Method and apparatus for predicting and diagnosing risk level of cardiovascular disease by type

Abstract

One embodiment of the cardiovascular disease type risk prediction diagnostic device is characterized by including at least one processor, which uses a diagnostic model to predict cardiovascular disease type risk, the diagnostic model including: a first model that analyzes input fundus images to predict cardiovascular risk; and a second model that receives input of cardiovascular risk score predicted by the first model and disease type-specific factors, and predicts cardiovascular disease type risk, and the at least one processor provides diagnostic information based on the cardiovascular risk predicted by the first model and the cardiovascular disease type risk predicted by the second model.

Description

Technical Field

[0001] This application relates to a method and apparatus for predicting and diagnosing the risk level of cardiovascular diseases by type using machine learning that learns from fundus images and cardiovascular biomarkers. Background Technology

[0002] Fundus images (i.e., fundus photographs or retinal images) are images taken of the posterior surface of the retina, including the retina, retinal vessels, optic nerve, macula, and choroid. Fundus images have traditionally been used primarily for the diagnosis of ophthalmic diseases, but due to the rapid development of artificial intelligence (AI) technology in recent years, their application has been significantly expanded to assessing the risk of systemic diseases such as cardiovascular diseases. This technology is not only non-invasive and cost-effective, but is also recognized as a particularly useful tool in disease screening processes targeting large populations. AI has demonstrated superior performance in analyzing retinal vascular changes associated with cardiovascular diseases, thus creating a new paradigm in the field of medical image analysis. As a technology utilizing existing fundus images to assist in the diagnosis of cardiovascular diseases, the applicant has previously filed Korean patent applications with application numbers 10-2018-0166720, 10-2018-0166721, and 10-2018-0166722. These patent applications relate to technologies that provide auxiliary information related to comprehensive risk assessment of cardiovascular diseases, focusing on more accurately predicting the risk of cardiovascular diseases through fundus images and supporting clinical decision-making based on this.

[0003] On the other hand, cardiovascular diseases include detailed types such as myocardial infarction (MI), stroke, peripheral artery disease (PAD), heart failure (HF), and atrial fibrillation (AF). While the applicant's prior patent applications aim to predict the overall risk of cardiovascular diseases, predicting individual risks based on the specific type of cardiovascular disease would provide more detailed auxiliary information. This could provide crucial support for healthcare professionals to make more accurate clinical decisions based on each patient's condition. Summary of the Invention

[0004] Technical issues

[0005] One problem this invention aims to solve is to provide a method and an apparatus for implementing the method, which utilizes a machine learning model that learns the correlation between fundus images and cardiovascular disease type risk, receives fundus images of a patient as input, and is able to predict the patient's risk according to cardiovascular disease type. Specifically, the object of this invention is to provide an apparatus for predicting risk according to cardiovascular disease type, which, based on a comprehensive cardiovascular risk score derived from features extracted from fundus images by a machine learning model, additionally considers disease type-specific factors such as age, sex, blood biomarkers, smoking status, and obesity to predict risk according to cardiovascular disease type.

[0006] One problem this invention aims to solve is to provide an apparatus for predicting risk based on cardiovascular disease type, which additionally considers the type of cardiovascular plaque in the subject patient, i.e., whether it is a soft plaque or a hard plaque, in order to predict risk based on cardiovascular disease type.

[0007] One problem this invention aims to solve is to provide an apparatus for predicting risk based on cardiovascular disease type, comprising a model for determining the appropriate clinical examination to be recommended for a patient based on the risk of cardiovascular disease type.

[0008] The problems to be solved by this application are not limited to those mentioned above, and any problems not mentioned can be clearly understood by those skilled in the art from this specification and the accompanying drawings.

[0009] Technical solution

[0010] According to one aspect of this application, an embodiment of a diagnostic device for predicting risk by cardiovascular disease type may include: a first model (cardiovascular risk prediction model) that analyzes input fundus images to predict cardiovascular risk; and a second model (cardiovascular disease type risk prediction model) that additionally receives inputs of a cardiovascular risk score predicted by the first model and disease type-specific factors to predict risk by cardiovascular disease type, and the diagnostic device for predicting risk by cardiovascular disease type is characterized in that, by integrating the cardiovascular risk predicted by the first model and the risk by cardiovascular disease type predicted by the second model, it provides risk levels and related information for each cardiovascular disease type.

[0011] According to one aspect of this application, an embodiment of a diagnostic device for predicting risk by cardiovascular disease type may include: a first model that analyzes input fundus images to predict cardiovascular risk; a second model that receives input of a cardiovascular risk score predicted by the first model and disease type-specific factors to predict risk by cardiovascular disease type; and a third model (recommendation examination judgment model) that, based on the cardiovascular disease type risk predicted by the second model, determines recommended clinical examinations for the patient, and the diagnostic device for predicting risk by cardiovascular disease type is characterized in that, by integrating the cardiovascular risk predicted by the first model and the cardiovascular disease type risk predicted by the second model, it provides risk levels and related information for each cardiovascular disease type, and related information for recommended examinations.

[0012] According to one aspect of this application, an embodiment of a diagnostic device for predicting risk by cardiovascular disease type may include: a first model that analyzes input fundus images to predict cardiovascular risk and cardiovascular plaque type; a second model that additionally receives input of cardiovascular risk score and disease type-specific factors predicted by the first model to predict risk by cardiovascular disease type; and a third model that determines recommended clinical examinations for patients based on the cardiovascular disease type risk predicted by the second model, and the diagnostic device for predicting risk by cardiovascular disease type is characterized in that, by integrating the cardiovascular risk and plaque type predicted by the first model and the cardiovascular disease type risk predicted by the second model, it provides risk levels and related information for each cardiovascular disease type, and related information for recommended examinations.

[0013] Beneficial effects

[0014] According to the present invention, a machine learning model that learns the correlation between fundus images and the risk of cardiovascular disease types can predict not only the patient's overall cardiovascular risk score but also the risk specific to each cardiovascular disease type by receiving fundus images as input. Furthermore, the machine learning model can predict the patient's cardiovascular plaque type. Through these methods, optimized personalized treatment plans can be developed for each patient, and patients with high risk of specific cardiovascular diseases can be identified early to strengthen preventative measures.

[0015] The effects of this application are not limited to those described above, and any effects not mentioned can be clearly understood by those skilled in the art from this specification and the accompanying drawings. Attached Figure Description

[0016] Figure 1 This is a block diagram illustrating a diagnostic device according to an embodiment.

[0017] Figure 2 This is a diagram used to illustrate that cardiovascular risk (Reti-CVD) can serve as a biomarker for predicting individual risks of cardiovascular stroke, myocardial infarction, and atrial fibrillation.

[0018] Figure 3 and Figure 4 This is a graph used to illustrate the correlation between Reti-CVD and various types of cardiovascular disease.

[0019] Figure 5 and Figure 6 This is a graph used to illustrate the cardiovascular disease prediction performance for diabetic patients involved in this embodiment.

[0020] Figure 7 This is a diagram illustrating a method for predicting risk by type of cardiovascular disease in one embodiment.

[0021] Figure 8 This is a graph used to illustrate the predictive performance of atrial fibrillation risk in one embodiment.

[0022] Figure 9 This is a diagram illustrating a method for predicting risk by type of cardiovascular disease in one embodiment.

[0023] Figure 10 This is a diagram illustrating a method for predicting risk by type of cardiovascular disease in another embodiment. Detailed Implementation

[0024] The above-described objects, features, and advantages of this application will become more apparent from the following detailed description in relation to the accompanying drawings. However, various modifications and embodiments can be applied to this application; specific embodiments are illustrated in the drawings and described in detail below.

[0025] Throughout this specification, the same reference numerals generally denote the same constituent elements. Furthermore, functionally identical constituent elements appearing in the drawings of various embodiments and falling within the same conceptual scope are described using the same reference numerals, and repeated descriptions of these elements are omitted.

[0026] Detailed descriptions of well-known functions or components related to this application are omitted where it is deemed that they might unnecessarily obscure the essence of this application. Furthermore, the numbers used in this specification (e.g., first, second, etc.) are merely identification symbols to distinguish one component from others.

[0027] Furthermore, the suffixes "module" and "part" used for constituent elements in the following embodiments are assigned or used interchangeably only for the convenience of writing the specification, and do not have any distinguishing meaning or function.

[0028] In the following embodiments, the singular expression includes the plural expression, provided that there is no obvious difference in meaning in the context.

[0029] In the following embodiments, terms such as "comprising" or "having" are intended to refer to the presence of the features or components described in the specification, without pre-excluding the possibility of adding more than one other feature or component.

[0030] In the accompanying drawings, the sizes of the components may be exaggerated or reduced for ease of illustration. For example, the sizes and thicknesses of the components shown in the drawings are arbitrarily illustrated for ease of explanation, and the invention is not necessarily limited to the contents shown.

[0031] In cases where a particular embodiment can be implemented in different ways, the order of a specific step may differ from the order described. For example, two steps described consecutively may be performed substantially simultaneously, or they may be performed in the reverse order of the description.

[0032] In the following embodiments, when referring to the connection of components, etc., it includes not only the case where the components are directly connected, but also the case where they are indirectly connected by intervening other components. For example, in this specification, when referring to the electrical connection of components, it includes not only the case where the components are directly electrically connected, but also the case where they are indirectly electrically connected by intervening other components.

[0033] The following describes a diagnostic device and method for predicting the risk of cardiovascular disease types based on eye images to assist medical personnel in making judgments. In this specification, the term "diagnosis" may not mean directly diagnosing a disease, but rather refers to diagnostic assistance used to aid in the diagnosis of a disease. The term "diagnosis" is used here for ease of explanation, but it can be interpreted as diagnostic assistance. The machine learning models described in this specification can be designed based on various machine learning libraries.

[0034] Cardiovascular diseases include hypertension, atherosclerosis, coronary artery disease (CAD), valvular disease, pericardial disease, myocarditis, cardiomyopathy, arrhythmia, heart failure (HF), rheumatic heart disease, congenital heart disease, aortic aneurysm, peripheral vascular disease, cerebrovascular disease, and pulmonary vascular disease. More specifically, these include angina pectoris, myocardial infarction (MI), pulmonary artery hypertension (PAH), cor pulmonale, and left heart failure. Failure, right heart failure, aortic valve stenosis (AS), aortic valve regurgitation (AR), pulmonary valve stenosis (PS), pulmonary valve regurgitation (PR), tricuspid valve stenosis (TS), tricuspid valve regurgitation (TR), mitral valve stenosis (MS), mitral valve regurgitation, peripheral artery disease,Cardiovascular disease may include at least one of the following: PAD (pericardial aortic aneurysm, thoracic aortic aneurysm), aortic aneurysm (e.g., abdominal aortic aneurysm, thoracic aortic aneurysm), pericarditis, stroke, cerebral infarction, cerebral hemorrhage, cerebral aneurysm, thrombotic disease (e.g., deep vein thrombosis, pulmonary embolism), atrial fibrillation (AF), atrial tachycardia (AT), paroxysmal supraventricular tachycardia (PSVT), ventricular tachycardia (VT), bradycardia, ischemic stroke, hemorrhagic stroke, and transient ischemic attack (TIA). Cardiovascular disease may include complications. In addition, complications can include heart attack, death due to cardiovascular disease, cardiogenic shock, kidney disease, aspiration pneumonia, dysphagia, decreased motor function, decreased language function, decreased cognitive function, sleep disorders, mood disorders, neuropathic pain, urinary tract infection, malnutrition, deep vein thrombosis, pressure sores, falls, pain, seizures, depression, etc. Additionally, detailed subtypes based on the type of cardiovascular disease can be included. For example, heart failure is classified as heart failure with reduced left ventricular ejection fraction (HFrEF), heart failure with mildly reduced ejection fraction (HFmrEF), and heart failure with preserved ejection fraction (HFpEF). Regarding stroke, it can be classified as ischemic stroke, thrombotic stroke (cerebral thrombosis), cerebral embolism (embolic stroke), lacunar infarction, hemorrhagic stroke, intracerebral hemorrhage, subarachnoid hemorrhage, transient ischemic attack, brainstem stroke, and stroke of unknown cause. Regarding atrial fibrillation, it includes detailed subtypes such as paroxysmal atrial fibrillation, persistent atrial fibrillation, long-term persistent atrial fibrillation, and permanent atrial fibrillation. Regarding myocardial infarction, it is classified into detailed subtypes such as type 1 (spontaneous myocardial infarction), type 2 (myocardial infarction due to ischemic imbalance), type 3 (myocardial infarction leading to death when biomarker values ​​are unavailable), type 4a (myocardial infarction associated with percutaneous coronary intervention (PCI), type 4b (myocardial infarction associated with stent thrombosis), and type 5 (myocardial infarction associated with coronary artery bypass grafting (CABG)).

[0035] 1. Biomarkers for specific types of cardiovascular diseases

[0036] Fundus images provide a non-invasive way to observe the structure of blood vessels in the human body, thus laying the foundation for predicting the likelihood of cardiovascular disease (CVD). Machine learning models based on fundus images and cardiovascular biomarkers are used to predict the risk of future CVD. Key CVD biomarkers include the Coronary Artery Calcium (CAC) score, Pooled Cohort Equation (PCE) score, QRISK score, modified Framingham Score (FRS), Carotid Intima-Media Thickness (CIMT) score, brachial pulse wave velocity (baPWV) score, and ankle-brachial index (ABI). The machine learning model generates a cardiovascular risk score, Reti-CVD, based on the input fundus images. Based on this score, individuals are categorized into low-risk, intermediate-risk, and high-risk groups. The classification criteria adopt the baseline values ​​established in the applicant's previous study, "Deep-learning-based cardiovascular risk stratification using coronary artery calcium scores predicted from retinal photographs," *The Lancet Digital Health* 3(5). The applicant's previous research revealed that the Reti-CVD score can be used as a biomarker to predict the risk of future cardiovascular diseases. This invention newly reveals the close correlation between the Reti-CVD score and the risk of future cardiovascular diseases of various types, and provides a detailed algorithm for predicting risk by cardiovascular disease type based on Reti-CVD.

[0037] The diagnostic device can use machine learning models to predict various information and risks, including cardiovascular risk, risk by type of cardiovascular disease, and risk by subtype of type, as described in this manual.

[0038] Figure 1 This is a block diagram illustrating a diagnostic device according to one embodiment. (Refer to...) Figure 1 The diagnostic device (10) may include a processor (12), a storage module (11), and a communication module (13).

[0039] The processor (12) may include one or more of the following: CPU (Central Processing Unit), RAM (Random Access Memory), GPU (Graphics Processing Unit), one or more microprocessors, and electronic components capable of processing the input data according to other predetermined logic. Furthermore, the processor (12) may be at least one.

[0040] The processor (12) can read the system program and various processing programs stored in the storage module (11). As an example, the processor (12) can expand the procedures, methods, etc., for performing the diagnosis described later in RAM and perform various processing according to the expanded program. For example, the processor (12) can process the algorithms described in this specification. The processor (12) can use machine learning models to predict cardiovascular risk. In addition, the processor (12) can implement the various methods described in this specification.

[0041] The storage module (11) can store the diagnostic model. The storage module (11) can store the machine learning model, the parameters of the machine learning model, variables, algorithms described in this specification, etc.

[0042] The storage module (11) can be implemented using non-volatile semiconductor memory, hard disk, flash memory, RAM, ROM (Read Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory) or other tangible non-volatile recording media.

[0043] The storage module (11) can store various processing programs, parameters used to execute program processing, or data resulting from such processing. As an example, the storage module (11) can store programs used for diagnostics described later, parameters, and data obtained from the execution of such programs. Furthermore, the storage module (11) can store various machine learning models described later.

[0044] Although not illustrated, the diagnostic device (20) may also include an input module. The input module can acquire user input. For example, the input module can acquire user input. Furthermore, the input module can acquire at least one of the subject's physical information described in this specification (height, weight, age, sex, race, smoking status, blood pressure (e.g., blood pressure value, hypertension), diabetes (or, blood glucose level), cholesterol level, etc.). Additionally, the input module can acquire disease-type specific factors, as described later. The processor (12) can acquire the input information through the input module.

[0045] The communication module (13) can communicate with various devices (e.g., learning devices, client devices, etc.). As an example, the diagnostic device (10) can be implemented as a server that communicates with client devices. Alternatively, the diagnostic device (10) can be implemented as a client device.

[0046] <Example 1 - Risk Prediction for Stroke, Myocardial Infarction, and Atrial Fibrillation>

[0047] Figure 2 This is a graph illustrating that cardiovascular risk (Reti-CVD) can serve as a biomarker for predicting the individual risks of cardiovascular stroke, myocardial infarction, and atrial fibrillation. (a) shows the incidence of stroke, myocardial infarction, and atrial fibrillation in low-risk, intermediate-risk, and high-risk groups based on Reti-CVD scores. Furthermore, (b) shows the adjusted hazard ratio (HR), HR trend, and C-index for each risk group. Here, CI stands for confidence interval.

[0048] (1) Stroke: The Reti-CVD score showed a significant association with the risk of stroke. The analysis revealed that the risk of stroke trended upward with increasing Reti-CVD score (adjusted HR trend, 1.50; 95% CI, 1.16–1.95; p = 0.002). This suggests that the Reti-CVD score could be used as a useful biomarker for identifying high-risk patient populations requiring early preventative intervention.

[0049] (2) Myocardial Infarction (MI): The Reti-CVD score also showed a significant correlation with the risk of myocardial infarction (MI). The risk of MI trended upward with increasing Reti-CVD score (adjusted HR trend, 1.22; 95% CI, 1.01–1.47; p = 0.036). This score can be used to identify individuals at high risk of myocardial infarction early and to enhance preventative measures.

[0050] (3) Atrial Fibrillation (AF): The Reti-CVD score was also closely associated with the risk of atrial fibrillation (AF) (adjusted HR trend, 1.32; 95% CI, 1.15–1.51; p<0.001). This suggests that the Reti-CVD score can be useful in assessing the risk of atrial fibrillation, and that Reti-CVD can be an important tool for early risk assessment and the development of preventive management strategies for atrial fibrillation.

[0051] <Example 2 - Risk Prediction for 14 Types of Cardiovascular Diseases, Including Myocardial Infarction and Peripheral Artery Disease>

[0052] This example illustrates the correlation between Reti-CVD scores and 13 different cardiovascular diseases (CVDs) and arterial hypertension. A cross-sectional analysis was conducted on 45,980 participants from the UK Biobank study. Logistic regression analysis was used to confirm differential correlations with various cardiovascular diseases, and adjustments were made for factors such as hypertension, diabetes, dyslipidemia, and smoking during the analysis.

[0053] Figure 3 and Figure 4 This is a graph used to illustrate the correlation between Reti-CVD and various types of cardiovascular disease. (See reference...) Figure 3 and Figure 4 Tables (a) and (b) illustrate the relationship between Reti-CVD scores and risk by cardiovascular disease type. In (a) and (b), OR represents the odds ratio, LCL represents the lower control limit, and UCL represents the upper control limit. These tables confirm that the high-risk group based on Reti-CVD scores is significantly associated with detailed cardiovascular disease risk compared to the low-risk group.

[0054] (1) Coronary artery disease or myocardial infarction (MI): The higher the Reti-CVD score, the significantly increased risk of coronary artery disease. In the high-risk group, the odds ratio (OR) showed 10.37 (95% CI, 7.58–14.18), a very significant increase observed in the overall trend.

[0055] (2) Peripheral vascular disease or peripheral artery disease (PAD): The risk of peripheral vascular disease increases with increasing Reti-CVD score. The odds ratio for the high-risk group was 9.65 (95% CI, 2.94–31.64), showing a statistically significant correlation.

[0056] (3) Atrial fibrillation (AF): A strong correlation was also observed between Reti-CVD score and risk of atrial fibrillation, with an odds ratio of 9.36 (95% CI, 6.51-13.45) in the high-risk group.

[0057] (4) Aortic Valve Stenosis (AVS): The higher the Reti-CVD score, the greater the risk of aortic stenosis. The odds ratio for the high-risk group was 8.13 (95% CI, 1.87-35.35).

[0058] (5) Other diseases: Significant correlation with Reti-CVD score has also been found in various cardiovascular diseases such as stroke, heart failure (HF), and pulmonary embolism (PE).

[0059] Examples 1 and 2 focus on using the Reti-CVD score to analyze the correlation between various types of cardiovascular diseases and related states, and to predict the risk for each type. These results can be used to detect the likelihood of cardiovascular diseases currently unknown to the patient and to assess their risk. That is, based on the correlation data between the Reti-CVD score and various CVDs, the diagnostic device can detect cardiovascular diseases that the patient may currently be unaware of. This is useful for identifying diseases in which the patient has not clearly felt symptoms or which are not diagnosed in their early stages. Furthermore, based on the research results, the diagnostic device can analyze the Reti-CVD score and existing patient health data to prioritize recommendations for the patient's most likely current cardiovascular disease. This system can guide patients and healthcare professionals to consider additional examinations for potentially risky diseases and can promote early diagnosis and preventative treatment. For example, when a patient shows a high risk associated with atrial fibrillation (AF) based on their Reti-CVD score, the patient is more likely to have undiagnosed atrial fibrillation, and therefore the diagnostic device can recommend additional electrocardiogram (ECG) examinations or monitoring. When another patient presents with an intermediate risk associated with peripheral vascular disease (PVD) or peripheral artery disease (PAD), the diagnostic device can guide them to undergo more focused testing for the disease on which they are more likely to develop.

[0060] 2. Cardiovascular predictive biomarkers for patients with pre-existing diseases

[0061] Reti-CVD can serve as a cardiovascular risk index that more accurately predicts the risk of secondary cardiovascular-related diseases by considering a patient's sex, ethnicity, and pre-existing disease status. This score can be derived from a model learned from fundus images and coronary artery calcification (CAC) data. While the Reti-CVD score primarily assesses cardiovascular disease risk by analyzing a patient's fundus images, it can be extended to a predictive tool that comprehensively considers a patient's pre-existing disease status. This allows for more accurate prediction of the risk of secondary cardiovascular disease in high-risk groups, such as patients with chronic diseases like diabetes, hypertension, and hyperlipidemia, or those who have already experienced cardiovascular disease (heart failure, coronary artery disease, stroke, etc.). For example, the incidence patterns of cardiovascular disease may differ based on sex (male / female) and ethnicity. By additionally reflecting such demographic variables in the Reti-CVD score, individual predictions of cardiovascular risk can be made for specific sexes or ethnicities. Furthermore, patients with chronic diseases such as hypertension, hyperlipidemia, and diabetes may have an additional increased risk of cardiovascular disease. The Reti-CVD-based risk prediction model of the present invention can take into account such pre-existing conditions and predict the risk of secondary cardiovascular disease (e.g., heart failure, coronary artery disease, stroke, etc.). Patients with pre-existing cardiovascular disease (heart failure, coronary artery disease, stroke, etc.) have an additional risk of developing new types of cardiovascular disease. The Reti-CVD score can be used to more accurately assess the risk of secondary cardiovascular disease in this patient population.

[0062] <Example 1 - Prediabetes and Diabetes Patients>

[0063] This embodiment uses data from Biobank UK on prediabetes and diabetes patients, classifying them into three groups—low-risk, intermediate-risk, and high-risk—based on the Reti-CVD score, and tracking and observing patients to assess fatal and non-fatal cardiovascular disease types (coronary artery disease, ischemic stroke, and transient ischemic attack).

[0064] Figure 5 and Figure 6These graphs illustrate the cardiovascular disease prediction performance of the Reti-CVD score in diabetic patients as described in this embodiment. They represent results based on the Reti-CVD score, categorized in a 2:1:1 ratio at the 50th and 75th percentiles into low-risk (n=550), intermediate-risk (n=276), and high-risk (n=275) groups. This cardiovascular disease prediction score is derived from retinal images of prediabetic and diabetic patients. To evaluate the predictive ability of the Reti-CVD score for fatal and non-fatal cardiovascular events, survival analyses using Cox proportional hazards models and hazard ratios (HRs) were performed using longitudinal data from UK Biobank. Figure 5 and Figure 6 As shown, among 1101 patients with prediabetes and diabetes, 138 (12.5%) experienced cardiovascular events. During the median follow-up period of 11 years, the incidence rates were 8.2% (45 / 550) in the low-risk group, 15.2% (42 / 276) in the intermediate-risk group, and 18.5% (51 / 275) in the high-risk group. A significant association was observed between the cardiovascular disease prediction score and the incidence of cardiovascular events after adjusting for factors such as age, sex, use of antihypertensive medications, statin use, and smoking history. Specifically, the risk of cardiovascular events was significantly increased in the intermediate-risk group (hazard ratio 1.57, 95% CI, 1.00–2.47) and the high-risk group (hazard ratio 1.88, 95% CI, 1.19–2.98) compared to the low-risk group. Furthermore, a trend of gradually increasing hazard ratios was observed (hazard ratio trend 1.36, 95% CI, 1.09–1.70). These results suggest that the Reti-CVD score can be used as a useful tool for risk stratification between patients with prediabetes and diabetes, and indicate its significant potential in managing high-risk patients.

[0065] 3. Methods for predicting risk based on cardiovascular disease type

[0066] 3.1 Risk prediction model based on cardiovascular disease type (classification model)

[0067] To directly predict the risk of cardiovascular disease types, a diagnostic device can perform the following actions. A machine learning (deep learning) model executed within the diagnostic device can learn using the input fundus image and the presence or absence of cardiovascular disease types such as atrial fibrillation (AF), coronary artery disease (CAD), peripheral artery disease (PAD), heart failure (HF), and stroke as labels. Through this learning process, features associated with cardiovascular diseases are learned and identified. Based on the analysis results, the processor of the diagnostic device can predict the likelihood of the subject's cardiovascular disease type from the fundus image. Deep learning algorithms for predicting the risk of multiple cardiovascular disease types from fundus images can be designed in various ways. These can include single-model, parallel, and multi-task models. A single-model approach uses a single deep learning model to simultaneously predict multiple cardiovascular disease types (e.g., atrial fibrillation (AF), coronary artery disease (CAD), peripheral artery disease (PAD), heart failure (HF), and stroke). This model receives the fundus image as input and has output nodes that individually predict the likelihood of each cardiovascular disease type. The final layer of the model has multiple output nodes, each returning the probability of a specific cardiovascular disease type. In parallel modeling methods, individual deep learning models are built for each cardiovascular disease type. For example, models predicting atrial fibrillation (AF), coronary artery disease (CAD), and peripheral artery disease (PAD) exist independently. Each model receives fundus images as input and predicts only the cardiovascular disease type of its target. Multi-task modeling methods are similar to single-modeling methods, but they enhance the learning of shared features for predicting multiple cardiovascular disease types. One model learns shared features and makes individual predictions for each cardiovascular disease type based on these features. After learning features related to multiple cardiovascular disease types in the shared layer, the model is designed to branch according to each cardiovascular disease type. Each path after the branching performs detailed predictions for that cardiovascular disease type.

[0068] Figure 7 This is a diagram illustrating a method for predicting risk by type of cardiovascular disease in one embodiment.

[0069] Reference Figure 7 The processor of the diagnostic device may include a step of acquiring fundus images (S100) and a step of acquiring cardiovascular disease diagnostic information (S200).

[0070] In step S100, the processor of the diagnostic device can acquire fundus images. Furthermore, according to an embodiment, the processor of the diagnostic device can perform preprocessing, enhancement, and serialization on the acquired fundus images. Various techniques can be applied to this, therefore detailed descriptions are omitted.

[0071] Furthermore, in step S200, the processor of the diagnostic device can acquire cardiovascular disease diagnostic information. In this specification, the diagnostic information may be represented as diagnostic auxiliary information. Cardiovascular disease diagnostic information may include risk prediction information by cardiovascular disease type. In this specification, risk prediction information by cardiovascular disease type may include risk prediction information for cardiovascular disease types such as stroke, myocardial infarction (MI), atrial fibrillation (AF), heart failure (HF), and peripheral artery disease (PAD). For example, risk prediction information by cardiovascular disease type may include information on the probability value (score) and / or grade of the occurrence of that cardiovascular disease type in the subject based on fundus images. Furthermore, risk prediction information by cardiovascular disease type may include information on the probability value (score) and / or grade of the occurrence of that cardiovascular disease type in the subject based on fundus images within a predetermined period (e.g., within 10 years, within 5 years). Additionally, secondary auxiliary information, described later, may be acquired in step S200. Specific embodiments will be described below.

[0072] <Example 1 - Reti-AF Single Model>

[0073] This embodiment illustrates the conceptual operation of the Reti-AF algorithm for predicting the risk of atrial fibrillation (AF). The Reti-AF algorithm uses fundus images (retinal images) of the patient as input and learns features from these images to identify patterns associated with the presence or absence of atrial fibrillation (AF). By analyzing the results, the algorithm predicts whether the input image is associated with AF or not. The Reti-AF algorithm can be executed by the processor of a diagnostic device.

[0074] To evaluate the performance of the deep learning algorithm Reti-AF involved in this embodiment, univariate Cox regression analysis and multivariate Cox regression analysis were performed. Figure 8 This is a graph illustrating the predictive performance of atrial fibrillation risk according to one embodiment. (Refer to...) Figure 8(a) is a table summarizing the relevant information on the risk of atrial fibrillation (Reti-AF) using univariate Cox regression analysis, and (b) is a table summarizing the relevant information on the risk of atrial fibrillation (Reti-AF) using multivariate Cox regression analysis.

[0075] Univariate Cox regression analysis showed that the incidence of atrial fibrillation (AF) was significantly increased with higher Reti-AF scores (without adjustment for variables such as age and gender). The hazard ratio (HR) for the high-risk group was 5.50 (95% CI, 4.08–7.41), demonstrating the effectiveness of Reti-AF in predicting AF risk. In multivariate Cox regression analysis, Reti-AF maintained a significant association with AF occurrence even after adjusting for variables such as age, sex, smoking status, diabetes, hyperlipidemia, and hypertension. In particular, age played a significant role as an independent variable in predicting AF occurrence, strengthening the predictive performance of Reti-AF. The C-index of the multivariate model was 0.76, showing improved performance compared to the univariate analysis. Ultimately, the Reti-AF algorithm demonstrated significant performance in predicting the risk of AF occurrence.

[0076] 3.2 Model for predicting risk by cardiovascular disease type based on cardiovascular biomarkers (Reti-CVD)

[0077] 3.2.1 Basic Algorithm Structure

[0078] Patients with pre-existing cardiovascular disease (heart failure, coronary artery disease, stroke, etc.) have an additional risk of developing new cardiovascular diseases. Therefore, it is necessary to predict the risk of developing each cardiovascular disease.

[0079] Figure 9 This is a diagram illustrating a method for predicting risk by type of cardiovascular disease in one embodiment. Figure 9 The process is illustrated by assessing a patient’s overall cardiovascular disease risk using a Reti-CVD prediction model (diagnostic model (1000)) (first model (1100)) and predicting the risk by cardiovascular disease type using an analysis that includes disease type-specific factors (second model (1200)). The first model (1100) and the second model (1200) can be included in the diagnostic model (1000).

[0080] Cardiovascular Risk Prediction Model ("First Model"): In this step, the patient's overall cardiovascular risk is assessed using the first model (1100). The first model (1100) learns from the patient's fundus images and coronary artery calcification (CAC) scores, and can predict a cardiovascular risk score, i.e., Reti-CVD, based on the input fundus images. In addition to the Reti-CVD score, the first model (1100) can also predict various disease-type-specific factors. For example, disease-type-specific factors may include at least one of the parameters representing the subject's physical information, such as biological age, sex, height, weight, BMI, and information about obesity (obesity score, BMI, body mass, body fat percentage, body muscle mass, whether obese, etc.).

[0081] Disease-type specific factors (laboratory markers) may include at least one of the diagnostic numerical parameters, such as hematocrit, red blood cell count, white blood cell count, hemoglobin level, platelet count, total iron-binding capacity (TIBC), iron content, ferritin (storage ferritin) level, total protein level, albumin level, aspartate aminotransferase (AST) level, alanine aminotransferase (ALT) level, γ-GT, alkaline phosphatase (ALP) level, globulin level, hepatitis antigen level, hepatitis antibody level, blood glucose level, glycated hemoglobin level (HbA1c), blood urea nitrogen (BUN) level, creatinine (Cr) level, uric acid level, total cholesterol level, high-density lipoprotein (HDL) cholesterol level, low-density lipoprotein (LDL) cholesterol level, triglyceride (TG) level, apolipoprotein B (ApoB), lipoprotein(a) (Lp(a)), and C-reactive protein (C-Reactive Protein). CRP, erythrocyte sedimentation rate (ESR), bicarbonate level, pH value, hematuria, occult blood in urine, protein in urine, glucose in urine, pH value, bilirubin in urine, ketone bodies in urine, urobilinogen in urine, glucose in urine, white blood cells in urine, creatinine in urine, albumin in urine, and albumin / creatinine ratio, etc. In addition, disease-type specific factors may include at least one of the following: electrocardiogram (ECG), systolic blood pressure (SBP), diastolic blood pressure (DBP), family history information, or patient clinical history information.

[0082] On the other hand, the first model (1100) can learn the presence or absence of cardiovascular plaque types, namely soft plaques and hard plaques, together with fundus images, and can predict the cardiovascular plaque type based on the input fundus images. The content related to cardiovascular plaque type prediction is explained in detail in the algorithm section reflecting cardiovascular plaque types in 3.2.4.

[0083] The model for predicting risk by cardiovascular disease type ("Second Model"): In this step, the risk for each cardiovascular disease type (e.g., atrial fibrillation, coronary artery disease, stroke, etc.) is predicted using data derived from the cardiovascular risk prediction model. The risk for each disease type is assessed by combining the Reti-CVD score obtained from the First Model (1100) with disease type-specific factors. The data input to the Second Model (1200) can replace the Reti-CVD score as a feature vector of the fundus image extracted by the First Model (1100). On the other hand, disease type-specific factors can be predicted by the First Model as described above, and in this case, the prediction results (output results) of the type-specific factors can be input to the Second Model (1200).

[0084] The second model (1200) may also include an input module. This input module can obtain disease-type-specific factors through user input. For example... Figure 9 As shown, when additional patient data (e.g., gender, age, blood test results, smoking status, obesity, etc.) is available, this patient data can be input into a model that predicts risk by cardiovascular disease type. This model performs an individual analysis of the probability of occurrence for each specific cardiovascular disease to predict which cardiovascular disease a patient is more susceptible to.

[0085] Output of Cardiovascular Disease Diagnostic Information: In the final step, cardiovascular disease diagnostic information can be obtained and output by combining overall cardiovascular risk (comprehensive score) and risk categories. This diagnostic information may include the current patient status, risk categories, and related explanatory information. In addition, it may provide auxiliary information for developing personalized prevention and treatment plans (e.g., secondary auxiliary information described later).

[0086] <Example - Specific Patient Case>

[0087] The model for predicting risk by cardiovascular disease type considers the Reti-CVD score and type-specific covariates to independently assess type-specific risk for each patient.

[0088] Table 1

[0089] 1. Patient A Case Overall Status Higher Reti-CVD score, hypertension, smoking history, female, 65 years old Stroke risk Prediction: Very high. Explanation: Patient A has a very high predicted risk of stroke due to the combined effects of a high Reti-CVD score, hypertension, smoking history, age (65 years), and being female. In stroke models, all of these factors contribute significantly to the increased risk; therefore, the patient's likelihood of developing a stroke is very high. Risk of myocardial infarction (MI) Prediction: High. Explanation: Patient A has a high risk of myocardial infarction due to high Reti-CVD, hypertension, smoking history, and age (65 years). However, menopause and LDL cholesterol levels are the main influencing factors; the risk of myocardial infarction may be further increased if the patient is menopausal or has high LDL cholesterol levels. Risk of atrial fibrillation (AF) Prediction: Moderate. Explanation: Patient A has a moderate predicted risk of atrial fibrillation due to high Reti-CVD, hypertension, smoking history, and age (65 years). Obesity is a major factor, but the risk can remain at a moderate level in the absence of obesity. Risk of heart failure (HF) Prediction: High. Explanation: For patient A, age (65 years), hypertension, and smoking history are the main risk factors for heart failure. Due to the high Reti-CVD score and the combination of multiple risk factors, the predicted risk of heart failure is high. Risk of peripheral artery disease (PAD) Prediction: High. Explanation: For patient A, smoking history, hypertension, age (65 years), and female sex all contribute to an increased risk of peripheral artery disease (PAD). Combined with a high Reti-CVD score, the risk of PAD is predicted to be high.

[0090] Table 2

[0091] 2. Patient B Case Overall Status Moderate Reti-CVD score, low cholesterol, obesity, family history of heart disease, male, 50 years old. Stroke risk Prediction: Moderate. Explanation: Patient B has a moderate Reti-CVD score and is obese, but lacks major risk factors for stroke, namely hypertension or a history of smoking. Considering his male sex and age of 50, the risk of stroke is predicted to be moderate. Risk of myocardial infarction (MI) Prediction: High. Explanation: Patient B has a high risk of myocardial infarction due to factors such as obesity, family history of heart disease, and being male. Although the Reti-CVD score is moderate and the LDL cholesterol level is low, obesity and family history play important predictive roles for myocardial infarction, therefore the risk is predicted to be high. Risk of atrial fibrillation (AF) Prediction: High. Explanation: Patient B has a high risk of atrial fibrillation due to obesity and age (50 years). Despite having a moderate Reti-CVD score, the risk of atrial fibrillation is predicted to be high due to male sex, obesity, and age. Risk of heart failure (HF) Prediction: Moderate. Explanation: Patient B has a family history of obesity and heart disease, but considering the relatively young age of 50 and a moderate Reti-CVD score, the predicted risk of heart failure is moderate. Risk of peripheral artery disease (PAD) Prediction: Moderate. Explanation: For patient B who is obese, the risk of PAD may be slightly higher, but considering the moderate Reti-CVD score and age (50 years), the risk is predicted to be moderate. The lower LDL cholesterol also contributes to the reduced risk.

[0092] 3.2.2 Detailed Algorithm of the Model for Predicting Risk by Cardiovascular Disease Type (Model Two)

[0093] For each type of cardiovascular disease (stroke, myocardial infarction (MI), atrial fibrillation (AF), heart failure (HF), and peripheral artery disease (PAD)), the model can consider Reti-CVD scores and specific covariates. Specific covariates, for example, for stroke, may include blood pressure, history of transient ischemic attack (TIA), atrial fibrillation status, and smoking status; for myocardial infarction (MI), may include LDL cholesterol, family history of heart disease, smoking, and hypertension; and for atrial fibrillation (AF), may include age, history of heart disease, obesity, and hypertension. Furthermore, for heart failure (HF), may include age, history of heart disease, and smoking; and for peripheral artery disease (PAD), may include smoking, diabetes, hypertension, hyperlipidemia, and age.

[0094] The processor of the diagnostic device can use a multivariate Cox proportional hazards model or other appropriate statistical or machine learning model for each detailed disease type. The model described below can be included in the second model (1200). Below are examples of risk formulas performed by each model for each disease type.

[0095] (1) Stroke model:

[0096] Stroke risk = exp(β1 × Reti-CVD + β2 × Blood Pressure + β3 × TIA History + β4 × Atrial Fibrillation Status + β5 × Smoking + ...)

[0097] (2) Myocardial infarction (MI) model:

[0098] Risk of myocardial infarction = exp(β1 × Reti-CVD + β2 × LDL cholesterol + β3 × Family history of heart disease + β4 × Smoking + β5 × Hypertension + …)

[0099] (3) Atrial fibrillation (AF) model:

[0100] Risk of atrial fibrillation = exp(β1 × Reti-CVD + β2 × Age + β3 × History of Heart Disease + β4 × Obesity + β5 × Hypertension + …)

[0101] (4) Heart failure (HF) model:

[0102] Risk of heart failure = exp(β1 × Reti-CVD + β2 × Age + β3 × Smoking + β4 × History of Heart Disease + …)

[0103] (5) Peripheral artery disease (PAD) model:

[0104] Peripheral artery disease risk = exp(β1 × Reti-CVD + β2 × Smoking + β3 × Diabetes + β4 × Hypertension + β5 × Hyperlipidemia + β6 × Age + …)

[0105] In the above formula, β1, β2, ... represent the coefficients for each covariate, reflecting the covariate's impact on risk. Each model outputs risk estimates (e.g., hazard ratios or probabilities) for a specific disease type (stroke, myocardial infarction (MI), atrial fibrillation (AF), heart failure (HF), peripheral artery disease (PAD)). These risk estimates are calculated independently, although all models use the Reti-CVD score as a common input, the results can be calculated individually. Furthermore, these risk estimates represent individual probabilities or hazard ratios for different outcomes and are therefore not summed. Each risk can be interpreted independently based on a combination of the Reti-CVD score and disease type-specific factors. Each model can run independently. That is, based on the Reti-CVD score and type-specific covariates, the risk for each cardiovascular type (stroke, myocardial infarction, atrial fibrillation, heart failure, peripheral artery disease) is calculated; these models do not influence each other, and risk predictions for each type can be performed independently.

[0106] 3.2.3 Includes the algorithm for the recommendation check and judgment model.

[0107] Figure 10 This is a diagram illustrating a method for predicting risk by type of cardiovascular disease in another embodiment.

[0108] Reference Figure 10 The recommended check and judgment model can be added to the basic algorithm structure described in 3.2.1 above. The aforementioned content applies to the first model (cardiovascular risk prediction model, 1100) and the second model (risk prediction model based on cardiovascular disease type, 1200), therefore detailed explanations are omitted. The third model (1300) can also be included in the diagnostic model (1000).

[0109] Recommended diagnostic model ("Third Model"): The Third Model (1300) determines what additional clinical examinations should be performed on the patient based on the risk of cardiovascular disease type. The processor of the diagnostic device can provide information about the additional clinical examinations. Here, additional clinical examinations may include electrocardiogram (ECG), echocardiogram, ABI (Ankle-Brachial Index), ABPI (Ankle-Brachial Pressure Index), blood test, urine test, ultrasound, etc.

[0110] For example, in cases where the risk of stroke is predicted to be high, additional blood pressure monitoring or brain imaging may be recommended. In cases of high risk of myocardial infarction, additional blood tests (e.g., LDL confirmation) or cardiac function tests may be necessary. In cases of high risk of heart failure, a more detailed assessment of cardiac function is required, such as echocardiography or BNP (B-type natriuretic peptide) testing. In cases of high risk of peripheral artery disease (PAD), lower extremity Doppler ultrasound, ankle-brachial index (ABI), or other vascular examinations may be recommended to assess blood flow. Furthermore, in cases of predicted high risk of atrial fibrillation (AF), ECG monitoring (e.g., 24-hour Holter monitoring) may be recommended to detect paroxysmal AF. For heart failure, peripheral artery disease, and atrial fibrillation, individual patient measurements can also be recommended through relevant mobile service applications such as heart rate monitoring apps and pain recording apps.

[0111] <Example - Recommended examination output for patients A and B>

[0112] This embodiment illustrates the process of determining the need for additional clinical examinations based on risk prediction of cardiovascular disease types for patients A and B. Appropriate additional examinations are recommended for disease types where the risk is predicted to be high based on each patient's condition, thereby enabling accurate assessment and management of the patient's cardiovascular health.

[0113] Table 3

[0114] 1. Patient A Case Overall Status Higher Reti-CVD score, hypertension, smoking history, female, 65 years old Stroke risk Prediction: Very High. Recommended Examination: Because patient A's stroke risk is predicted to be very high, additional clinical examinations are required. Specifically, blood pressure monitoring and brain imaging (e.g., MRI or CT) are recommended. These examinations will help determine the likelihood of stroke early and develop preventative treatment plans. Risk of myocardial infarction (MI) Prediction: High Recommendation: Because patient A's risk of myocardial infarction is predicted to be high, additional blood tests (e.g., LDL cholesterol confirmation) and cardiac function tests (e.g., echocardiography) are recommended. These tests help to more accurately assess myocardial infarction risk factors and modify the treatment plan if necessary. Risk of atrial fibrillation (AF) Prediction: Moderate. Recommended testing: Although patient A's risk of atrial fibrillation is predicted to be moderate, considering the age (65 years), ECG monitoring (e.g., 24-hour Holter monitoring) can be considered. This can be useful for the early detection of paroxysmal atrial fibrillation (AF). Risk of heart failure (HF) Prediction: High Recommendation: Because patient A's risk of heart failure is predicted to be high, a more detailed assessment of cardiac function is required through echocardiography and BNP (B-type natriuretic peptide) testing. These tests will help monitor the progression of heart failure and develop an appropriate treatment plan. Risk of peripheral artery disease (PAD) Prediction: High Recommendation: Given that patient A's risk of peripheral artery disease is predicted to be high, assessment of blood flow status via lower extremity Doppler ultrasound, ankle-brachial index, or other vascular examinations is recommended. This is used to confirm the presence of peripheral vascular stenosis and, if necessary, to initiate timely treatment.

[0115] Table 4

[0116] 2. Patient B Case Overall Status Moderate Reti-CVD score, low cholesterol, obesity, family history of heart disease, male, 50 years old. Stroke risk Prediction: Moderate. Recommended examinations: Since patient B's stroke risk is predicted to be moderate, additional examinations are not necessary. Regular blood pressure monitoring and lifestyle modifications are recommended. Continuous monitoring can be performed via periodic retinal imaging. Risk of myocardial infarction (MI) Prediction: High Recommendation: Given that patient B's risk of myocardial infarction is predicted to be high, non-invasive cardiac examinations (e.g., exercise / pharmacological stress tests) and echocardiography should be considered. This will aid in the early detection and prevention of heart disease. Risk of atrial fibrillation (AF) Prediction: High Recommendation: Due to the predicted high risk of atrial fibrillation in patient B, ECG monitoring (e.g., 24-hour Holter monitoring) is recommended. This can detect early signs of atrial fibrillation and help correct the rhythm early. Risk of heart failure (HF) Prediction: Moderate. Recommended testing: Since patient B's risk of heart failure is predicted to be moderate, it is best to monitor cardiac function through periodic echocardiography and BNP testing. Additional testing can be determined based on the current condition. Risk of peripheral artery disease (PAD) Prediction: Moderate. Recommended testing: Since patient B's peripheral artery disease risk is predicted to be moderate, periodic lower extremity Doppler ultrasound can be used to monitor blood flow. While no additional testing is required, lifestyle modifications to manage risk factors are recommended.

[0117] When the processor of the diagnostic device acquires new data regarding recommended examinations, the processor of the diagnostic device inputs the patient's data back into the second model (1200), such as... Figure 10 As shown, this can be used to reassess the risk by type of cardiovascular disease.

[0118] 3.2.4 Algorithm reflecting cardiovascular plaque type

[0119] Algorithms reflecting cardiovascular plaque types can be combined with the basic algorithm described in 3.2.1 above, or the algorithm of the recommendation and judgment model in 3.2.3 can also be included in the configuration. The foregoing applies to the first model (cardiovascular risk prediction model, 1100) and the second model (a model predicting risk by cardiovascular disease type, 1200), therefore detailed explanations are omitted. As briefly mentioned earlier, the first model (1100) can learn the correlation between the input fundus image and the cardiovascular plaque type. Cardiovascular plaque types can be obtained from CT coronary angiography (CT-CA), cardiovascular magnetic resonance imaging (CMR), 18F-FDG PET (Positron Emission Tomography), 18F-NaF PET (Positron Emission Tomography), coronary angiography, intravascular ultrasound (IVUS), intravascular ultrasound-RF analysis (IVUS-RF Analysis), optical coherence tomography (OCT), optical frequency domain imaging (OFDI), intracoronary thermography, Raman spectroscopy, and near-infrared spectroscopy (NIRS). Specifically, cardiovascular risk prediction models can be learned using labels representing soft plaques and their corresponding retinal images, as well as labels representing hard plaques and their corresponding retinal images. Therefore, the cardiovascular risk prediction model can obtain the probability value (and / or grade) of the probability that a subject has soft plaques or the probability value (and / or grade) of the probability that a subject has hard plaques in retinal images. The first model can predict plaque type with probability values, which can be expressed as scores, together with Reti-CVD and disease type-specific factors.

[0120] The plaque type predicted by the first model (1100) can be reflected in the risk prediction by cardiovascular disease type. For example, soft plaques have a high risk of acute events due to the possibility of rupture and embolism, and are particularly likely to be associated with the risk of stroke. Furthermore, soft plaques increase the risk of heart failure due to the increased risk of acute limb ischemia caused by the potential for embolization, which may increase the risk of peripheral artery disease. Conversely, hard plaques are associated with progressive occlusion and may lead to myocardial infarction over time. Furthermore, hard plaques increase the risk of heart failure due to the increased risk of chronic limb ischemia and intermittent claudication caused by progressive vascular stenosis, which may increase the risk of peripheral artery disease.

[0121] Therefore, the second model (1200) can consider the plaque type predicted by the first model (1100) as a covariate for detailed disease types, predicting the risk of cardiovascular diseases such as stroke, myocardial infarction, heart failure, and peripheral artery disease. Furthermore, the third model (1300) can consider this plaque type and determine recommended clinical examinations for the patient. For example, if the second model predicts a high risk of stroke and / or the third model predicts the presence of soft or hard plaques, the third model (1300) can generate blood biomarkers (e.g., coagulation tests, lipid profiles), blood pressure information, carotid artery status information, and information required for the diagnosis of atrial fibrillation (age, blood pressure information, etc.) as required information, and recommend clinical examinations to obtain this information. Furthermore, if a high risk of myocardial infarction and / or the presence of hard plaques is confirmed, the third model (1300) can generate blood biomarkers (e.g., lipid profiles), information on family history, etc., as required information, and recommend clinical examinations to obtain this information. Furthermore, when a high risk of heart failure and / or the presence of soft or hard plaques is confirmed based on heart failure diagnosis information, the third model (1300) can generate blood biomarkers (e.g., BNP levels), a history of myocardial infarction, and overall cardiac function indicators as required information, and recommend clinical examinations for obtaining this information. Additionally, when a high risk of peripheral artery disease and / or the presence of soft or hard plaques is confirmed based on peripheral artery disease diagnosis information, the third model can generate smoking information, diabetes information, and blood pressure information as required information, and recommend clinical examinations for obtaining this information.

[0122] 3.2.5 Detailed subtype prediction algorithm for cardiovascular disease types

[0123] Cardiovascular diseases exist in various forms (subtypes). Each subtype differs in its pathological mechanisms, symptoms, prognosis, and treatment methods; therefore, clearly distinguishing these aspects is crucial. This invention, based on fundus images and employing a machine learning model, surpasses existing comprehensive cardiovascular risk prediction methods by providing an algorithm for predicting risk by cardiovascular disease type. The processor of the diagnostic device, based on the algorithm involved in this invention, can predict the risk of detailed subtypes of cardiovascular diseases. For the basic model structure, it can be applied... Figure 9 The content described herein is omitted in detail.

[0124] As an example, heart failure (HF) can be categorized into three subtypes: heart failure with reduced left ventricular ejection fraction (HFrEF), heart failure with preserved left ventricular ejection fraction (HFpEF), and heart failure with mildly reduced left ventricular ejection fraction (HFmrEF). Heart failure with reduced left ventricular ejection fraction (HFrEF) is characterized by a left ventricular ejection fraction (LVEF) below 40%, indicating a decline in myocardial contractility and a significant reduction in the heart's ability to eject blood. In this case, a history of myocardial infarction (MI), chronic hypertension, valvular disease, diabetes, and smoking are disease-type-specific factors, and echocardiography, cardiac MRI, and BNP (B-type natriuretic peptide) testing may be recommended. Heart failure with preserved left ventricular ejection fraction (HFpEF) is characterized by a left ventricular ejection fraction (LVEF) above 50%, primarily due to problems with ventricular diastolic function, resulting in a decreased ability of the heart to fully relax and fill with blood. In this case, obesity, hypertension, diabetes, advanced age, and female gender can be considered as disease-type-specific factors, and echocardiography, cardiac MRI, and BNP testing can be recommended. Furthermore, heart failure with mildly reduced left ventricular ejection fraction (HFmrEF) is a borderline state of heart failure with a left ventricular ejection fraction (LVEF) of 40-49%, considered an intermediate form between HFrEF and HFpEF. Advanced age, hypertension, diabetes, obesity, and valvular heart disease can be considered as disease-type-specific factors, and echocardiography, cardiac MRI, and MBP testing can be recommended.

[0125] Similar to heart failure cases, stroke can be classified into ischemic stroke, thrombotic stroke (cerebral thrombosis), cerebral embolism (embolic stroke), lacunar infarction, hemorrhagic stroke, intracerebral hemorrhage, subarachnoid hemorrhage, transient ischemic attack, brainstem stroke, and stroke of unknown cause. Atrial fibrillation includes detailed subtypes such as paroxysmal atrial fibrillation, persistent atrial fibrillation, long-term persistent atrial fibrillation, and permanent atrial fibrillation. Myocardial infarction is classified into detailed subtypes such as type 1 (spontaneous myocardial infarction), type 2 (myocardial infarction due to ischemic imbalance), type 3 (myocardial infarction leading to death when biomarker values ​​are unavailable), type 4a (myocardial infarction associated with percutaneous coronary intervention (PCI), type 4b (myocardial infarction associated with stent thrombosis), and type 5 (myocardial infarction associated with coronary artery bypass grafting (CABG)).

[0126] Therefore, the second model (cardiovascular disease risk prediction model) according to the present invention can reflect detailed subtype-specific factors of each type in cardiovascular risk and predict the risk by disease type subtype. Furthermore, the third model (recommendation examination judgment model) can recommend clinical examinations required for the target patient's cardiovascular disease, taking into account the detailed subtype of the disease.

[0127] 3.2.6 Secondary Auxiliary Information

[0128] Diagnostic information may include the aforementioned Reti-CVD cardiovascular risk score, plaque-based diagnostic information, disease-type-specific factors for the target patient, risk and related information for each detailed disease type, recommended examination information, etc. The secondary auxiliary information generated based on the diagnostic information is explained here. That is, secondary auxiliary information includes secondary guidance information obtained based on the diagnostic information. As an example, secondary guidance information may include prescription information, treatment information, and management information.

[0129] Prescription information can refer to information about medications (e.g., prescription drugs) recommended to the examinee to maintain or improve cardiovascular disease risk based on diagnostic information. Furthermore, prescription information may include information about the prescribed medication, the period of administration, and the dosage. For example, prescription information may include information about prescriptions for one or more of the following medications: statins (including various formulations such as simvastatin, atorvastatin, and rosuvastatin) as HMG-CoA reductase inhibitors, PCSK9 inhibitors, combination therapy with fibrates, aspirin, bile acid sequestrants and nicotinic acid, omega-3 fatty acids, ezetimibe, and fibrates. Additionally, where the diagnostic information includes the presence of soft and / or hard plaques in the examinee, the prescription information may include information about high-intensity statin therapy or calcium channel blockers or ACE inhibitors for hypertension management.

[0130] Furthermore, treatment information can refer to information recommended to the examinee for future treatment in order to maintain or improve the cardiovascular disease risk based on diagnostic information. For example, supplemental examination information may include information on the examinee's secondary diagnosis or medical treatment. As an example, supplemental examination information may include additional required examinations, information on the hospital / medical personnel capable of performing the additional examinations, and information on the recommended procedures / surgeries.

[0131] In addition, management information may include information on non-medical treatments recommended to the examinee to maintain or improve cardiovascular disease risk based on diagnostic information. For example, management information may include information on lifestyle habits, dietary habits, exercise, and over-the-counter medications such as nutritional supplements to reduce cardiovascular disease risk. Furthermore, if the diagnostic information includes the presence of soft plaques in the examinee, the management information may include information on strict lifestyle modifications (e.g., diet, exercise, smoking cessation).

[0132] Furthermore, when the diagnostic information includes information about the presence of plaques in the subject, the management information may include information about long-term lifestyle improvements. Additionally, in one embodiment, the diagnostic device may be linked to an external monitoring device. Here, the monitoring device can refer to a device that monitors the subject's lifestyle habits or behaviors. For example, the monitoring device may include portable devices, wearable devices, health measurement devices, etc. Furthermore, the monitoring device may be the aforementioned client device. For example, the monitoring device includes a camera unit that captures images of the inside and outside of the eye to obtain images of the inside and outside of the eye.

[0133] Furthermore, the monitoring device can monitor various information such as the subject's activity level, exercise method, exercise time, food intake, intake amount, health supplement intake information, sleep time, sleep habits, heart rate, blood pressure, blood sugar level, body water content, oxygen content, body temperature, blood oxygen saturation, pulse wave, whether medical treatment has been sought, whether examinations have been conducted, whether surgery has been performed, and eye images.

[0134] The processor of the diagnostic device can communicate with the monitoring device via wired or wireless communication through the communication module.

[0135] The processor of the diagnostic device can provide guidance information to the monitoring device. Furthermore, the monitoring device can provide various information to the subject based on the guidance information and the monitored information. For example, the monitoring device obtains management information (e.g., lifestyle, dietary, and exercise information) from the diagnostic device as guidance information, compares the management information with the monitored information, determines whether the monitored information matches the management information, and can provide the determination result and / or additional information based on the determination result.

[0136] For example, if the monitored exercise time is less than the exercise time specified in the management information, the monitoring device can provide the subject with information indicating that the exercise was performed according to the management information. Furthermore, if the monitored food intake corresponds to the food intake specified in the management information, the monitoring device can provide the subject with information indicating that the food intake was performed well according to the management information.

[0137] Furthermore, the processor of the diagnostic device can acquire monitored information from the monitoring device. Based on the guidance information and the monitored information, the processor can provide various information to the patient. For example, the processor can compare the monitored information with the guidance information to determine whether the monitored information matches the management information, and can provide the determination result and / or additional information based on that determination result. The aforementioned example of the monitoring device can be applied to the operation of the processor of the diagnostic device.

[0138] Furthermore, the processor of the diagnostic device can reflect the monitoring information received from the monitoring device and generate guidance information. For example, the processor of the diagnostic device can obtain the examinee's status information (movement status, living status, eating habits, etc.) based on the monitoring information, and based on the examinee's status information, modify the guidance information determined by the diagnostic information to suit the examinee.

[0139] In one embodiment, the processor of the diagnostic device may utilize a predetermined database to provide guidance information. For example, the diagnostic device may include a database that matches diagnostic information scores and / or levels with guidance information. For instance, if the diagnostic information is represented in three levels, the database may include guidance information matching a low-risk level (e.g., prescription information - none, treatment information - information about the next appointment, management information - dietary information provided, exercise information provided), guidance information matching a medium-risk level (e.g., prescription information - none, treatment information - additional examination information provided, management information - dietary information provided, exercise information provided, over-the-counter drug information provided), and guidance information matching a high-risk level (e.g., prescription information - statin prescription information provided, treatment information - additional examination information, recommended procedure / surgery information provided, management information - dietary information provided, exercise information provided, over-the-counter drug information provided). The processor of the diagnostic device may provide guidance information matching the diagnostic information based on the database.

[0140] In another embodiment, the processor of the diagnostic device can utilize a machine learning model to provide guidance information. For example, the diagnostic device may include a guidance information model based on a machine learning model or a neural network model. The guidance information model can learn based on the diagnostic information's score and / or grade, along with the guidance information. Additionally, the guidance information model can also learn for at least one of the examinee's physical information (height, weight, age, gender, ethnicity, smoking status, blood pressure (e.g., blood pressure value, hypertension), diabetes (or blood glucose level), and cholesterol level). Thus, the processor of the diagnostic device can input the diagnostic information's score and / or grade, along with the examinee's physical information, into the guidance information model to obtain guidance information tailored to the examinee. Furthermore, according to embodiments, the guidance information model may be included in the aforementioned diagnostic model or may be constructed independently of the diagnostic model.

[0141] Various embodiments of this specification can be implemented in software including commands stored in machine-readable storage media (e.g., a computer). The machine, as a means of invoking stored commands from the storage media and capable of operating according to the invoked commands, can include electronic devices according to the disclosed embodiments. When the commands are executed by a processor, the processor can directly or under the control of the processor perform the function corresponding to the commands using other components. Commands can include code generated or executed by a compiler or interpreter. Machine-readable storage media can be provided in the form of non-transitory storage media. Here, "non-transitory storage media" simply means excluding signals and being tangible, without distinguishing whether data is stored semi-permanently or temporarily in the storage media. For example, "non-transitory storage media" can include buffers where data is temporarily stored.

[0142] According to one embodiment, the methods involved in the various embodiments disclosed in this specification may be provided in a computer program product. The computer program product can be traded as a commodity between a seller and a buyer. The computer program product may be in the form of a machine-readable storage medium (e.g., Compact DiscRead Only Memory, CD-ROM) or distributed online through an app store (e.g., Play Store™). In the case of online distribution, at least a portion of the computer program product, such as a downloadable application, may be temporarily stored on a storage medium such as the manufacturer's server, the app store's server, or the memory of a relay server, or may be temporarily generated.

[0143] As described above, although the embodiments have been illustrated with limited examples and drawings, those skilled in the art can make various modifications and variations based on the above description. For example, even if the described techniques are performed in a different order than the described methods, and / or the components of the described systems, structures, devices, circuits, etc., are combined or integrated in a different form than the described methods, or are replaced or substituted by other components or equivalents, suitable results can still be achieved.

[0144] Therefore, other implementations, other embodiments, and contents equivalent to the scope of the claims also fall within the scope of the following claims.

Claims

1. A cardiovascular disease type risk prediction and diagnostic device, characterized in that, include: At least one processor, The at least one processor utilizes a diagnostic model to predict the risk of cardiovascular disease types. The diagnostic model includes: The first model is used to analyze input fundus images to predict cardiovascular risk; and The second model is used to receive the cardiovascular risk score and disease type-specific factors predicted by the first model as input, and to predict the risk of cardiovascular disease type. The at least one processor provides diagnostic information based on the cardiovascular risk predicted by the first model and the cardiovascular disease type risk predicted by the second model.

2. The apparatus according to claim 1, characterized in that, The detailed types of cardiovascular diseases include one or more of myocardial infarction (MI), peripheral artery disease (PAD), stroke, atrial fibrillation (AF), and heart failure (HF).

3. The apparatus according to claim 1, characterized in that, The disease type-specific factors include one or more of the following: patient's age, sex, blood biomarkers, urine test values, smoking status, obesity, electrocardiogram (ECG), or patient's family history.

4. The apparatus according to claim 1, characterized in that, The disease type-specific factors are predicted by the first model or input from an external source.

5. The apparatus according to claim 1, characterized in that, The diagnostic information includes the risk level of each cardiovascular type and related information.

6. A cardiovascular disease type risk prediction and diagnostic device, characterized in that, include: At least one processor, The at least one processor utilizes a diagnostic model to provide cardiovascular diagnostic information. The diagnostic model includes: The first model is used to analyze input fundus images to predict cardiovascular risk; The second model receives inputs of the cardiovascular risk score and disease type-specific factors predicted by the first model, and predicts the risk of cardiovascular disease types; and The third model is used to determine the recommended clinical examinations for patients based on the cardiovascular disease type risk predicted by the second model. The at least one processor provides cardiovascular diagnostic information based on the cardiovascular risk predicted by the first model and the cardiovascular disease type risk predicted by the second model.

7. The apparatus according to claim 6, characterized in that, The detailed types of cardiovascular diseases include one or more of myocardial infarction (MI), peripheral artery disease (PAD), stroke, atrial fibrillation (AF), and heart failure (HF).

8. The apparatus according to claim 6, characterized in that, The disease type-specific factors include one or more of the following: patient's age, sex, blood biomarkers, smoking status, obesity, urine test values, electrocardiogram (ECG), or patient's family history.

9. The apparatus according to claim 6, characterized in that, The disease type-specific factors are predicted by the first model or input from an external source.

10. The apparatus according to claim 6, characterized in that, The additional clinical examinations include at least one of the following: electrocardiogram (ECG), echocardiogram, ABI (ankle-brachial index), ABPI (ankle-brachial pressure index), blood test, urine test, and ultrasound.

11. The apparatus according to claim 6, characterized in that, The cardiovascular diagnostic information includes the risk level of each disease type and related information or information related to recommended examinations.

12. A cardiovascular disease type risk prediction and diagnostic device, characterized in that, include: At least one processor, The at least one processor utilizes a diagnostic model to provide cardiovascular diagnostic information. The diagnostic model includes: The first model is used to analyze input fundus images to predict cardiovascular risk and cardiovascular plaque type; The second model receives inputs of the cardiovascular risk score and disease type-specific factors predicted by the first model, and predicts the risk of cardiovascular disease types; and The third model is used to determine the recommended clinical examinations for patients based on the cardiovascular disease type risk predicted by the second model. Cardiovascular diagnostic information is provided based on the predictions made by the first to third models.

13. The apparatus according to claim 12, characterized in that, The detailed types of cardiovascular diseases include one or more of myocardial infarction (MI), peripheral artery disease (PAD), stroke, atrial fibrillation (AF), and heart failure (HF).

14. The apparatus according to claim 12, characterized in that, The disease type-specific factors include one or more of the following: patient's age, sex, blood biomarkers, smoking status, obesity, urine test values, electrocardiogram (ECG), or patient's family history.

15. The apparatus according to claim 12, characterized in that, The disease type-specific factors are predicted by the first model or input from an external source.

16. The apparatus according to claim 12, characterized in that, The additional clinical examinations include at least one of the following: electrocardiogram (ECG), echocardiogram, ABI (ankle-brachial index), ABPI (ankle-brachial pressure index), blood test, urine test, and ultrasound.

17. The apparatus according to claim 12, characterized in that, The cardiovascular diagnostic information includes the risk level and related information for each type of cardiovascular disease, plaque type, or information related to recommended examinations.

18. The apparatus according to claim 15, characterized in that, When disease type-specific factors are input from outside the model, the risk of cardiovascular disease type is reassessed based on the disease type factors obtained from the outside.

19. A cardiovascular disease type risk prediction and diagnostic device, characterized in that, include: At least one processor, The at least one processor utilizes a diagnostic model to predict the risk of cardiovascular disease types. The diagnostic model includes: The first model is used to analyze input fundus images to predict cardiovascular risk; and The second model is used to receive the cardiovascular risk score predicted by the first model and the detailed subtype-specific factors of the disease type as input, and to predict the detailed subtype risk of the disease type. The at least one processor provides diagnostic information based on the cardiovascular risk predicted by the first model and the detailed subtype risk of the type predicted by the second model.

20. The apparatus according to claim 19, characterized in that, The cardiovascular disease types include one or more of myocardial infarction (MI), peripheral artery disease (PAD), stroke, atrial fibrillation (AF), and heart failure (HF).

21. The apparatus according to claim 20, characterized in that, The heart failure includes one or more subtypes of heart failure with reduced left ventricular ejection fraction (HFrEF), heart failure with mildly reduced left ventricular ejection fraction (HFmrEF), or heart failure with preserved left ventricular ejection fraction (HFpEF). The stroke includes one or more subtypes of ischemic stroke, thrombotic stroke (cerebral thrombosis), cerebral embolism (embolic stroke), lacunar infarction, hemorrhagic stroke, intracerebral hemorrhage, subarachnoid hemorrhage, transient ischemic attack, brainstem stroke, and stroke of unknown cause. The atrial fibrillation includes one or more subtypes of paroxysmal atrial fibrillation, persistent atrial fibrillation, long-term persistent atrial fibrillation, or permanent atrial fibrillation. The myocardial infarction includes one or more subtypes of type 1 (spontaneous myocardial infarction), type 2 (myocardial infarction due to ischemic imbalance), type 3 (myocardial infarction leading to death when biomarker values ​​are unavailable), type 4a (myocardial infarction associated with percutaneous coronary intervention (PCI), type 4b (myocardial infarction associated with stent thrombosis), or type 5 (myocardial infarction associated with coronary artery bypass grafting (CABG)).

Images