obstruction (Significant / Severe LVOT obstruction), and the predicted point-based M-mode generation process can be used as an auxiliary task to improve the explainability of artificial intelligence models.

[0123] Referring to FIG. 6b, the processor (140) evaluates the performance of the artificial intelligence model trained in FIG. 6a using an internal validation set and an external validation set.

[0124] In Figure 6b, a box plot and a Receiver Operating Characteristic curve (ROC) for the prediction results are respectively plotted, and the Area Under the Curve (AUC) is approximately 0.8344 for the internal validation set and approximately 0.8954 for the external set, confirming that the artificial intelligence model of the present disclosure exhibits high prediction accuracy and generalization performance.

[0125] Additionally, as illustrated in FIG. 6c, the processor (140) analyzes the time series change of the predicted probability using echocardiogram images before and after drug administration as input to analyze the drug treatment effect (Mavacamten effect) in a patient with obstructive hypertrophic cardiomyopathy.

[0126] Each graph is echo date on the axis, The axis represents the predicted probability of left ventricular outflow tract closure, where the dotted line indicates the first administration date and the black dots represent the model prediction probability.

[0127] For example, since most patients show a tendency for the probability of left ventricular outflow tract obstruction to gradually decrease after drug administration, it can be confirmed that the artificial intelligence model of the present disclosure can quantitatively reflect the actual clinical treatment response.

[0128] Accordingly, as illustrated in FIGS. 6a to 6c, the artificial intelligence model of the present disclosure can accurately predict whether there is left ventricular outflow tract obstruction by integrally analyzing spatial and temporal features from a single echocardiogram image, and can also be effectively applied to the analysis of clinical treatment effects.

[0129] FIGS. 7a and 7b are drawings illustrating an example of a prediction point-based M-mode generation process and a sampling guide application in an artificial intelligence model according to one embodiment of the present disclosure.

[0130] First, Fig. 7a illustrates a process in which an artificial intelligence model automatically predicts the location of the left ventricular outflow tract within an echocardiogram image and automatically generates an M-mode image using the result.

[0131] Referring to FIG. 7a, the processor (140) estimates a predicted point distribution representing the location of the left ventricular outflow tract for the input echocardiogram image. From this predicted distribution, sample points ( , ) is selected, and an M-mode sampling line is set based on the corresponding coordinates to generate a predicted M-mode image in the time axis direction.

[0132] In this process, the processor (140) can perform learning using pseudo-point labels (MV tip at mid-systole) corresponding to actual measurement points during the pre-learning process, which act as a mutual correction relationship with the prediction points to form a sampling guide.

[0133] Meanwhile, Fig. 7b shows an example of performing prediction point visualization depending on whether a sampling guide is applied.

[0134] Referring to FIG. 7b, the processor (140) can perform prediction point visualization depending on whether a sampling guide is applied. When a sampling guide is not applied (w / o Sampling Guide), the prediction point ( , Although the distribution is irregular, when a sampling guide is applied, it can be seen that prediction points corresponding to the actual mitral tip (MV tip) appear aligned on the image.

[0135] As shown in the right image of Fig. 7b, each prediction point ( , )₁… ( , ) N It is distributed at a location corresponding to the pseudo-point label indicated by the orange line, which confirms that the artificial intelligence model reliably detects the anatomical reference point of the left ventricular outflow tract.

[0136] Accordingly, according to the configuration illustrated in FIGS. 7a and 7b, the artificial intelligence model of the present disclosure can probabilistically predict the location of the left ventricular outflow tract from a single echocardiogram image and automate the generation of time-axis M-mode images using the predicted points, thereby simultaneously improving the accuracy of myocardial motion pattern analysis and the explainability of the model.

[0137] FIG. 8 is a diagram showing the structure of an artificial intelligence model according to an M-mode sampling method according to one embodiment of the present disclosure.

[0138] Referring to FIG. 8, the artificial intelligence model of the present disclosure can be divided into three learning structures, each including a baseline, deterministic, and stochastic M-mode sampling structures.

[0139] First, the reference model (Vanila) takes the patient's echocardiogram images (B-mode images) as input and uses spatial structural features ( ) extracts. At this time, the extracted features ( ) is input into a classifier based on pre-trained B-mode weights to predict whether there is obstructive HCMP, and the prediction result ( ) outputs either "Obstructive HCMP(+)" or "Obstructive HCMP(-)". Here, the reference model (Vanilla) is set as the baseline for experimental comparison and refers to the basic model to which no M-mode sampling structure or auxiliary learning module is applied.

[0140] Meanwhile, according to the deterministic M-mode sampling structure, the deterministic model uses B-mode features ( Location coordinates of the left ventricular outflow tract through the point predictor from ) , ) is estimated, and based on this, an M-mode image in the time axis direction is generated through an M-mode sampler.

[0141] This generated M-mode image is processed by a 2D convolutional neural network (2D CNN) to produce temporal cardiac motion features ( Extracts ), and these two features ( ) is combined to form the final output ( Produces ).

[0142] Therefore, since deterministic models integrate and reflect spatial and temporal features through fixed-coordinate-based sampling, they can improve the accuracy of left ventricular outflow tract obstruction prediction.

[0143] Finally, according to the stochastic M-mode sampling structure, the stochastic model is configured so that the point predictor produces a Predict Point Distribution instead of a single coordinate value.

[0144] The processor (140) samples a point ( , ) is probabilistically extracted, and an M-mode image is generated based on this.

[0145] Subsequently, M-mode features extracted through a 2D convolutional neural network (2D CNN) ( ) is a B-mode characteristic( Combined with ) to produce a prediction result for left ventricular outflow tract closure.

[0146] This probability-based sampling method can reflect uncertainty when predicting the location of the left ventricular outflow tract, thereby deriving more reliable results that account for minute variability in myocardial motion and improving the explainability of the model.

[0147] Accordingly, according to the configuration illustrated in FIG. 8, the artificial intelligence model of the present disclosure applies a deterministic or probabilistic M-mode sampling structure compared to a reference model that uses only B-mode images, thereby integrally reflecting spatial structural features and temporal motion features, classifying the presence and severity of left ventricular outflow tract obstruction more accurately, and providing the advantage of being able to visually explain the internal reasoning process of the artificial intelligence model.

[0148] FIGS. 9a and 9b are drawings showing the results of a performance evaluation using internal and external validation datasets of an artificial intelligence model according to one embodiment of the present disclosure, wherein FIG. 9a shows the results for the internal validation set (SNUBH) and FIG. 9b shows the results for the external validation set (SEV).

[0149] Referring to Figures 9a and 9b respectively, three versions based on structural differences of each model are illustrated and include, from left to right, the results for the reference model (Vanila: Pre-trained weight model), the deterministic M-mode explainable model (hereinafter referred to as the 'deterministic model'), and the stochastic M-mode explainable model (hereinafter referred to as the 'stochastic model').

[0150] For each model, the graph on the left is a box plot showing the distribution of predicted values ​​relative to the actual ground truth (GT), and the graph on the right visualizes the model's classification performance (AUC) through the ROC curve.

[0151] First, according to the internal validation results in Figure 9a, the AUC of the reference model is approximately 0.8316, the deterministic model is approximately 0.8337, and the probabilistic model is approximately 0.8344, all showing similar or slightly improved prediction performance. This suggests that the M-mode sampling structure improves the explainability of the model while maintaining the prediction accuracy of the existing B-mode based model.

[0152] In addition, according to the external validation results in Fig. 9b, the reference model's AUC is approximately 0.8994, the deterministic model's is approximately 0.9132, and the probabilistic model's is approximately 0.8954, confirming that high generalization performance is achieved even with external data. In particular, the deterministic model exhibits the highest AUC in the external validation dataset, demonstrating that it can reliably predict whether there is left ventricular outflow tract obstruction without relying on internal data learning.

[0153] In addition, as can be seen from the box plot results in Figures 9a and 9b, the distribution of predicted values ​​between the non-occlusive patient group and the obstructive patient group is clearly distinguished, which demonstrates that the artificial intelligence model of the present disclosure can determine whether there is left ventricular outflow tract obstruction with high reliability.

[0154] Accordingly, according to the results illustrated in FIGS. 9a and 9b, the artificial intelligence model of the present disclosure achieves high classification accuracy in both internal and external verification environments, and in particular, demonstrates that the explainability and clinical reliability of the model can be improved simultaneously through an M-mode-based structure.

[0155] FIG. 10 is a diagram showing the results of a visual comparison based on whether the left ventricular outflow tract is blocked, as predicted by an artificial intelligence model according to one embodiment of the present disclosure.

[0156] FIG. 10 (a) is an example of a patient with obstruction (LVOT obstruction(+)), in which a processor (140) performs prediction point-based M-mode sampling in the mid-systole interval of an echocardiogram to produce an M-mode image generated along the left ventricular outflow tract region.

[0157] In the predicted M-mode image shown at the top of (a), the anterior wall of the myocardium and the mitral tip (MV tip) appear to be abnormally close toward the left ventricular septum, which is identified as a characteristic pattern reflecting stenosis or occlusion of the left ventricular outflow tract.

[0158] Meanwhile, the pseudo-label shown at the bottom of (a) represents reference coordinates defined based on the actual mitral tip at mid-systole, and the visual consistency of the model can be verified through its alignment with the prediction result. That is, the prediction point (blue line) and the pseudo-label (orange line) almost coincide, indicating that the artificial intelligence model of the present disclosure accurately detects the actual anatomical reference point.

[0159] Figure 10 (b) is an example of a non-obstructive patient (LVOT obstruction(-)), showing a comparison of predicted M-mode images and pseudo-labels generated during the same mid-systole interval.

[0160] In the predicted M-mode image shown at the top of (b), the movement of the mitral valve apex maintains a constant interval, and the opening and closing changes of the left ventricular outflow tract region show a pattern that remains normal. Additionally, the alignment between the prediction points and the pseudo-labels shown at the bottom of (b) is stably maintained, confirming that the model accurately reflects normal myocardial movement patterns.

[0161] Accordingly, according to the results illustrated in FIG. 10, the artificial intelligence model of the present disclosure can automatically generate M-mode images that accurately reflect actual anatomical structures and movement patterns for both patients with obstructed and non-obstructed left ventricular outflow tracts, thereby simultaneously improving the explainability and clinical reliability of the model.

[0162] FIG. 11 is a diagram showing an example of interpretation of an M-mode image generated by an artificial intelligence model according to one embodiment of the present disclosure using a prediction point-based sampling guide.

[0163] Figure 11 (a) shows a case of a closed patient (LVOT obstruction(+)) and (b) shows a case of a non-closed patient (LVOT obstruction(-)). In both cases, the left side shows the input B-mode (PLAX) image and prediction / guideline, and the right side shows the M-mode image automatically generated along the corresponding line.

[0164] First, referring to Fig. 11(a), the M-mode image generated during the mid-systole phase shows a pattern in which the anterior wall and the mitral valve tip (MV tip) structures approach or come into contact with the septum in an abnormal manner (SAM pattern) (red arrow).

[0165] As a result, the intervals between waveforms narrow or discontinuous hyperechoic bands form on M-mode images, which reflects characteristic findings associated with left ventricular outflow tract stenosis or occlusion.

[0166] The artificial intelligence model of this embodiment can automatically predict the corresponding line from a single echocardiogram image and visualize it in the time axis direction, thereby enabling both quantitative and qualitative analysis of the closure pattern.

[0167] Meanwhile, referring to Fig. 11(b), in the M-mode image generated by the same procedure, the gap between the apex of the mitral valve and the septum (E-septal separation) is maintained at a constant level, and a periodic and regular wave pattern appears (blue double-sided arrow).

[0168] Consequently, no SAM pattern or abnormal contraction pattern is observed, suggesting that left ventricular outflow tract opening and closing are maintained normally.

[0169] Accordingly, according to the results illustrated in FIG. 11, the artificial intelligence model of the present disclosure can visually explain the presence or absence of left ventricular outflow tract obstruction through a consistent pipeline of prediction lines, M-mode generation, and pattern reading, and can simultaneously improve the interpretability and clinical reliability of the artificial intelligence model.

[0170] FIG. 12 is a diagram showing an example of tracking the response to Mavacamten treatment in a patient with obstructive hypertrophic cardiomyopathy using an artificial intelligence model according to one embodiment of the present disclosure.

[0171] Each graph in Fig. 12 is based on the results of continuous echocardiography of different patients, The axis is the inspection date (echo date), The axes represent the predicted probabilities of left ventricular outflow tract closure. Here, the dotted line indicates the time of the first drug administration, and each dot represents the closure probability predicted by the model.

[0172] As shown in Fig. 12, most patients show a tendency for the predicted probability to gradually decrease after drug administration, which suggests that the artificial intelligence model of the present disclosure can quantitatively reflect the effect of relieving left ventricular outflow tract obstruction following actual drug treatment.

[0173] Meanwhile, for some patients, the rate of change or trend of predicted probability varies, and these individual differences can support clinical judgments, such as predicting treatment response and adjusting personalized drug dosages.

[0174] Accordingly, Figure 12 shows that the artificial intelligence model of the present disclosure can be used as a clinical application tool capable of quantitatively and visually evaluating changes in a patient's condition before and after treatment based on a single echocardiogram image input.

[0175] As described above, according to the present disclosure, by using a single echocardiogram image of a patient as input to perform preprocessing, determination of left ventricular outflow tract obstruction, quantification of the degree of obstruction (calculation of DLI-LVOTO), and prediction of treatment response as a single integrated pipeline, it is possible to automatically provide an AI-based information service for myocardial function assessment and left ventricular outflow tract obstruction prediction applicable in actual clinical settings. This resolves the issues of insufficient objective indicators and inefficiency in tracking treatment response, which were pointed out as limitations of existing manual interpretation and qualitative evaluation methods, and improves the accuracy of patient-specific prognosis prediction and treatment decision-making, thereby simultaneously ensuring clinical reliability and practical utility.

[0176] Although embodiments of the present disclosure have been described in more detail with reference to the attached drawings, the present disclosure is not necessarily limited to these embodiments and may be modified in various ways within the scope of the technical spirit of the present disclosure. Accordingly, the embodiments of the present disclosure are intended to explain, not limit, the technical spirit of the present disclosure, and the scope of the technical spirit of the present disclosure is not limited by these embodiments. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of protection of the present disclosure shall be interpreted by the claims below, and all technical spirits within an equivalent scope shall be interpreted as being included within the scope of rights of the present disclosure. Explanation of the symbols

[0177] 1000: Information provision system 100: Service Server 110: Communication interface 111: Wired communication port 112: Wireless circuit 120: Memory 121: Operating System 122: Communication Module 123: User Interface Module 124: Application 130: I / O Interface 140: Processor 200: User terminal 210: Memory interface 220: Processor 230: Peripheral Interfaces 240: I / O Subsystem 241: Touchscreen Controller 242: Other Input Controllers 243: Touch screen 244: Other Input Control Devices 250: Memory 251: Operating System 252: Communication Module 253: GUI Module 254: Sensor processing module 255: Phone module 256: Applications 256-1, 256-2: Application 257: Digital Assistant Client Module 258: User data 260: Motion sensor 261: Light sensor 262: Proximity sensor 263: Other sensors 270: Camera Subsystem 271: Optical sensor 280: Communications subsystem 290: Audio Subsystem 291: Speaker 292: Microphone 2100: Display Module 300: Medical information providing device

Claims

Claim 1 A method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, performed by a device, comprising: a step of acquiring an echocardiogram image of a patient and performing preprocessing; a step of inputting the preprocessed echocardiogram image to an artificial intelligence model that is pre-trained to take a single echocardiogram image as input, extract spatial structural features and temporal motion features of the myocardium, either individually or in combination, and predict whether there is left ventricular outflow tract obstruction based on the extracted features and output a prediction result; and a step of generating prognosis evaluation information for the patient based on the prediction result when a prediction result based on the preprocessed echocardiogram image is output from the artificial intelligence model. Claim 2 A method for providing information for predicting left ventricular outflow tract occlusion in hypertrophic cardiomyopathy, wherein the step of performing the preprocessing comprises performing at least one of noise removal, luminance correction, contrast correction, image resolution normalization, region of interest (ROI) extraction, time interval segmentation, or inter-frame alignment on the echocardiogram image. Claim 3 A method for providing information for predicting left ventricular outflow tract occlusion in hypertrophic cardiomyopathy, wherein the artificial intelligence model comprises a first neural network for extracting spatial structural features from the preprocessed echocardiogram image and a second neural network for extracting temporal motion features from the frame sequence of the preprocessed echocardiogram image. Claim 4 A method for providing information for predicting left ventricular outflow tract occlusion in hypertrophic cardiomyopathy, wherein the first neural network is implemented as a 3D Convolutional Neural Network (3D-CNN), and the second neural network is a 2D Convolutional Neural Network (2D-CNN) or Transformer-based network configured to analyze time-axis changes of the myocardium using automatically generated M-mode representations. Claim 5 A method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, wherein, in paragraph 3, the second neural network automatically tracks the myocardial motion trajectory within the echocardiogram image to generate a multi-level M-mode feature map, and embeds the multi-level M-mode feature map to improve the accuracy of predicting left ventricular outflow tract obstruction. Claim 6 A method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, wherein, in claim 5, the artificial intelligence model predicts whether left ventricular outflow tract obstruction is achieved by combining the embedding result output from the second neural network with the output of the first neural network. Claim 7 A method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, wherein the prediction result comprises at least one of a binary classification result classifying the presence or absence of left ventricular outflow tract obstruction or a continuous value (score) that quantitatively predicts the left ventricular outflow tract pressure gradient. Claim 8 A method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, wherein the prediction result is calculated as a prediction index (DLI-LVOTO) that quantifies the degree of left ventricular outflow tract obstruction, and the prediction index is calculated based on at least one of the contractile speed, displacement distance, or timing difference of the myocardium in the preprocessed echocardiogram. Claim 9 A method for providing information for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, wherein the step of generating the prognostic evaluation information comprises combining the prediction result with at least one clinical information (age, gender, heart rate, drug administration history, etc.) regarding the patient to calculate a prognostic grade or risk, and outputting a treatment response prediction result according to the calculated prognostic grade or risk. Claim 10 An information providing device for predicting left ventricular outflow tract obstruction in hypertrophic cardiomyopathy, comprising: a communication interface; a memory; and a processor operably connected to the communication interface and the memory, wherein the processor acquires an echocardiogram image of a patient and performs preprocessing, extracts spatial structural features and temporal motion features of the myocardium, either individually or in combination, using a single echocardiogram image as input, and inputs the preprocessed echocardiogram image to a pre-trained artificial intelligence model that predicts whether there is left ventricular outflow tract obstruction based on the extracted features and outputs a prediction result, and when a prediction result based on the preprocessed echocardiogram image is output from the artificial intelligence model, generates prognosis evaluation information for the patient based on the prediction result.

Images