Method and device for diagnosing kidney disease

Abstract

The invention discloses a kidney disease diagnosis method and device. A control method of a diagnostic apparatus according to one embodiment includes: a step of acquiring a retinal image of a subject; based on the retina image, kidney disease diagnosis information about the object is obtained through a machine learning model, the machine learning model comprises a first model and a second model, the first model is a neural network model, and the second model can be a regression-based machine learning model.

Description

(19) State Intellectual Property Office (12) Invention Patent Application (10) Application Publication Number (43) Application Publication Date (21) Application Number 202380100669.5 (22) Application Date 2023.11.01 (30) Priority Data 10-2023-0074433 2023.06.09 KR (85) PCT International Application Entering National Phase Date 2026.01.21 (86) PCT International Application Application Data PCT / KR2023 / 017322 2023.11.01 (87) PCT International Application Publication Data WO2024 / 253266 KO 2024.12.12 (71) Applicant Medical Whale Co., Ltd. Address Seoul, South Korea (72) Inventors Choi Tae-geun, Lee Geun-young, Lim Hyung-taek, Park Jung-tak, Joo Young-soo (74) Patent Agency Beijing Gaowo Law Firm 11569 Patent Attorney Li Qiao (51) Int.Cl. A61B 3 / 00 (2006.01) A61B 3 / 12 (2006.01) A61B 3 / 14 (2006.01) G16H 50 / 20 (2006.01) G16H 30 / 20 (2006.01) (54) Invention Title: Method and Apparatus for Diagnosing Kidney Disease (57) Abstract: This invention discloses a method and apparatus for diagnosing kidney disease. According to one embodiment, the control method of the diagnostic apparatus includes: a step of acquiring a retinal image of a subject; and a step of acquiring kidney disease diagnostic information about the subject using a machine learning model based on the retinal image, wherein the machine learning model includes a first model and a second model, the first model being a neural network model, and the second model being a regression-based machine learning model. Claims 2 pages, Description 43 pages, Drawings 18 pages, CN 121586537 A 2026.02.27 CN 1 21 58 65 37 A 1. A control method for a diagnostic device, characterized in that it includes: a step of acquiring a retinal image of a subject; and a step of acquiring diagnostic information about kidney disease of the subject based on the retinal image using a machine learning model, wherein the machine learning model includes a first model and a second model, the first model being a neural network model and the second model being a regression-based machine learning model. 2. The control method for a diagnostic device according to claim 1, characterized in that the first model is trained based on first training data and result values ​​corresponding to the first training data based on a first biomarker, the first training data including multiple retinal images, and in the step of acquiring diagnostic information about kidney disease of the subject, the retinal image is input into the first model, and a first result value is obtained from the first model.The first score and the physical information of the subject are input into the second model, a second result value is obtained from the second model, and the kidney disease diagnosis information is obtained based on the second result value. 3. The control method of the diagnostic device according to claim 2, wherein the first result value represents at least one of the probability that the eGFR (Estimated Glomerular Filtration Rate) value is equal to or less than a first value, or information about the presence of albuminuria in the urine of the subject, wherein the first result value represents at least one of the probability that the eGFR value of the subject is equal to or less than the first value, or the probability that albuminuria is present in the urine of the subject. 4. The control method of the diagnostic device according to claim 3, wherein the second machine learning model is trained based on second training data, the second training data including kidney disease event tracking results of objects in a predetermined population. 5. The control method of the diagnostic device according to claim 4, wherein data about objects in the predetermined population whose eGFR value is less than a second value are excluded from the second training data, and the second value is higher than the first value. 6. The control method of the diagnostic device according to claim 1, wherein the first result value includes information about the current risk of kidney disease in the subject, and the second result value includes information about the probability of future kidney disease-related events occurring in the subject. 7. The control method of the diagnostic device according to claim 6, wherein the second result value includes information about the probability of kidney disease-related events occurring in the subject within 5 years. 8. The control method of the diagnostic device according to claim 1, wherein in the step of obtaining kidney disease diagnostic information about the subject, a predetermined cutoff value is applied to the second result value as the kidney disease diagnostic information, and any one of multiple grades corresponding to the subject is obtained. 9. The control method of the diagnostic device according to claim 8, wherein the cutoff value is set based on the population distribution in multiple grade groups divided according to the tracking observation results of a predetermined population. 10. The control method of the diagnostic device according to claim 1, further comprising the step of providing guidance information to the subject based on the kidney disease diagnostic information. 11. The control method of the diagnostic device according to claim 10, characterized in that, in the step of providing guidance information to the subject based on the kidney disease diagnostic information, guidance information corresponding to the kidney disease diagnostic information is provided using a pre-stored database or a machine learning model.12. The control method of the diagnostic device according to claim 10, characterized in that it further includes the step of determining the rate of progression of the kidney disease in the subject based on the kidney disease diagnostic information, wherein in the step of determining the rate of progression of the kidney disease in the subject based on the kidney disease diagnostic information, the result value of the subject based on a biomarker is obtained, and the rate of progression of the kidney disease in the subject is determined using the result value based on the biomarker and the second result value. 13. The control method of the diagnostic device according to claim 12, characterized in that when the second result value is greater than or equal to a predetermined value, it is determined that the rate of progression of the kidney disease is faster than the rate of progression of the kidney disease expected based on the result value of the biomarker; when the second result value is less than the predetermined value, it is determined that the rate of progression of the kidney disease is slower than the rate of progression of the kidney disease expected based on the result value of the biomarker. 14. The control method of the diagnostic device according to claim 12, characterized in that in the step of providing guidance information to the subject based on the kidney disease diagnostic information, the guidance information is provided to the subject using the rate of progression of the kidney disease in the subject and the kidney disease diagnostic information. 15. A recording medium having a program for performing the method of any one of claims 1 to 14. 16. A diagnostic apparatus, characterized in that it comprises: a storage module; and at least one processor, the at least one processor being configured to: acquire a retinal image of a subject; and, based on the retinal image, acquire diagnostic information about kidney disease of the subject using a machine learning model stored in the storage module; the machine learning model comprising a first model and a second model, the first model being a neural network model and the second model being a regression-based machine learning model. Claims 2 / 2 Page 3 CN 121586537 A Method and Apparatus for Diagnosing Kidney Disease Technical Field

[0001] The present invention relates to a method and apparatus for diagnosing kidney disease. Background Art

[0002] Retinal examination can observe abnormalities in the retina, optic nerve, and macula, and the results can be confirmed relatively easily through imaging, thus it is a frequently used diagnostic data in ophthalmology. On the other hand, in recent years, due to the rapid development of artificial intelligence technology, the development of diagnostic artificial intelligence is actively underway in the field of medical diagnosis, especially in the field of image-based diagnosis. Global companies are also investing heavily in the development of artificial intelligence to analyze various medical imaging data through collaborations with the medical community and the provision of large-scale data. Some companies have even successfully developed AI diagnostic tools that output excellent diagnostic results.

[0003] Retinal images can be used to non-invasively observe blood vessels in the body, thus expanding the application of retinal image diagnosis not only for ophthalmological diseases but also for kidney diseases. Summary of the Invention

[0004] Technical Problem

[0005] The technical problem of this application is to provide a method for diagnosing kidney disease by utilizing retinal images and machine learning models to obtain information about kidney disease with high accuracy.

[0006] The technical problem of this application is not limited to the above-mentioned problem. Problems not mentioned can be clearly understood by those skilled in the art from this specification and the accompanying drawings.

[0007] Solution to the Problem

[0008] According to one aspect, a control method of a diagnostic device according to one embodiment includes: the step of acquiring a retinal image of a subject; and the step of acquiring kidney disease diagnostic information about the subject using a machine learning model based on the retinal image. The machine learning model may include a first model and a second model, wherein the first model may be a neural network model and the second model may be a regression-based machine learning model.

[0009] The technical solution is not limited to the above-mentioned solution. Technical solutions not mentioned can be clearly understood by those skilled in the art from this specification and the accompanying drawings.

[0010] Effects of the Invention

[0011] According to this application, information about kidney disease can be obtained with high accuracy by utilizing retinal images and machine learning models.

[0012] The effects of this invention are not limited to the effects described above. Effects not mentioned will be clearly understood by those skilled in the art to which this application pertains from this specification and the accompanying drawings. Brief Description of the Drawings

[0013] FIG1 illustrates a diagnostic system according to an embodiment.

[0014] FIG2 is a block diagram illustrating a learning device according to an embodiment.

[0015] FIG3 is a block diagram illustrating a diagnostic device according to an embodiment. Specification 1 / 43 pages 4 CN 121586537 A

[0016] FIG4 illustrates a diagnostic system according to an embodiment.

[0017] FIG5 is a block diagram illustrating a client device according to an embodiment of the present invention.

[0018] FIG6 is a diagram illustrating a diagnostic process according to an embodiment of the present invention.

[0019] FIG7 is a diagram illustrating the configuration of a learning unit according to an embodiment of the present invention.

[0020] FIG8 is a conceptual diagram illustrating an image dataset according to an embodiment of the present invention.

[0021] FIG9 is a block diagram illustrating a diagnostic model learning process according to an embodiment of the present invention.

[0022] FIG10 is a diagram illustrating the configuration of a diagnostic unit according to an embodiment of the present invention.

[0023] FIG11 is a diagram illustrating a diagnostic process according to an embodiment of the present invention.

[0024] FIG12 is a block diagram illustrating a diagnostic unit according to an embodiment of the present invention.

[0025] FIG13 is a diagram illustrating a diagnostic process according to an embodiment of the present invention.

[0026] FIG14 is a diagram illustrating a diagnostic system according to an embodiment of the present invention.

[0027] Figure 15 is a diagram illustrating a serial diagnostic model according to one embodiment.

[0028] Figure 16 is a diagram illustrating a serial diagnostic model according to another embodiment.

[0029] Figure 17 is a diagram illustrating a serial diagnostic model according to yet another embodiment.

[0030] Figure 18 is a diagram illustrating a diagnostic method using a diagnostic model according to one embodiment.

[0031] Figure 19 is a diagram illustrating a kidney disease diagnostic method according to one embodiment.

[0032] Figure 20 illustrates a diagnostic model for obtaining kidney disease diagnostic information according to one embodiment.

[0033] Figure 21 shows the clinical characteristics of the subject according to Embodiment 1.

[0034] Figures 22a and 22b are graphs illustrating the incidence of kidney disease-related events based on kidney disease diagnostic information according to Embodiment 1.

[0035] Figure 23 is a graph illustrating the predictive performance of kidney disease diagnostic information according to Embodiment 1 for the occurrence of kidney disease-related events within 5 years.

[0036] Figure 24 is a diagram illustrating a kidney disease diagnostic method according to another embodiment.

[0037] FIG25 is a diagram illustrating a method for providing guidance information on kidney disease diagnosis information according to one embodiment.

[0038] FIG26 is a diagram illustrating a method for providing guidance information using existing prescription information and kidney disease diagnosis information according to one embodiment.

[0039] FIG27 is a diagram illustrating a method for predicting the rate of progression of kidney disease using kidney disease diagnosis information according to one embodiment.

[0040] FIG28 is a diagram illustrating the clinical characteristics of an object according to Embodiment 2.

[0041] FIG29a and FIG29b are diagrams illustrating the incidence of kidney disease-related events using KDIGO classification and kidney disease diagnosis information according to Embodiment 2. Detailed Description

[0042] Best Mode for Carrying Out the Invention

[0043] A method for controlling a diagnostic device according to one embodiment includes: the step of acquiring a retinal image of a subject; and the step of acquiring kidney disease diagnosis information about the subject using a machine learning model based on the retinal image, the machine learning model including a first model and a second model, the first model being a neural network model and the second model being a regression-based machine learning model. Specification 2 / 43 pages 5 CN 121586537 A

[0044] Modes for Carrying Out the Invention

[0045] The above-mentioned objects, features and advantages of the present invention will become clearer from the following detailed description in relation to the accompanying drawings. However, the present invention can have many modifications and various embodiments, and therefore specific embodiments will be illustrated and described in detail below.

[0046] In the drawings, the thickness of layers and regions is exaggerated for clarity, and when an element or layer is referred to as inWhen another element or layer is "on" or "on," it includes cases where it is not only directly on another element or layer, but also with other layers or other elements intervening in between. Throughout the specification, the same reference numerals generally denote the same constituent elements. Furthermore, the same reference numerals are used to describe the same constituent elements whose functions are within the same conceptual scope, appearing in the figures of various embodiments.

[0047] Detailed descriptions of well-known functions or constituent elements related to the present invention are omitted if it is determined that such detailed descriptions may unnecessarily obscure the gist of the invention. Furthermore, the numbers used in the description (e.g., first, second, etc.) are merely identification symbols used to distinguish one constituent element from another.

[0048] Furthermore, the suffixes "module" and "part" used for constituent elements in the following description are merely assigned or used interchangeably for convenience in writing the specification and do not inherently have a distinguishing meaning or function.

[0049] The method according to the embodiments can be implemented in the form of program instructions executable by various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., individually or in combination. The program instructions recorded on the medium may be specifically designed and configured for the embodiments, or may be known and usable by those skilled in the art of computer software. Examples of computer-readable recording media include: magnetic media such as hard disks, floppy disks and magnetic tapes; optical media such as CD-ROMs and DVDs; magnetic-optical media such as floppy disks; and hardware devices specifically configured for storing and executing program instructions, such as ROMs, RAMs, flash memory, etc. Examples of program instructions include machine language code generated by a compiler, and high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to run as one or more software modules to perform the operations of the embodiments, and vice versa.

[0050] 1. Diagnosis using retinal images

[0051] 1.1. Diagnostic system and process

[0052] 1.1.1. Purpose and definition

[0053] The following will describe a diagnostic system and method that assists medical personnel in determining the presence of a disease or the presence of abnormalities as a basis for judgment based on ocular images. In this specification, the term "diagnosis" may mean diagnostic aid to assist in the diagnosis of a disease, rather than a direct diagnosis of the disease. The term "diagnosis" is used hereinafter for ease of explanation, but it can mean diagnostic aid.

[0054] In particular, it will be described how to construct a machine learning model for diagnosing diseases using deep learning techniques, and how to utilize...The constructed model assists in the diagnostic method of detecting the presence or abnormality of diseases. In this specification, an eye image is an image including the eye of the subject, and may include various images such as retinal images and / or fundus images. In this specification, for the sake of convenience, the techniques described herein will focus on retinal images, but are not limited thereto, and the descriptions in this specification can of course be applied to other eye images.

[0055] The machine learning models described in this specification can be designed based on various machine learning libraries. For example, machine learning models can refer to various forms of models designed based on supervised, unsupervised, semi-supervised, or reinforcement learning artificial intelligence algorithms, such as decision trees, random forest algorithms, stochastic gradient descent algorithms, neural network algorithms, k-nearest neighbors algorithms, linear regression, logistic regression, Cox proportional hazards model (survival model) (based on regression), support vector machines, k-means, hierarchical cluster analysis (HCA), expectation maximization, principal component analysis (PCA), kernel PCA, locally-linear embedding (LLE), t-SNE (t-distributed stochastic neighbor embedding), Apriori, Eclat, etc.

[0056] Unless otherwise specified, the machine learning model is mainly described below using neural network models, but it does not necessarily refer only to models based on neural network algorithms. Within the scope of the functions and purposes of the invention as described in this specification, it can be replaced by models based on other algorithms, which is obvious.

[0057] According to one embodiment of the invention, a diagnostic system or method based on retinal image-assisted diagnosis of at least one ophthalmic disease, cardiovascular disease (and / or cerebrovascular disease), kidney disease, or other systemic disease can be provided.

[0058] For example, in this specification, ophthalmic diseases may include cataracts, glaucoma, macular degeneration, diabetic retinopathy, epiretinal membrane, macular hole, high / degenerative myopia, melanoma, and retinal detachment.Detachment, dry eye syndrome, presbyopia, and astigmatism.

[0059] In addition, cardiovascular diseases may include at least one of coronary artery disease (CAD), aortic stenosis, hypertension, arterial hypertension, heart failure, arrhythmia, atrial fibrillation, valvular heart disease, cardiomyopathy, peripheral vascular disease, peripheral artery disease (PAD), heart attack, aneurysm, aortic aneurysm (e.g., abdominal aortic aneurysm, thoracic aortic aneurysm), thrombotic diseases (e.g., deep vein thrombosis, pulmonary embolism), myocardial infarction, and cardiovascular and cerebrovascular diseases. In addition, cardiovascular and cerebrovascular diseases can include at least one of stroke, ischemic stroke, cerebral infarction, cerebral hemorrhage, subarachnoid hemorrhage, transient ischemic attack, and death due to cardiovascular disease. Cardiovascular diseases can include complications. In addition, complications can include 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, bedsores, falls, pain, attacks, depression, etc.

[0060] In addition, kidney diseases can include at least one of chronic kidney disease (CKD), acute kidney injury (AKI), kidney stones, nephrotic syndrome, glomerulonephritis, polycystic kidney disease (PKD), kidney cancer, and pyelonephritis. In addition, kidney diseases can include complications. In addition, complications may include dialysis-related side effects, hypotension, muscle cramps, nausea and vomiting, headache, dialysis disequilibrium syndrome, itching, and arrhythmia.

[0061] Furthermore, other systemic diseases may include diabetes, hypertension, hypotension, Alzheimer's disease, and cytomegalovirus disease.At least one of the following: toxins, arteriosclerosis.

[0062] Furthermore, according to another embodiment of the present invention, various parameters of the subject can be predicted based on retinal images. For example, the parameters may include at least one of the parameters representing the subject's physical information, such as biological age, sex, height, weight, BMI index, body mass, body fat percentage, body muscle mass, etc. In addition, the parameters may include at least one of the diagnostic numerical parameters, such as hematocrit, red blood cell count, white blood cell count, hemoglobin value, platelet count, total iron-binding capacity (TIBC), iron value, ferritin (storage ferritin) value, total protein value, albumin value, aspartate aminotransferase (AST) value, aminotransferase value, γ-GTP, γ-GT, alkaline phosphatase (ALP) value, globulin value, hepatitis antigen value, hepatitis antibody value, glycated hemoglobin value (HbA1c), blood urea nitrogen (BUN) value, creatinine value, uric acid value, total cholesterol value, high-density lipoprotein (HDL) cholesterol value, low-density lipoprotein (LDL) cholesterol value, triglyceride (TG) value, bicarbonate value, systolic blood pressure (SBP), diastolic blood pressure (DBP), etc.

[0063] According to another embodiment of the present invention, a diagnostic system or method may be provided for detecting abnormal retinal features that can be used to diagnose ophthalmic diseases or other conditions. For example, a diagnostic system or method may be provided to obtain information on the following: color abnormalities of the entire retina, lens opacity, abnormal cup-to-disc ratio of the optic disc, macular abnormalities (e.g., macular hole), abnormalities in vessel diameter and course, abnormalities in retinal artery diameter, retinal hemorrhage, microaneurysms, hard exudates, epiretinal membrane, myelinated nerve fibers, chorioretinal atrophy, retinal nerve fiber layer defect (RNFL defect), exudation, drusen, cataracts, glaucoma, diabetic retinopathy, tessellated fundus, large optic cup, retinal vein occlusion (RVO), branch retinal vein occlusion (BRVO), central retinal vein occlusion (CRVO), retinal artery occlusion (RAO), rhegmatogenous retinal detachment, and posterior serous / exudative retinal detachment. serous / exudativeRD), central serous chorioretinopathy (CSCR), VKH disease, macular disease, epiretinal membrane (ERM), macular hole (MH), pathological myopia, optic nerve degeneration, optic atrophy, severe hypertensive retinopathy, disc swelling and elevation, traction disc, pigmentary degeneration, congenital disc abnormality, retinitis pigmentosa, bietti crystalline dystrophy, peripheral retinal degeneration and break, myelinated nerve fiber, vitreous particles, fundus neoplasm, and massive hard exudates. Exudates), yellow-white spots / flecks, cotton-wool spots, vascular tortuosity, chorioretinal atrophy / coloboma, preretinal hemorrhage, fibrosis, laser spots, silicone oil in the eye, blurred fundus, blurred fundus without PDR, blurred fundus with suspected PDR, etc.

[0064] In this specification, diagnostic information can be understood as including diagnostic information based on the determination of the presence or absence of a disease, or the observed information as the basis therefor.

[0065] 1.1.2. Diagnostic System Composition

[0066] According to an embodiment of the present invention, a diagnostic system can be provided.

[0067] FIG1 illustrates a diagnostic system according to an embodiment of the present invention. Referring to FIG1, the diagnostic system 1 may include: a learning device 10 for training a diagnostic model, a diagnostic device 20 for performing diagnosis using the diagnostic model, and a client device 30 for obtaining a diagnostic request. The diagnostic system 1 may include multiple learning devices, multiple diagnostic devices, or multiple client devices.

[0068] The learning device 10 may include a learning unit 100. The learning unit 100 may perform training of a diagnostic model. For example, the learning unit 100 may acquire a retinal image dataset and perform training of a diagnostic model for detecting diseases or abnormalities seen in retinal images. The learning unit 100 may be included in the processor of the learning device 10 described later, meaning that the processor learns the functional expression of the processing described in the specification 5 / 43 page 8 CN 121586537 A.

[0069] The diagnostic device 20 may include a diagnostic unit 200. The diagnostic unit 200 may perform disease diagnosis or obtain auxiliary information for diagnosis using a diagnostic model. For example, the diagnostic unit 200 can use a diagnostic model trained by the learning unit to acquire diagnostic information. The diagnostic unit 200 may be included in the processor of the diagnostic device 20 described later, meaning a functional expression of the processor's diagnostic processing.

[0070] The client device 30 may include an imaging unit 300. The imaging unit 300 can capture retinal images. The client device may be an ophthalmic retinal imaging device. Alternatively, the client device 30 may be a handheld device such as a smartphone or tablet.

[0071] In the diagnostic system 1 of this embodiment, the learning device 10 can determine a diagnostic model for diagnosis by acquiring a dataset and performing diagnostic model learning. When the diagnostic device obtains an information request from the client, it uses the determined diagnostic model to acquire diagnostic information corresponding to the diagnostic target image. The client device can request information from the diagnostic device and obtain the diagnostic information transmitted in response.

[0072] A diagnostic system according to another embodiment may include: a diagnostic device that performs diagnostic model learning and performs diagnosis using the model, and a client device. A diagnostic system according to yet another embodiment may include: a diagnostic device that performs diagnostic model learning, obtains diagnostic requests, and performs diagnosis. A diagnostic system according to yet another embodiment may include: a learning device for performing diagnostic model learning, and a diagnostic device for acquiring a diagnostic request and performing a diagnosis.

[0073] The diagnostic system disclosed in this specification is not limited to the above embodiments and can be implemented in any form, including: a learning unit for performing model learning, a diagnostic unit for acquiring diagnostic information based on the learned model, and an imaging unit for acquiring a diagnostic target image.

[0074] Some embodiments of the devices constituting the system will be described below.

[0075] 1.1.2.1. Learning Device

[0076] A learning device according to an embodiment of the present invention can perform training of a diagnostic model for assisted diagnosis.

[0077] FIG2 is a block diagram illustrating a learning device according to an embodiment of the present invention. Referring to FIG2, the learning device 10 may include a processor 12 and a storage module 11.

[0078] The learning device 10 may include a processor 12. The processor 12 can control the operation of the learning device 10.

[0079] The processor 12 may include one or more of a central processing unit (CPU), random access memory (RAM), graphics processing unit (GPU), one or more microprocessors, and other electronic components that process input data according to predetermined logic. In addition, the processor 12 may be one or more.

[0080] The processor 12 can read system programs and various processing programs stored in the storage module 11. For example, the processor 12 can expand the processes, methods, etc. for performing the diagnosis described later onto RAM and perform various processing according to the expanded program. The processor 12 can perform learning of the diagnostic model described later.

[0081] The learning device 10 may include a storage module 11. The storage module 11 can store the data and learning model required for learning.

[0082] The storage module 11 may be implemented in the form of a non-volatile semiconductor memory, a hard disk, flash memory, RAM, read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), or other tangible non-volatile recording media.

[0083] The storage module 11 may store various processing programs, parameters for performing program processing, or such processing result data. For example, the storage module 11 may store a data processing flow program for performing the diagnosis described later, a diagnostic flow program, parameters for performing each program, and data obtained by performing such programs (e.g., processed data or diagnostic result values). In addition, the storage module 11 may store various diagnostic models described later. Specification 6 / 43 pages 9 CN 121586537 A

[0084] The learning device 10 may include a separate learning unit. The learning unit may perform the learning of diagnostic models.

[0085] The learning unit may be included in the processor 12 described above. The learning unit may be stored in the storage module 11 described above. The learning unit may be implemented by a portion of the processor 12 and the storage module 11 described above. For example, the learning unit can be stored in the storage module 11 and driven by the processor 12.

[0086] The learning device 10 may further include a communication module 13. The communication module 13 can communicate with external devices. For example, the communication module 13 can communicate with a diagnostic device, a server device, or a client device described later. The communication module 13 can perform wired or wireless communication. The communication module 13 can perform bidirectional or unidirectional communication.

[0087] 1.1.2.2. Diagnostic Device

[0088] A diagnostic device can acquire diagnostic information using a diagnostic model.

[0089] FIG3 is a block diagram illustrating a diagnostic device according to an embodiment of the present invention. Referring to FIG3, the diagnostic device 20 may include a processor 22 and a storage module 21.

[0090] The processor 22 may include one or more of a central processing unit (CPU), random access memory (RAM), graphics processing unit (GPU), one or more microprocessors, and other electronic components that process input data according to predetermined logic. Furthermore, the processor 22 may be one or more.

[0091] The processor 22 can read system programs and various processing programs stored in the storage module 21. For example, the processor 22 can expand processes, methods, etc., for performing the diagnostics described later onto RAM and perform various processing according to the expanded program. The processor 22 can generate diagnostic information using a diagnostic model. The processor 22 can acquire diagnostic data (e.g., retinal data of the subject) for diagnosis and acquire diagnostic information predicted from the diagnostic data using a learned diagnostic model.

[0092] The storage module 21 can store the diagnostic model. The storage module 21 can store parameters, variables, etc., of the diagnostic model.

[0093] The storage module 21 may be implemented in the form of a non-volatile semiconductor memory, hard disk, flash memory, RAM, read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), or other tangible non-volatile recording media.

[0094] The storage module 21 may store various processing programs, parameters for performing program processing, or such processing result data. For example, the storage module 21 may store a data processing flow program for performing the diagnosis described later, a diagnostic flow program, parameters for performing each program, and data obtained by performing such programs (e.g., processed data or diagnostic result values). In addition, the storage module 21 may store various diagnostic models described later.

[0095] Although not shown, the diagnostic device 20 may further include an input module. The input module may acquire user input. For example, the input module may acquire user input requesting diagnostic information. In addition, the input module may acquire at least one of the following: height, weight, age, gender, race, smoking status, blood pressure (e.g., blood pressure value, hypertension status), diabetes status (or blood glucose value), and cholesterol value. The processor 22 can acquire information input through the input module.

[0096] The diagnostic device 20 may further include a communication module 23. The communication module 23 can communicate with the learning device and / or the client device. For example, the diagnostic device 20 may be configured as a server communicating with the client device. This will be described in more detail below.

[0097] 1.1.2.3. Server Device

[0098] According to one embodiment of the present invention, the diagnostic system may include a server device. The diagnostic system according to one embodiment of the present invention may also include multiple server devices.

[0099] The server device may store and / or drive a diagnostic model. The server device may store weight values ​​constituting the learned diagnostic model. The server device may collect or store data for diagnosis.

[0100] The server device may output the results of a diagnostic process using the diagnostic model to a client device. The server device may obtain feedback from the client device. The server device may operate similarly to the diagnostic device described above. Specification 7 / 43 pages 10 CN 121586537 A

[0101] FIG4 illustrates a diagnostic system according to an embodiment of the present invention. Referring to FIG4, a diagnostic system 20 according to an embodiment of the present invention may include a diagnostic server 40, a learning device, and a client device.

[0102] The diagnostic server 40, i.e., the server device, may communicate with multiple learning devices or multiple diagnostic devices. Referring to FIG6, the diagnostic server 40 may communicate with a first learning device 10a and a second learning device 10b. Referring to FIG4, the diagnostic server 40 may communicate with a first client device 30a and a second client device 30b.

[0103] For example, the diagnostic server 40 may communicate with a first learning device 10a for training a first diagnostic model for acquiring first diagnostic information and a second learning device 10b for training a second diagnostic model for acquiring second diagnostic information.

[0104] The diagnostic server 40 may store a first diagnostic model for acquiring first diagnostic information and a second diagnostic model for acquiring second diagnostic information, and in response to a diagnostic information acquisition request from a first client device 30a or a second client device 30b, acquire diagnostic information and transmit the acquired diagnostic information to the first client device 30a or the second client device 30b.

[0105] Alternatively, the diagnostic server 40 may also communicate with a first client device 30a that requests first diagnostic information and a second client device 30b that requests second diagnostic information.

[0106] 1.1.2.4. Client Device

[0107] The client device may request diagnostic information from the diagnostic device or the server device. The client device may acquire data required for diagnosis and transmit the acquired data to the diagnostic device.

[0108] FIG5 is a block diagram illustrating a client device according to an embodiment of the present invention. Referring to FIG5, a client device 30 according to an embodiment of the present invention may include an imaging module 31, a processor 32, and a communication module 33.

[0109] The imaging module 31 can acquire image or video data. The imaging module 31 can acquire retinal images. However, the client device 30 may also be replaced by a data acquisition unit other than the imaging module 31.

[0110] The communication module 33 can communicate with external devices, such as diagnostic devices or server devices. The communication module 33 can perform wired or wireless communication.

[0111] The processor 32 can control the imaging module 31 to acquire images or data. The processor 32 can control the imaging module 31 to acquire retinal images. The processor 32 can transmit the acquired retinal images to the diagnostic device. The processor 32 can transmit the images acquired by the imaging module 31 to the server device through the communication module 33 and obtain diagnostic information generated therefrom.

[0112] Although not shown, the client device may further include an output module. The output module may include a display that outputs video or images or a speaker that outputs sound. The output module may output video or image data acquired by the imaging unit. The output module may output diagnostic information acquired from the diagnostic device.

[0113] Although not shown, the client device may further include an input module. The input module can acquire user input. For example, the input module can acquire user input requesting diagnostic information. The input module can acquire user information for evaluating the diagnostic information acquired from the diagnostic device. In addition, the input module can acquire at least one of the following physical information of the subject: height, weight, age, gender, race, smoking status, blood pressure (e.g., blood pressure value, hypertension status), diabetes status (or blood glucose value), and cholesterol value.

[0114] In addition, although not shown, the client device may further include a storage module. The storage module can store images acquired by the imaging unit.

[0115] 1.1.3. Overview of the diagnostic process

[0116] The diagnostic system or diagnostic device disclosed in this specification can perform a diagnostic process. The diagnostic process can be broadly divided into a learning process for learning a diagnostic model for diagnosis and a diagnostic process using the diagnostic model. Specification 8 / 43 pages 11 CN 121586537 A

[0117] FIG6 is a diagram illustrating a diagnostic process according to an embodiment of the present invention. Referring to FIG6, a diagnostic process according to an embodiment of the present invention may include: a learning process of acquiring and processing data S11, learning a diagnostic model S12, and acquiring parameters of the learned diagnostic model S13; and a diagnostic process of acquiring diagnostic target data S21 and acquiring diagnostic information S23 using a diagnostic model learned based on the diagnostic target data S22.

[0118] More specifically, the learning process may include: a data processing process of processing input learning image data into a state that can be used for model training, and a learning process of training a model using the processed data. The learning process may be executed by the above-described learning device.

[0119] The diagnostic process may include: a data processing process of processing input examination target image data into a state that can be used for diagnosis using a diagnostic model, and a diagnostic process of performing diagnosis using the processed data. The diagnostic process may be executed by the above-described diagnostic device or server device.

[0120] The various processes will be described below.

[0121] 1.2. Learning Process

[0122] According to an embodiment of the present invention, a process for training a diagnostic model can be provided. Specifically, a training process for a diagnostic model that performs or assists in diagnosis based on retinal images can be disclosed.

[0123] The learning process described below can be executed by the learning device described above.

[0124] 1.2.1. Learning Unit

[0125] According to an embodiment of the present invention, the learning process can be executed by a learning unit. The learning unit can be provided in the learning device described above.

[0126] FIG7 is a diagram for illustrating the configuration of a learning unit according to an embodiment of the present invention. Referring to FIG7, the learning unit 100 may include a data processing module 110, a queue module 130, a learning module 150, and a learning result acquisition module 170. Each module can execute the various steps of the data processing process and the learning process described later. However, the constituent elements and the functions performed by each element described in FIG7 are not all necessary; depending on the learning form, some elements may be added or omitted.

[0127] 1.2.2. Data Processing Flow

[0128] 1.2.2.1. Image Data Acquisition

[0129] According to an embodiment of the present invention, a dataset can be acquired. According to an embodiment of the present invention, the data processing module can acquire the dataset.

[0130] The dataset can be an image dataset.

[0131] For example, it can be a retinal image dataset. The retinal image dataset can be acquired using a general non-mydriatic retinal camera, etc. The retinal image can be a panoramic retinal image or a wide-field retinal image. The retinal image can be a red-free image. The retinal image can be an image captured by infrared light. The retinal image can be an image captured by autofluorescence. The image data can be acquired in any of the following formats: JPG, PNG, DCM (DICOM), BMP, GIF, TIFF.

[0132] For example, the dataset can be an image dataset, including any of the following: optical coherence tomography (OCT) images, OCT angiography images, or retinal angiography images. At this point, a diagnostic model learned using a dataset including OCT images, OCT angiography images, or retinal angiography images can predict or output diagnostic information (or labels) based on a target OCT image, target OCT angiography image, or target retinal angiography image.

[0133] The dataset may include a training dataset. The dataset may include a test dataset. The dataset may include a validation dataset. In other words, the dataset may be assigned to at least one of the training dataset, the test dataset, and the validation dataset.

[0134] The dataset may be determined based on the diagnostic information to be obtained using the diagnostic model learned through the dataset. Example: Specification 9 / 43, page 12CN 121586537 A For example, if you want to train a diagnostic model to obtain diagnostic information related to cataracts, the dataset you obtain can be determined as an infrared retinal image dataset. Or, if you want to train a diagnostic model to obtain diagnostic information related to macular degeneration, the dataset you obtain can be a dataset of retinal images taken with autofluorescence.

[0135] Individual data included in the dataset may include labels. There may be multiple labels. In other words, individual data included in the dataset may be labeled for at least one feature. For example, the dataset is a retinal image dataset that includes multiple retinal image data, each retinal image data may include diagnostic information labels (e.g., presence or absence of a specific disease) and / or seen information (e.g., abnormality or absence of a specific area) labels corresponding to the image.

[0136] For another example, the dataset is a retinal image dataset, each retinal image data may include peripheral information labels about the image. For example, each retinal image data may include: left or right eye information about whether the retinal image is a left or right eye image, gender information about whether it is a female or male retinal image, age information about the age of the subject who took the retinal image, and other peripheral information labels.

[0137] FIG8 is a conceptual diagram illustrating an image dataset according to an embodiment of the present invention. Referring to FIG8, an image dataset DS according to an embodiment of the present invention may include multiple image data (IDs). Each image data (ID) may include an image (I) and a label (L) assigned to the image. Referring to FIG10, the image dataset DS may include a first image data ID1 and a second image data ID2. The first image data ID1 may include a first image I1 and a first label L1 corresponding to the first image.

[0138] In FIG8, the description is based on the case where one image data includes one label, but as described above, one image data may include multiple labels.

[0139] 1.2.2.2. Image Preprocessing

[0140] According to an embodiment of the present invention, image preprocessing may be performed. If the image is input as is for learning, overfitting may occur due to learning unnecessary features, and the learning efficiency will also decrease.

[0141] To prevent this, the data processing module may perform appropriate preprocessing on the image data to meet the learning purpose, thereby improving learning efficiency and performance.

[0142] In one embodiment, the data processing module may utilize or appropriately combine various techniques to perform image preprocessing, such as image resizing, grayscale conversion, histogram equalization, normalization, feature enhancement, image enhancement, denoising, edge detection, segmentation, morphological operations, color space conversion, etc. Image preprocessing using some of these techniques will be described below.

[0143] 1.2.2.2.1. Image Resizing

[0144] According to one embodiment of the present invention, the size of the acquired image data can be adjusted. That is, the image can be resized. According to one embodiment of the present invention, the image resizing can be performed by the data processing module of the learning unit described above.

[0145] The size or aspect ratio of the image can be adjusted. Multiple acquired images can be resized to have a constant size. Alternatively, the image can be resized to have a constant aspect ratio. Resizing the image can be done by applying an image conversion filter to the image.

[0146] If the size or capacity of a single acquired image is too large or too small, the size or capacity of the image can be adjusted to convert it to an appropriate size. Alternatively, if the size or capacity of a single image is different, the size or capacity can be unified by resizing.

[0147] According to one embodiment, the capacity of the image can be adjusted. For example, if the capacity of the image exceeds an appropriate range, the image can be reduced by downsampling. Alternatively, if the capacity of the image does not meet an appropriate range, the image can be enlarged by upsampling or interpolating.

[0148] According to another embodiment, the size or aspect ratio of an image can be adjusted by cropping the image or adding pixels to the acquired image. For example, if the image contains regions that are unnecessary for learning, a portion of the image can be cropped to remove these regions. Alternatively, if cropping a portion of the image results in an aspect ratio mismatch, the image aspect ratio can be adjusted by adding columns or rows. In other words, the aspect ratio can be adjusted by adding margins or padding to the image.

[0149] According to yet another embodiment, the image capacity and size or aspect ratio can be adjusted together. For example, if the image capacity is large, the image capacity can be reduced by downsampling the image and converted to appropriate image data by cropping unnecessary regions contained in the reduced image.

[0150] Furthermore, according to another embodiment of the invention, the orientation of the image data can also be changed.

[0151] Specifically, when using a retinal image dataset as the dataset, the capacity or size of each retinal image can be adjusted. Cropping can be performed to remove blank areas in the retinal image other than the retinal portion, or padding can be performed to supplement the cropped portion of the retinal image to adjust the aspect ratio.

[0152] 1.2.2.2.2. Feature Enhancement

[0153] According to one embodiment of the present invention, the data processing module can perform preprocessing to highlight features of the retinal image. For example, the data processing module can perform preprocessing on the retinal image to more easily detect abnormal signs of ophthalmic diseases, or perform preprocessing to highlight changes in retinal vessels or blood flow.

[0154] For example, image preprocessing can be performed on an image that has undergone the resizing process described above. However, the invention disclosed in this specification is not limited thereto, and image preprocessing can also be performed without resizing. Image preprocessing can be performed by applying a preprocessing filter to the image.

[0155] According to one embodiment, a blur filter can be applied to the image. A Gaussian filter can be applied to the image. A Gaussian blur filter can also be applied to the image. Alternatively, a deblur filter can be applied to the image to make the image clearer.

[0156] According to another embodiment, a filter that adjusts or modulates the color of the image can be applied. For example, a filter that changes the values ​​of some components in the RGB values ​​that constitute the image can be applied, or a filter that binarizes the image can be applied.

[0157] According to yet another embodiment, a filter that highlights specific elements can be applied. For example, preprocessing can be performed on retinal image data to highlight vascular elements in each image. In this case, the preprocessing to highlight vascular elements can be performed by applying one or more filters sequentially or in combination.

[0158] Furthermore, according to an embodiment of the present invention, image preprocessing can be performed according to the characteristics of the diagnostic information to be acquired. For example, if it is desired to acquire diagnostic information related to retinal hemorrhage, drusen, microaneurysms, exudates, etc., preprocessing can be performed to convert the acquired retinal image into a red-light-free retinal image form.

[0159] 1.2.2.2.3. Image Augmentation

[0160] According to an embodiment of the present invention, the image can be augmented or expanded. Image augmentation can be performed by the data processing module of the learning unit described above.

[0161] The augmented image can be used to improve the training performance of the diagnostic model. For example, if the amount of data used to train the diagnostic model is insufficient, the amount of data used for training can be expanded by modulating the existing training image data, and the amount of training image data can be increased by using the modulated (or modified) image and the original image. Therefore, overfitting can be suppressed, deeper model layers can be formed, and prediction accuracy can be improved.

[0162] For example, image data expansion can be performed by horizontally flipping the image, cropping a portion of the image, correcting the color values ​​of the image, or adding artificial noise. Specifically, cropping a portion of the image can be performed by cropping a portion of the elements constituting the image or by randomly cropping a portion of the image. Furthermore, image data can be expanded by horizontally flipping, vertically flipping, proportionally resizing, cropping, filling, color adjustment, or brightness adjustment of the image data.

[0163] In addition, in one embodiment, the data processing module can enhance the image by rotating the retinal image. (See specification page 11 / 43, 14 CN 121586537 A)The retina is circular in shape, and there is no data loss even if the retina is rotated. Therefore, when image enhancement is performed by rotating the retinal image, multiple retinal images can be obtained without data loss. For example, the data processing module can rotate the retinal image based on a predetermined angle (e.g., 15 degrees, 30 degrees, 60 degrees, etc.) to obtain multiple retinal images.

[0164] Furthermore, in one embodiment, the above-mentioned image data enhancement or expansion can generally be applied to the training dataset. However, it can also be applied to other datasets, such as test datasets, i.e., test datasets used to test models that have completed learning using training data and validation using validation data.

[0165] Specifically, when using a set of retinal images as a dataset, one or more of the following processes can be randomly applied: horizontal flipping, cropping, adding noise, or changing colors, to increase the amount of data, thereby obtaining an enhanced retinal image dataset.

[0166] 1.2.2.3. Image Serialization

[0167] According to an embodiment of the present invention, image data can be serialized. The image can be serialized by the data processing module of the learning unit described above. The serialization module can serialize the preprocessed image data and transmit it to the queue module.

[0168] If the image data is used for learning as is, it needs to be decoded because the image data is in the form of image files such as JPG, PNB, and DCM. If the model learning performance is reduced every time it is decoded before learning is performed, the image file can be serialized instead of being used for learning. Therefore, the image data can be serialized to improve learning performance and speed. The serialized image data can be image data that has been applied to one or more of the above-mentioned image resizing and image preprocessing steps, or it can be the original image data that has not undergone any processing.

[0169] Each image data contained in the image dataset can be converted into string form. The image data can be converted into binary data form. In particular, the image data can be converted into a data form suitable for diagnostic model learning. For example, the image data can be converted into TFRecord form for diagnostic model learning using TensorFlow (tensorflow).

[0170] Specifically, when using a retinal image set as the dataset, the acquired retinal image set can be converted into TFRecord format and used for learning the diagnostic model.

[0171] 1.2.2.4. Queue

[0172] A queue can be used to solve the data bottleneck problem. The queue module of the learning unit described above can store image data in the queue and transmit it to the learning model module.

[0173] Especially when using both a central processing unit (CPU) and a graphics processing unit (GPU) to execute the learning process, the bottleneck between the CPU and GPU can be minimized by using queues, making access to the database smoother and improving memory usage efficiency.

[0174] Queues can store data for diagnostic model learning. Queues can store image data. The image data stored in the queue can be image data that has undergone at least one of the above data processing procedures (i.e., resizing, preprocessing, and enhancement), or it can be the acquired raw image.

[0175] Queues can store image data, preferably image data that has been serialized as described above. Queues can store image data and supply image data to the diagnostic model. Queues can transfer image data to the diagnostic model in batch size units.

[0176] Queues can provide image data. Queues can provide data to the learning module described later. As data is extracted from the learning module, the amount of data accumulated in the queue may decrease.

[0177] As the diagnostic model learns, when the amount of data stored in the queue decreases below the baseline, the queue can request additional data. The queue can request additional data of a specific type. When additional data is requested, the learning unit can add data to the queue.

[0178] The queue can be set in the system memory of the learning device. For example, the queue can be formed in the random access memory (RAM) of the central processing unit (CPU). In this case, the size of the queue, i.e., its capacity, can be determined according to the RAM capacity of the CPU. As a queue, a first-in-first-out (FIFO) queue, a primary queue, or a random queue can be used.

[0179] 1.2.3. Learning Process

[0180] According to an embodiment of the present invention, a learning process for the diagnostic model can be disclosed.

[0181] According to an embodiment of the present invention, the learning of the diagnostic model can be performed by the learning device described above. The learning process can be performed by the processor of the learning device described above. The learning process can be performed by the learning module of the learning unit described above.

[0182] FIG9 is a block diagram illustrating a diagnostic model learning process according to an embodiment of the present invention. Referring to FIG9, a diagnostic model learning process according to an embodiment of the present invention can be performed through the following steps: acquiring data S31, learning a diagnostic model S32, validating the learned model S33, and acquiring the variables of the learned model S34.

[0183] 1.2.3.1. Data Input

[0184] A dataset for diagnostic model learning can be acquired.

[0185] The acquired data can be an image dataset processed by the above data processing process. For example, the dataset can beThis includes resized, pre-processed, augmented, and subsequently serialized retinal image data.

[0186] During the learning phase of the diagnostic model, training datasets can be acquired and used. During the validation phase of the diagnostic model, validation datasets can be acquired and used. During the testing phase of the diagnostic model, test datasets can be acquired and used. Each dataset may include retinal images and labels.

[0187] Datasets can be acquired from a queue. Datasets can be acquired from the queue in batch size units. For example, if the batch size is specified as 60, the dataset can be extracted from the queue in units of 60. The size of the batch size may be limited by the RAM capacity of the GPU.

[0188] Datasets can be randomly acquired from the queue to the learning module. Datasets can also be acquired in the order they accumulate in the queue.

[0189] The learning module can specify the composition of the datasets acquired from the queue for extraction. For example, the learning module can extract retinal images with left-eye labels and retinal image data with right-eye labels for a specific subject for use in learning.

[0190] The learning module can acquire datasets with specific labels from the queue. For example, the learning module can obtain a dataset of retinal images labeled as abnormal from the queue. The learning module can obtain the dataset from the queue by specifying the proportion of data corresponding to the label. For example, the learning module can obtain a dataset of retinal images from the queue such that the number of retinal images labeled as abnormal is 1:1 with the number of retinal images labeled as normal.

[0191] 1.2.3.2. Model Design

[0192] The diagnostic model can be designed as a neural network model, a machine learning model, or various other models. In one embodiment, if the diagnostic model includes a neural network model, the neural network model may include multiple layers or levels.

[0193] The neural network model can be implemented as a classifier that generates diagnostic information. The classifier can perform binary or multi-class classification. For example, the neural network model can be a binary classification model that classifies input data into normal or abnormal categories for target diagnostic information such as specific diseases or abnormal signs. Alternatively, the neural network model can be a multi-class classification model that classifies input data into multiple levels of categories for specific characteristics (e.g., the degree of disease progression). Alternatively, the neural network model can also be implemented as a regression model that outputs a specific value related to a specific disease.

[0194] The neural network model may include a convolutional neural network (CNN).As a CNN structure, at least one of AlexNet, LENET, NIN, VGGNet, ResNet, WideResNet, GoogleNet, FractaNet, DenseNet, FitNet, RitResNet, HighwayNet, MobileNet, and DeeplySupervisedNet can be used. The neural network model can be implemented using multiple CNN structures.

[0195] For example, the neural network model can be implemented as including multiple VGGNet blocks. More specifically, the neural network model can be prepared by combining the following structure: a first structure that sequentially combines a CNN layer with 64 filters of size 3x3, a BN (Batch Normalization) layer, and a ReLU layer, and a second block that sequentially combines a CNN layer with 128 filters of size 3x3, a ReLU layer, and a BN layer.

[0196] The neural network model can include a max pooling layer after each CNN block and, at the end, a global average pooling (GAP) layer, a fully connected (FC) layer, and an activation layer (e.g., Sigmoid, Softmax, etc.).

[0197] Furthermore, in another embodiment, if the diagnostic model includes a machine learning model, the machine learning model may include a linear regression model, a Cox proportional hazards model, etc.

[0198] 1.2.3.3. Model Learning

[0199] The diagnostic model may be learned using a training dataset.

[0200] The diagnostic model may be learned using a labeled dataset. However, the learning process of the diagnostic model described in this specification is not limited to this, and the diagnostic model may also be learned in an unsupervised manner using unlabeled data.

[0201] The learning of the diagnostic model may be performed as follows: based on the training image data, the result value is obtained using the diagnostic model assigned arbitrary weight values, the obtained result value is compared with the label value of the training data, and backpropagation is performed according to its error to optimize the weight values. In addition, the learning of the diagnostic model may be affected by the model validation results, test results and / or feedback from the diagnostic phase described later.

[0202] The learning of the above-described diagnostic model may be performed using TensorFlow. However, this invention is not limited to this; frameworks such as Theano, Keras, Caffe, Torch, and CNTK (Microsoft Cognitive Toolkit) can also be used for learning diagnostic models.

[0203] 1.2.3.4. Model Validation

[0204] The diagnostic model can be validated using a validation dataset. The validation of the diagnostic model can be performed in the following ways.Line: Obtain the result values ​​of the validation dataset from the learned diagnostic model and compare the result values ​​with the labels of the validation dataset. Validation can be performed by measuring the accuracy of the result values. Based on the validation results, the parameters (e.g., weights and / or biases) or hyperparameters (e.g., learning rate) of the diagnostic model can be adjusted.

[0205] For example, a learning device according to an embodiment of the present invention can train a diagnostic model based on retinal image prediction diagnostic information and perform the validation of the diagnostic model by comparing the diagnostic information of the validated retinal images of the learned model with the validation labels corresponding to the validated retinal images.

[0206] For the validation of the diagnostic model, a separate validation set can be used, i.e., a dataset with discriminative factors not included in the training dataset. For example, the separate validation set can be a dataset that differs from the training dataset in terms of race, environment, age, gender, etc.

[0207] 1.2.3.5. Model Testing

[0208] The diagnostic model can be tested using a test dataset. Specification 14 / 43 pages 17 CN 121586537 A

[0209] According to an embodiment of the present invention, the learning process can be used to test a diagnostic model using a test dataset that is separate from the training dataset and the validation dataset. Based on the test results, the parameters (e.g., weights and / or biases) or hyperparameters (e.g., learning rate) of the diagnostic model can be adjusted.

[0210] For example, a learning device according to an embodiment of the present invention can obtain result values ​​from a diagnostic model that has been learned to predict diagnostic information based on retinal images, with test retinal image data not used for training and validation as input, to perform a test on the learned and validated diagnostic model.

[0211] For testing the diagnostic model, a separately prepared external data set can be used, i.e., a dataset with factors that are separate from the training and / or validation data.

[0212] 1.2.3.6. Output of Results

[0213] As a result of learning the diagnostic model, optimized parameter values ​​of the model can be obtained. As described above, as the model is repeatedly learned using the test dataset, more suitable parameter (or variable) values ​​can be obtained. Once learning has been fully completed, optimized values ​​for the weights and / or biases can be obtained.

[0214] According to one embodiment of the present invention, the learned diagnostic model and / or the parameters or variables of the learned diagnostic model can be stored in the learning device and / or the diagnostic device (or server). The learned diagnostic model can be used by the diagnostic device and / or client device, etc., for predicting diagnostic information. Furthermore, the parameters or variables of the learned diagnostic model can also be updated through feedback obtained from the diagnostic device or client device.

[0215] 1.2.3.7. Model Ensemble

[0216] According to an embodiment of the present invention, multiple sub-models can be learned simultaneously during the learning of a diagnostic model. The multiple sub-models can have different hierarchical structures.

[0217] At this time, the diagnostic model according to an embodiment of the present invention can be implemented by combining multiple sub-diagnostic models. In other words, the learning of the diagnostic model can be performed using the ensemble technique of combining multiple sub-diagnoses.

[0218] When a diagnostic model is constructed by forming an ensemble, prediction can be performed by synthesizing the results predicted from various forms of sub-diagnostic models, thereby further improving the accuracy of the result prediction.

[0219] 1.3. Diagnostic Process

[0220] According to an embodiment of the present invention, a process (or diagnostic process) for obtaining diagnostic information using a diagnostic model can be provided. Specifically, through the diagnostic process, diagnostic information (e.g., diagnostic information or seen information) can be predicted using retinal images and the learned diagnostic model.

[0221] The diagnostic process described below can be performed by a diagnostic device.

[0222] 1.3.1. Diagnostic Unit

[0223] According to an embodiment of the present invention, the diagnostic process can be executed by the processor of the above-described diagnostic device. The processor can be provided in the above-described diagnostic device.

[0224] FIG10 is a diagram illustrating the configuration of a diagnostic unit according to an embodiment of the present invention. Referring to FIG10, the diagnostic unit 200 may include a diagnostic request acquisition module 210, a data processing module 230, a diagnostic module 250, and an output module 270.

[0225] Each module can execute various steps of the data processing flow and the learning flow described later. However, the constituent elements and the functions performed by each element illustrated in FIG10 are not all necessary; depending on the aspect of the diagnosis, some elements may be added or omitted.

[0226] 1.3.2. Data Acquisition and Diagnostic Request

[0227] According to an embodiment of the present invention, the diagnostic device can acquire diagnostic target data and acquire diagnostic information based thereon. The diagnostic target data may be image data. Data acquisition and diagnostic request acquisition can be performed by the diagnostic request acquisition module described in the diagnostic instructions (page 15 / 43, CN 121586537 A).

[0228] For example, diagnostic target data (TD) may include diagnostic target image (TI) and diagnostic target patient information (PI; patient information).

[0229] The diagnostic target image (TI) may be an image used to acquire diagnostic information about the diagnostic target patient. For example, the diagnostic target image may be a retinal image and / or a retinal image. The diagnostic target (TI) may have any of the following formats: JPG, PNG, DCM (DICOM), BMP, GIF, TIFF.

[0230] The diagnostic object information (PI) can be information used to identify the diagnostic target object. Alternatively, the diagnostic object information (PI) can be characteristic information of the diagnostic target object or image. For example, the diagnostic object information (PI) can include the shooting time of the diagnostic target image, the shooting device, the identification number, ID, name, gender, age, weight, ethnicity, smoking status, blood pressure (hypertension), diabetes status, etc. If the diagnostic target image is a retinal image, the diagnostic object information (PI) can further include information related to the eyeball, such as binocular information indicating whether it is a left eye image or a right eye image.

[0231] The diagnostic device can acquire a diagnostic request. The diagnostic device can acquire diagnostic target data together with the diagnostic request. When acquiring a diagnostic request, the diagnostic device can acquire diagnostic information using a learned diagnostic model. The diagnostic device can acquire a diagnostic request from a client device. Alternatively, the diagnostic device can acquire a diagnostic request from a user through a separately set input means.

[0232] 1.3.3. Data processing flow

[0233] The acquired data can be processed. The data processing can be performed by the data processing module of the diagnostic unit described above.

[0234] The data processing flow can generally be performed similarly to the data processing flow in the learning flow described above. The data processing flow in the diagnostic flow will now be described, focusing on the differences from the data processing flow in the learning flow.

[0235] In the diagnostic flow, the diagnostic device can acquire data as in the learning flow. At this time, the acquired data can have the same format as the data acquired in the learning flow. For example, if the learning device trained a diagnostic model using image data in DCM format in the learning flow, the diagnostic device can acquire DCM images and use the learned diagnostic model to obtain diagnostic information.

[0236] In the diagnostic flow, the acquired diagnostic target image can be resized as in the image data used in the learning flow. The diagnostic target image can be resized to have appropriate capacity, size, and / or aspect ratio in order to effectively perform diagnostic information prediction using the learned diagnostic model.

[0237] For example, if the diagnostic target image is a retinal image, resizing can be performed, such as cropping unnecessary parts of the image or reducing its size, in order to predict diagnostic information based on the retinal image.

[0238] In the diagnostic process, preprocessing filters can be applied to the acquired diagnostic target image, similar to the image data used in the learning process. Appropriate filters can be applied to the diagnostic target image to further improve the accuracy of diagnostic information predictions made through the learned diagnostic model.

[0239] For example, if the diagnostic target image is a retinal image, preprocessing that helps predict diagnostic information can be applied to the diagnostic target image, such as image preprocessing that emphasizes blood vessels or image preprocessing that emphasizes or weakens specific colors.

[0240] In the diagnostic process, the acquired diagnostic target images can be serialized like the image data used in the learning process. The diagnostic target images can be converted or serialized into a form that facilitates driving the diagnostic model in a specific working framework.

[0241] The serialization of the diagnostic target images can be omitted. This is because, unlike the learning phase, the amount of data processed by the processor at one time in the diagnostic phase is small, so the burden on data processing speed is relatively small. Specification 16 / 43 pages 19 CN 121586537 A

[0242] In the diagnostic process, the acquired diagnostic target images can be stored in a queue like the image data used in the learning process. However, since the amount of data processed in the diagnostic process is smaller than that in the learning process, the step of storing the data in the queue can be omitted.

[0243] On the other hand, in the diagnostic process, since it is not necessary to increase the amount of data, unlike the learning process, data augmentation or image augmentation processes can be omitted in order to obtain accurate diagnostic information.

[0244] 1.3.4. Diagnostic Process

[0245] According to an embodiment of the present invention, a diagnostic process utilizing a learned diagnostic model can be disclosed. The diagnostic process can be executed in the diagnostic device described above. The diagnostic process can be executed in the diagnostic server described above. The diagnostic process can be executed in the control unit of the diagnostic device described above. The diagnostic process can be executed by the diagnostic module of the diagnostic unit described above.

[0246] FIG11 is a diagram illustrating a diagnostic process according to an embodiment of the present invention. Referring to FIG11, the diagnostic process can be executed by the following steps: acquiring diagnostic target data S31, utilizing a learned diagnostic model S42, and acquiring a result corresponding to the acquired diagnostic target data S43. However, data processing can be selectively performed.

[0247] The various steps of the diagnostic process will be described below with reference to FIG11.

[0248] 1.3.4.1. Data Input

[0249] According to an embodiment of the present invention, the diagnostic module can acquire diagnostic target data. The acquired data can be data that has been processed as described above. For example, the acquired data can be retinal image data of a subject that has been resized and preprocessed with prominent blood vessels applied. According to one embodiment of the present invention, the left eye image and the right eye image of a subject can be input together as diagnostic target data.

[0250] 1.3.4.2. Data Classification

[0251] A diagnostic model configured as a classifier can classify the input diagnostic target image into a positive or negative category according to a predetermined label.

[0252] A learned diagnostic model can receive diagnostic target data as input and output predicted labels. A learned diagnostic model can output predicted values ​​of diagnostic information. Diagnostic information can be obtained using the learned diagnostic model. Diagnostic information can be determined based on the predicted labels.

[0253] For example, a diagnostic model can predict diagnostic information (i.e., information about the presence or absence of disease) or observed information (i.e., information about the presence or absence of abnormalities) of an examinee's ophthalmological or systemic diseases. In this case, the diagnostic information or observed information can be output in probabilistic form. For example, the probability that the examinee has a specific disease or the probability that a specific abnormality is observed in the examinee's retinal image can be output. When using a diagnostic model set up as a classifier, the predicted label can be determined by considering whether the output probability value (or prediction score) exceeds a threshold.

[0254] Specifically, the diagnostic model can use the examinee's retinal image as the diagnostic target image and output the probability value of whether the examinee has diabetic retinopathy. When using a diagnostic model in the form of a classifier where normal is 1, the examinee's retinal photograph can be input into the diagnostic model, and the probability value of whether the examinee has diabetic retinopathy can be obtained in the form of normal:abnormal = 0.74:0.26, etc.

[0255] This description is based on the case where a diagnostic model in the form of a classifier is used to classify data, but the present invention is not limited thereto. A diagnostic model implemented in the form of a regression model can also be used to predict specific diagnostic values ​​(e.g., blood pressure).

[0256] According to another embodiment of the present invention, applicability information of an image can be obtained. Applicability information can indicate whether a diagnostic target image is suitable for obtaining diagnostic information using a diagnostic model.

[0257] The applicability information of an image can be image quality information. Quality information or applicability information can indicate whether the target image meets a baseline level.

[0258] For example, if a diagnostic target image is defective due to defects in the imaging equipment or lighting effects during imaging, an inapplicability result can be output as applicability information for that diagnostic target image. If the noise contained in the diagnostic target image exceeds a certain level, the diagnostic target image can be determined to be inapplicable.

[0259] Applicability information can be a value predicted by using a diagnostic model. Alternatively, applicability information can be information obtained through a separate image analysis process.

[0260] According to one embodiment, even if an image is classified as inapplicable, diagnostic information obtained based on the inapplicable image can still be acquired.

[0261] According to one embodiment, images classified as inapplicable can be re-examined by a diagnostic model.

[0262] In this case, the diagnostic model performing the re-examination can be different from the diagnostic model performing the initial examination. For example, the diagnostic device can store a first diagnostic model and a second diagnostic model, and images classified as inapplicable by the first diagnostic model can be examined by the second diagnostic model.

[0263] According to yet another embodiment of the invention, a map can be acquired from a learned diagnostic model. Diagnostic informationInformation may include maps. Maps may be acquired together with other diagnostic information. For example, maps may include saliency maps, class activation maps (CAMs), heat maps, etc. In addition, CAMs may be selectively acquired. For example, when the diagnostic information or seen information acquired by the diagnostic model is classified as an abnormal category, CAMs may be extracted and / or output.

[0264] 1.3.5. Output of Diagnostic Information

[0265] Diagnostic information may be determined based on the results output by the diagnostic model. The output of diagnostic information may be performed by the output module of the diagnostic unit described above. Diagnostic information may be output from the diagnostic device to the client device. Diagnostic information may be output from the diagnostic device to the server device. Diagnostic information may be stored in the diagnostic device or the diagnostic server. Diagnostic information may be stored in a separately configured server device, etc.

[0266] Diagnostic information may be database-based and managed. For example, the acquired diagnostic information may be stored and managed together with the diagnostic target image of the corresponding subject according to the identification number of the subject. At this time, the diagnostic target image and diagnostic information of the subject may be managed in chronological order. By managing diagnostic information and diagnostic target images in a time-series manner, the history of individual diagnostic information can be easily tracked and managed.

[0267] Diagnostic information can be provided to the user. Diagnostic information can be provided to the user through the output means of a diagnostic device or a client device. Diagnostic information can be output through visual or auditory output means provided in the diagnostic device or client device for the user to perceive.

[0268] According to an embodiment of the present invention, an interface for effectively providing diagnostic information to the user can be provided.

[0269] Furthermore, if an image is classified as inapplicable, applicability information of the image can be provided together. For example, if an image is classified as inapplicable, diagnostic information and inapplicability determination information corresponding to the image can be provided together.

[0270] Diagnostic target images that are determined to be inapplicable can also be classified as target images to be re-shot. In this case, guidance for re-shooting objects of images classified as re-shot targets can be provided together with applicability information. On the other hand, in response to providing diagnostic information obtained through the diagnostic model, feedback related to the learning of the diagnostic model can be obtained. For example, feedback for adjusting parameters or hyperparameters related to the learning of the diagnostic model can be obtained. Feedback can be obtained through an input module located in the diagnostic device or client device.

[0271] According to one embodiment of the present invention, the diagnostic information corresponding to the diagnostic target image may include grade information. The grade information can be selected from multiple grades. The grade information may be based on diagnostic information obtained through the diagnostic model and / or the specification page 18 / 43 21 CN 121586537 AThe grading information is determined by the observed information. The grading information may be determined by considering the applicability or quality information of the diagnostic target image. If the diagnostic model is a classifier model that performs multi-class classification, the grading information may be determined by considering the category into which the diagnostic target image is classified by the diagnostic model. If the diagnostic model is a regression model that outputs a numerical value related to a specific disease, the grading information may be determined by considering the output numerical value. In addition, the grading information may be determined by applying a predetermined cutoff value to the score corresponding to the diagnostic information.

[0272] For example, the diagnostic information corresponding to the diagnostic target image may include any grading information selected from first grading information or second grading information. If abnormal observed information or abnormal diagnostic information is obtained through the diagnostic model, the grading information may be selected as first grading information. If abnormal observed information or abnormal diagnostic information is not obtained through the diagnostic model, the grading information may be selected as second grading information. Alternatively, if the numerical value obtained through the diagnostic model exceeds the baseline numerical value, the grading information may be selected as first grading information, and if the obtained numerical value does not meet the baseline numerical value, the grading information may also be selected as second grading information. The first grading information may indicate that there is stronger abnormal information in the diagnostic target image compared to the second grading information.

[0273] On the other hand, if the quality of the diagnostic target image is determined to be lower than the baseline by image analysis or a diagnostic model, the level information can be selected as the third level information. Alternatively, the diagnostic information may include the third level information and the first or second level information.

[0274] 1.4. Diagnostic System Using Multiple Diagnostic Models

[0275] According to an embodiment of the present invention, diagnostic information can be output using a diagnostic model as described above. The diagnostic model described above may be composed of one diagnostic model or multiple diagnostic models. Furthermore, if the diagnostic model is composed of multiple diagnostic models, the multiple diagnostic models may be configured in parallel or in sequence. Parallel diagnostic models and serial diagnostic models will be described below.

[0276] 1.4.1.1. Parallel Diagnostic System Configuration

[0277] According to an embodiment of the present invention, a parallel diagnostic system for acquiring multiple diagnostic information can be provided. The parallel diagnostic system can train multiple diagnostic models for acquiring multiple diagnostic information and acquire multiple diagnostic information using the learned multiple diagnostic models.

[0278] For example, a parallel diagnostic system can train a first diagnostic model related to the presence or absence of ophthalmic diseases in a subject and a second diagnostic model related to the presence or absence of systemic diseases in a subject, based on retinal images, and output diagnostic information about the presence or absence of ophthalmic diseases and systemic diseases in the subject using the learned first and second diagnostic models.

[0279] Multiple diagnostic models can be learned in parallel and / or independently. By using multiple diagnostic models in this way...Learning a model by testing different labels can improve the prediction accuracy for each label and improve the efficiency of the prediction operation. Since the above-described matters can be applied to the learning of multiple diagnostic models, detailed descriptions are omitted.

[0280] 1.4.1.2. Parallel Diagnostic Process

[0281] According to an embodiment of the present invention, a diagnostic process for acquiring multiple diagnostic information can be provided. The diagnostic process for acquiring multiple diagnostic information can be implemented in the form of a parallel diagnostic process including multiple mutually independent sub-diagnostic processes.

[0282] According to an embodiment of the present invention, the diagnostic process can be executed by multiple diagnostic modules. Each diagnostic process can be executed independently.

[0283] FIG12 is a block diagram for illustrating a diagnostic unit according to an embodiment of the present invention.

[0284] Referring to FIG12, a diagnostic unit 200 according to an embodiment of the present invention may include a diagnostic request acquisition module 211, a data processing module 231, a first diagnostic module 251, a second diagnostic module 253, and an output module 271. Unless otherwise specified, each module of the diagnostic unit 200 (pages 19 / 43, CN 121586537 A) can operate similarly to the diagnostic module of the diagnostic unit shown in FIG10.

[0285] In FIG12, even though the diagnostic unit 200 includes multiple diagnostic modules, the diagnostic request acquisition module 211, data processing module 231, and output module 271 are shown as common, but the present invention is not limited to this configuration, and the diagnostic request acquisition module, data processing module, and / or output module may also be configured as multiple. Multiple diagnostic request acquisition modules, data processing modules, and / or output modules may also operate in parallel.

[0286] For example, the diagnostic unit 200 may include a first data processing module that performs a first processing on an input diagnostic target image and a second processing module that performs a second data processing on the diagnostic target image. The first diagnostic module may acquire first diagnostic information based on the first processed diagnostic target image, and the second diagnostic module may acquire second diagnostic information based on the second processed diagnostic target image. The first and / or second processing can be any one of image resizing, image color modulation, applying a blur filter, vascular emphasis processing, red light removal conversion, partial region cropping, or partial element extraction.

[0287] Multiple diagnostic modules can acquire different diagnostic information. Multiple diagnostic modules can use different diagnostic models to acquire diagnostic information. For example, the first diagnostic module can use a first diagnostic model to predict whether the subject has an ophthalmological disease to acquire first diagnostic information related to the presence or absence of the subject's ophthalmological disease, and the second diagnostic module can use a second diagnostic model to predict whether the subject has a systemic disease to acquire second diagnostic information related to the presence or absence of the subject's systemic disease.

[0288] More specifically, the first diagnostic module can use a retinal image-based model to predict whether the subject has diabetic retinopathy.A first diagnostic model for retinal lesions obtains first diagnostic information about whether the subject has diabetic retinopathy. A second diagnostic module can use a second diagnostic model based on retinal images to predict whether the subject has hypertension to obtain second diagnostic information related to whether the subject has hypertension.

[0289] Furthermore, the diagnostic process according to an embodiment of the present invention may include multiple sub-diagnostic processes. Each sub-diagnostic process may be executed using a different diagnostic model. Each sub-diagnostic process may be executed in a different diagnostic process. For example, the first diagnostic module may execute a first sub-diagnostic process for obtaining first diagnostic information using a first diagnostic model. Alternatively, the second diagnostic module may execute a second sub-diagnostic process for obtaining second diagnostic information using a second diagnostic model.

[0290] Multiple learned diagnostic models may receive diagnostic target data as input and output predicted labels or probabilities. Each diagnostic model may be set up as a classifier and classify the input diagnostic target data against predetermined labels. At this time, multiple diagnostic models may be set up as classifiers that learn for different characteristics.

[0291] On the other hand, a map may be obtained from each diagnostic model. The map may be selectively obtained. The map may be extracted when predetermined conditions are met. For example, if the first diagnostic information indicates that the subject is abnormal for a first characteristic, a first map can be obtained from the first diagnostic model.

[0292] FIG13 is a diagram illustrating a diagnostic process according to an embodiment of the present invention.

[0293] Referring to FIG13, a diagnostic process according to an embodiment of the present invention may include: obtaining diagnostic target data S51, and obtaining diagnostic information corresponding to the diagnostic target data using a first diagnostic model and a second diagnostic model S51a, S51b). The diagnostic target data may be processed data.

[0294] A diagnostic process according to an embodiment of the present invention may include: obtaining first diagnostic information through a learned first diagnostic model, and obtaining second diagnostic information through a learned second diagnostic model. The first diagnostic model and the second diagnostic model may obtain the first diagnostic information and the second diagnostic information respectively based on the same diagnostic target data.

[0295] For example, the first diagnostic model and the second diagnostic model may obtain first diagnostic information about the subject's ophthalmological disease and second diagnostic information about the presence or absence of the subject's kidney disease based on the diagnostic target retinal image. Instruction manual, pages 20 / 43, CN 121586537 A

[0296] Furthermore, unless otherwise specified, the diagnostic process related to FIG13 can be implemented similarly to the diagnostic process described above with reference to FIG11.

[0297] 1.4.1.3. Output of diagnostic information

[0298] According to an embodiment of the present invention, diagnostic information obtained through a parallel diagnostic process can be acquired. The acquired...Diagnostic information can be stored in a diagnostic device, a server device, and / or a client device. The acquired diagnostic information can be transmitted to an external device.

[0299] Multiple diagnostic information can respectively indicate multiple labels predicted by multiple diagnostic models. Multiple diagnostic information can respectively correspond to multiple labels predicted by multiple diagnostic models. Alternatively, diagnostic information can be information determined based on multiple labels predicted by multiple diagnostic models. Diagnostic information can correspond to multiple labels predicted by multiple diagnostic models.

[0300] In other words, the first diagnostic information can be diagnostic information corresponding to a first label predicted by a first diagnostic model. Alternatively, the first diagnostic information can be diagnostic information determined by combining the first label predicted by the first diagnostic model and the second label predicted by a second diagnostic model.

[0301] On the other hand, an image of a map obtained from multiple diagnostic models can be output. The map image can be output when predetermined conditions are met. For example, if the first diagnostic information indicates that the subject is abnormal for a first characteristic or the second diagnostic information indicates that the subject is abnormal for a second characteristic, a map image obtained from a diagnostic model that outputs diagnostic information indicating an abnormality can be output.

[0302] Multiple diagnostic information and / or map images can be provided to the user. Multiple diagnostic information, etc., can be provided to the user through the output means of a diagnostic device or a client device.

[0303] According to an embodiment of the present invention, the diagnostic information corresponding to the diagnostic target image may include grade information. The grade information can be selected from multiple grades. The grade information can be determined based on multiple diagnostic information and / or seen information obtained through a diagnostic model. The grade information can be determined considering the applicability information or quality information of the diagnostic target image. The grade information can be determined considering the category in which the diagnostic target image is classified by multiple diagnostic models. The grade information can be determined considering the numerical values ​​output from multiple diagnostic models.

[0304] For example, the diagnostic information corresponding to the diagnostic target image may include any grade information selected from first grade information or second grade information. If at least one abnormal seen information or abnormal diagnostic information is obtained from the diagnostic information obtained through multiple diagnostic models, the grade information can be selected as first grade information. If no abnormal seen information or abnormal diagnostic information is obtained from the diagnostic information obtained through the diagnostic model, the grade information can be selected as second grade information.

[0305] If at least one of the values ​​obtained through the diagnostic model exceeds a baseline value, the level information can be selected as first-level information; and if none of the obtained values ​​meet the baseline value, the level information can also be selected as second-level information. First-level information can indicate a stronger abnormality in the diagnostic target image compared to second-level information.

[0306] If the quality of the diagnostic target image is determined to be lower than the baseline by image analysis or diagnostic model, the level information can be selected as the third level information. Alternatively, the diagnostic information may include the third level information and the first or second level information.

[0307] Furthermore, a diagnostic system according to an embodiment of the present invention may include a retinal image acquisition unit, a first processing unit, a second processing unit, a third processing unit, and a diagnostic information output unit.

[0308] According to an embodiment of the present invention, a diagnostic system may include a diagnostic device. The diagnostic device may include a retinal image acquisition unit, a first processing unit, a second processing unit, a third processing unit, and / or a diagnostic information output unit. However, this is not limited to this, and the various parts included in the diagnostic system may be located at appropriate positions on the learning device, the diagnostic device, the learning diagnostic server, and / or the client device, respectively. For convenience, the following description will be based on the case where the diagnostic device of the diagnostic system includes a retinal image acquisition unit, a first processing unit, a second processing unit, a third processing unit, and a diagnostic information output unit.

[0309] FIG14 is a diagram for illustrating a diagnostic system according to an embodiment of the present invention. Referring to FIG14, the diagnostic system includes a diagnostic device, which may include a retinal image acquisition unit, a first processing unit, a second processing unit, a third processing unit, and a diagnostic information output unit.

[0310] According to an embodiment of the present invention, a diagnostic system for assisting in the diagnosis of multiple diseases, based on a retinal image, may include: a retinal image acquisition unit for acquiring a target retinal image as the basis for acquiring diagnostic information about a subject; a first processing unit for acquiring a first result about the subject using a first diagnostic model (the first diagnostic model is based on a first set of retinal images for machine learning) on ​​the target retinal image; a second processing unit for acquiring a second result about the subject using a second diagnostic model (the second diagnostic model is based on a second set of retinal images that is at least partially different from the first set of retinal images) on the target retinal image; a third processing unit for determining diagnostic information about the subject based on the first result and the second result; and a diagnostic information output unit for providing the determined diagnostic information to a user.

[0311] The third processing unit may combine the first result and the second result to determine whether the diagnostic information corresponding to the target retinal image is normal information or abnormal information.

[0312] The third processing unit may determine the diagnostic information about the subject by giving priority to abnormal results, thereby improving diagnostic accuracy.

[0313] When both the first and second results are normal, the third processing unit can determine the diagnostic information as normal; when either the first or second result is abnormal, the diagnostic information can be determined as abnormal.

[0314] The first and second results can be for the same disease or for different diseases.

[0315] At least one map related to the first / second result can be obtained through the first processing unit and / or the first / second diagnostic model, and the diagnostic information output unit can output an image of at least one map.

[0316] When the diagnostic information obtained by the third processing unit is abnormal diagnostic information, the diagnostic information output unit can output an image of at least one map.

[0317] The diagnostic system may further include a fourth processing unit that obtains quality information of the target retinal image, and the diagnostic information output unit can output the quality information of the target retinal image obtained by the fourth processing unit.

[0318] If the fourth processing unit determines that the quality information of the target retinal image is lower than a predetermined quality level, the diagnostic information output unit can provide the user with the determined diagnostic information and information indicating that the quality information of the target retinal image is lower than the predetermined quality level.

[0319] 1.4.2.1. Configuration of a serial diagnostic system

[0320] According to one embodiment, a serial diagnostic system including multiple diagnostic models connected in a serial manner can be provided. The serial type diagnostic model will be described below by example.

[0321] FIG15 is a diagram for illustrating a serial diagnostic model according to one embodiment. Referring to Figure 15, the diagnostic model 1000 may include a first serial model 1100 and a second serial model 1200.

[0322] The first diagnostic model 1100 may acquire input data including retinal images and acquire a first output (or intermediate output).

[0323] The first diagnostic model 1100 may acquire input data including retinal images and / or other medical diagnostic images and / or non-visual diagnostic data. The input data may include retinal images, OCT images, iris images, ophthalmic angiography images, lung CT images, lung CT images, cardiac CT images, lung X-ray images, cardiac X-ray images, kidney X-ray images, other computed tomography images, MRI images, or X-ray images. The input data may include data representing the subject's age, height, gender, smoking status, family history, etc.

[0324] The first output may be the value acquired by the output layer of the first diagnostic model 1100. For example, the first diagnostic model 1100 is a classifier model, and the first output may include the output values ​​at multiple nodes of the output layer of the first diagnostic model 1100. As another example, the first diagnostic model 1100 is a regression model, and the first output may include the numerical values ​​obtained by the first diagnostic model 1100.

[0325] The first output may be a numerical value provided by a portion of the layers of the first diagnostic model 1100. For example, the first output may be a value obtained based on the values ​​of the output layer of the first diagnostic model 1100. Alternatively, the first output may also be a value obtained based on the values ​​of the hidden layers of the first diagnostic model 1100.

[0326] According to one embodiment, the first output may be a value obtained through an activation function in the output layer of the first diagnostic model 1100.The value obtained. If the output layer of the first diagnostic model 1100 includes multiple nodes (or neurons), the first output may include the respective output values ​​corresponding to the multiple nodes or the values ​​obtained by a predetermined function (e.g., summation) based on the respective output values.

[0327] The activation function may be any one of the following: Sigmoid function, hyperbolic tangent function, Rectified Linear Unit (ReLU) function, PReLU, Leaky ReLU function, Identity Function, Exponential Linear Unit (ELU) function, and Maxout function.

[0328] The first output may be a feature map or feature value related to the target disease. The first output may be a probability map, saliency map, heatmap, etc. related to the target disease. The second diagnostic model 1200 may be configured to obtain diagnostic information based on the feature map or feature value related to the target disease.

[0329] The first output may be a probabilistic phenotype related to the target disease. For example, if the target disease is coronary artery disease, and the diagnostic information is numerical information related to the target coronary artery disease, then the first output may be the probability that the subject has the target coronary artery disease based on the eye image. The second diagnostic model 1200 may be configured to obtain diagnostic information for the target disease based on a probability expression related to the target disease.

[0330] The second diagnostic model 1200 may obtain a second output (or diagnostic information) based on the first output. The second diagnostic model 1200 may be a diagnostic model 1000 that has been learned to obtain a second output with the first output as input. The second output may be diagnostic information in various forms as described in this specification.

[0331] FIG16 is a diagram illustrating a serial diagnostic model according to another embodiment. Referring to FIG16, the diagnostic model 1000 may include a first diagnostic model 1100 and a second diagnostic model 1200.

[0332] The first diagnostic model 1100 may acquire input data including an eye image and acquire a first output (intermediate output). The second diagnostic model 1200 may acquire diagnostic information based on the first output and the second input.

[0333] The second input may be the same input data as the first input. For example, the first diagnostic model 1100 can obtain a first output based on an eye image, and the second diagnostic model 1200 can obtain diagnostic information based on the first output and the eye image.

[0334] The second input can be input data obtained based on the first input. For example, the second input can be eye image data obtained after performing image processing. For example, the second input can be a black and white processed eye image, an eye image with blood vessels emphasized, a blood vessel image extracted from an eye image, or an eye image with blood vessels removed.

[0335] The second input can be input data that is at least partially different from the first input.

[0336] The second input may be image data different from the first input. The second input may include retinal images, OCT images, iris images, ophthalmic angiography images, lung CT images, lung CT images, cardiac CT images, lung X-ray images, cardiac X-ray images, kidney X-ray images, other tomographic images, MRI images, or X-ray images.

[0337] The second input may include non-visual information about the subject. For example, the first diagnostic model 1100 may obtain a first output about the target disease based on eye images on page 23 / 43 of the specification, CN 121586537 A, and the second diagnostic model 1200 may obtain diagnostic information based on the first output and the second input (e.g., the subject's physical information (e.g., age, sex, race, smoking status, blood pressure (e.g., blood pressure value, hypertension status), diabetes status (or blood glucose value), cholesterol value, family history, etc.)).

[0338] FIG17 is a diagram illustrating a serial diagnostic model according to yet another embodiment. Referring to FIG17, the diagnostic model 1000 may include a first diagnostic model 1100 and a second diagnostic model 1200.

[0339] Compared with FIG16, the diagnostic model 1000 may further acquire first diagnostic information acquired by the first diagnostic model 1100. The diagnostic model 1000 may acquire intermediate diagnostic information and secondary diagnostic information acquired based on the intermediate diagnostic information.

[0340] The diagnostic model 1000 may acquire first diagnostic information and second diagnostic information. The first diagnostic model 1100 may acquire input data including an eye image and acquire a first output (first diagnostic information or intermediate output). The diagnostic model 1000 may acquire first diagnostic information based on the first output. The second diagnostic model 1200 may acquire second diagnostic information at least partially based on the first output. The diagnostic model 1000 may acquire second diagnostic information at least partially based on the first output and in combination with other information extracted from the eye image.

[0341] The first diagnostic information and the second diagnostic information may be diagnostic information for the same target disease. The first diagnostic information can be diagnostic information obtainable from eye images (clinically or through machine learning models), such as the probability expression of the presence or absence of eye diseases, vascular abnormalities, kidney diseases, etc.

[0342] The second diagnostic information can represent more detailed diagnostic information than the first diagnostic information. For example, the second diagnostic information may include rating information representing the risk level of the target disease or score information representing a score related to the target disease for the same target disease as the first diagnostic information.

[0343] The second diagnostic information may be associated with the first diagnostic information, but may include diagnostic information that further considers information obtained beyond the image, such as the rate of disease progression, guidance information related to the diagnostic information, etc.

[0344] On the other hand, the first diagnostic information and the second diagnostic information can be diagnostic information for different diseases. The first diagnostic information and the second diagnostic information can be diagnostic information for different diseases belonging to the same group. For example, the first diagnostic information can be diagnostic information related to glaucoma belonging to the ophthalmology disease group, and the second diagnostic information can be diagnostic information related to macular degeneration belonging to the ophthalmology disease group. For example, the first diagnostic information can be diagnostic information related to drusen belonging to the ophthalmology disease group, and the second diagnostic information can be diagnostic information related to diabetic retinopathy belonging to the ophthalmology disease group.

[0345] The first diagnostic information and the second diagnostic information can be diagnostic information for different diseases belonging to different groups. For example, the first diagnostic information can be diagnostic information related to macular degeneration or drusen belonging to the ophthalmology disease group, and the second diagnostic information can be diagnostic information related to hyperlipidemia belonging to the cardiovascular disease group.

[0346] In the above embodiment, the case where the diagnostic model 1000 has a first diagnostic model 1100 and a second diagnostic model 1200 has been described, but the diagnostic model 1000 may also include more diagnostic models. In addition, each diagnostic model can be connected by the parallel connection or serial connection described above.

[0347] 1.4.2.2. Diagnosis via a serial diagnostic model

[0348] According to an embodiment of the invention described in this specification, a diagnostic method utilizing a diagnostic model including serially connected sub-models can be provided.

[0349] FIG18 is a diagram illustrating a diagnostic method utilizing a diagnostic model according to an embodiment.

[0350] Referring to FIG18, a diagnostic method according to an embodiment may include: acquiring input data S61, acquiring first diagnostic information S62, and acquiring second diagnostic information S63.

[0351] The step of acquiring input data S61 may include acquiring a retinal image of the subject. The step of acquiring input data S61 may further include acquiring medical images of body parts other than the subject's eyeballs. The step of acquiring input data S61 may further include acquiring non-visual information about the subject (e.g., body information of the subject). The step of acquiring input data S61 may further include performing preprocessing on the eyeball images of the subject required to acquire diagnostic information.

[0352] The step of obtaining the first diagnostic information S62 may include obtaining first diagnostic information about the subject based on the eye image and through a first diagnostic model. Obtaining the first diagnostic information may be obtaining diagnostic information related to a first disease. For example, obtaining the first diagnostic information may include obtaining first diagnostic information based on the eye image that represents the probability that the subject's coronary artery calcium score is greater than or equal to 0.

[0353] The step of obtaining second diagnostic information S63 may include obtaining second diagnostic information about the subject based on the first diagnostic information and through the second diagnostic model.

[0354] Obtaining second diagnostic information may include obtaining diagnostic information related to the first disease and different from the first diagnostic information. For example, obtaining first diagnostic information may include obtaining first diagnostic information indicating the probability that the subject's coronary artery calcium score is greater than or equal to 0, and obtaining second diagnostic information may include obtaining second diagnostic information based on the subject's first diagnostic information, indicating a score related to the subject having a target cardiovascular disease (e.g., the probability of cardiovascular disease-related events occurring within 10 years).

[0355] Alternatively, obtaining second diagnostic information may include obtaining diagnostic information related to a second disease that is different from the first disease. For example, obtaining first diagnostic information may include obtaining first diagnostic information indicating that the subject has a target ophthalmological disease, and obtaining second diagnostic information may include obtaining second diagnostic information indicating that the subject has a cardiovascular or cerebrovascular disease.

[0356] The diagnosis using the above diagnostic model will be described in detail below through specific embodiments.

[0357] 2. Methods for Diagnosing Kidney Diseases

[0358] 2.1.1. Biomarkers for Kidney Diseases

[0359] In one embodiment, kidney disease biomarkers used to predict, measure, and confirm the presence, progression, etc., of kidney disease can be used in methods for diagnosing kidney disease. In this specification, kidney disease biomarkers may include risk assessment tools for kidney disease.

[0360] For example, kidney disease biomarkers may include eGFR (Estimated Glomerular Filtration Rate) values, albuminuria values, cystatin C values, KDIGO (Kidney Disease: Improving Global Outcomes) classifications, etc.

[0361] eGFR values ​​can be calculated based on serum creatinine concentration, age, sex, and ethnicity, and are used as indicators to assess kidney filtration function. For example, based on eGFR values, the risk of kidney disease can be divided into low risk (90 mL / min / 1.73m2 or above), intermediate risk (60-89 mL / min / 1.73m2), high-intermediate risk (45-59 mL / min / 1.73m2), very high risk (30-44 mL / min / 1.73m2), and highest risk (29 mL / min / 1.73m2 or below).

[0362] Albuminuria value is the proportion of albumin in urine and can be used as a major indicator for judging kidney disease. Albuminuria value can be the ratio of albumin to creatinine. For example, based on albuminuria value, the risk of kidney disease can be divided into low risk (less than 30 mg / g), intermediate risk (30-300 mg / g), and high risk (30-44 mL / min / 1.73m2).

[0363] Cystatin C levels can be a biomarker of kidney function measured in the blood. The higher the cystatin C level, the higher the risk of kidney function. For example, based on cystatin C levels, the risk of kidney disease can be divided into a low-risk group (1.0 mg / L) and a high-risk group (above 1.0 mg / L). In another example, based on cystatin C levels, the risk of kidney disease can be divided into a low-risk group (0.62–1.15 mg / L) and a high-risk group (above 1.15 mg / L). Furthermore, in another example, based on cystatin C values, the risk of kidney disease can be divided into a low-risk group (0.9 mg / L or less), a low-to-intermediate-risk group (0.9–1.2 mg / L), an intermediate-risk group (1.2–1.9 mg / L), an intermediate-to-high-risk group (1.9–3.0 mg / L), and a high-risk group (more than 3.0 mg / L).

[0364] The KDIGO classification can be a renal function biomarker determined by combining eGFR values ​​(or classifications) and albuminuria values ​​(or classifications). For example, the KDIGO classification can be divided into low-risk, intermediate-risk, and high-risk groups by combining eGFR values ​​(or classifications) and albuminuria values ​​(or classifications).

[0365] In addition, various renal disease biomarkers can be used in kidney disease diagnostic methods.

[0366] 2.1.2. Kidney Disease Diagnosis Method

[0367] A kidney disease diagnosis method according to an embodiment of this specification can be executed using at least one of the above-described diagnostic model, parallel diagnostic model, or serial diagnostic model. For ease of explanation, the following description will focus on executing the kidney disease diagnosis method using a serial diagnostic model.

[0368] FIG19 is a diagram illustrating a kidney disease diagnosis method according to an embodiment.

[0369] Referring to FIG19, the processor of the diagnostic device may include a step S100 of acquiring a retinal image and a step S200 of acquiring kidney disease diagnosis information.

[0370] In step S100, the processor of the diagnostic device may acquire a retinal image. Furthermore, according to an embodiment, the processor of the diagnostic device may perform preprocessing, enhancement, serialization, etc., on the acquired retinal image. Since the above can be applied here, detailed descriptions are omitted.

[0371] Furthermore, in step S200, the processor of the diagnostic device may acquire kidney disease diagnosis information. In this specification, kidney disease diagnosis information may be represented as Reti-CKD. The kidney disease diagnosis method will now be described with reference to FIG20.

[0372] Figure 20 illustrates a diagnostic model for obtaining diagnostic information about kidney disease according to one embodiment.

[0373] Referring to FIG20, the diagnostic model 1000 may be included in the processor and / or storage module of the diagnostic device. Furthermore, the descriptions of the diagnostic models in FIGS. 15 to 18 can be applied to the diagnostic model 1000.

[0374] The diagnostic model 1000 may include a first diagnostic model 1100 and a second diagnostic model 1200. The first diagnostic model 1100 and the second diagnostic model 1200 may be machine learning models. Furthermore, the first diagnostic model 1100 and the second diagnostic model 1200 may be models based on the same algorithm or models based on different algorithms. For example, the first diagnostic model 1100 and the second diagnostic model 1200 may be neural network models. Furthermore, the first diagnostic model 1100 may be a neural network model, and the second diagnostic model 1200 may be a non-neural network machine learning model. In the following description, for ease of explanation, the case where the first diagnostic model 1100 is a neural network model and the second diagnostic model 1200 is a non-neural network machine learning model will be emphasized, but the description in this specification is not limited thereto.

[0375] The processor of the diagnostic device can input a retinal image (or a preprocessed retinal image, a serialized retinal image) into the first diagnostic model 1100. Furthermore, the processor of the diagnostic device can obtain from the first diagnostic model 1100, which has input the retinal image, the probability value and / or classification that the subject in the retinal image is currently at high risk of kidney disease. For example, the processor of the diagnostic device can obtain from the first diagnostic model 1100 the probability that the subject's eGFR value in the current retinal image is less than or equal to 60 mL / min / 1.73 m2 and / or the probability that albuminuria is present in the current subject's urine (or the probability that the albuminuria value is greater than or equal to 300 mg / g) and / or the classification. Here, the probability value can be a value between 0 and 1.

[0376] Specifically, the first diagnostic model 1100 can be a neural network model with a CNN structure.

[0377] In one embodiment, in the training process of the first diagnostic model 1100, the training target retinal image can include retinal images of patients with kidney disease and retinal images of normal individuals without kidney disease.

[0378] Furthermore, in the training process of the first diagnostic model 1100, each monocular image of the left and right eyes of the target retinal image can be trained individually. In the training process, the ground truth can be whether or not the kidney disease is present. Additionally, predetermined information can be marked on each retinal image. For example, each retinal image can be marked with whether the eGFR value is less than or equal to 60 mL / min / 1.73 m2 and / or whether albuminuria is present in the subject's urine (e.g., whether the albuminuria value is greater than or equal to 300).A binary variable (mg / g). For example, if the eGFR value is less than or equal to 60 mL / min / 1.73 m2 or albuminuria is present in the subject's urine, the subject in the retinal image is labeled as having kidney disease; if the eGFR value is greater than 60 mL / min / 1.73 m2 and albuminuria is not present in the subject's urine, the subject in the retinal image can be labeled as not having kidney disease. Of course, the eGFR value and albuminuria value in the above example can be changed according to the embodiment. In addition, for example, the eGFR value and albuminuria value of the subject can be labeled on each retinal image.

[0379] In addition, during the evaluation of the first diagnostic model 1100, probability values ​​are assigned to the left eye retinal image and the right eye retinal image respectively, and the average of the probability values ​​can be regarded as the output of the test result.

[0380] In addition, the last fully connected layer of the first diagnostic model 1100 can perform a logit (one logit) probability prediction. The logit from the last fully connected layer can be converted into a probability using the Sigmoid function. Furthermore, the first diagnostic model 1100 can be trained to minimize the loss of the target and the prediction. In one embodiment, the first diagnostic model 1100 can be trained for 50 epochs using the AdamW Optimizer with a learning rate of 0.0002 and a cosine learning rate.

[0381] Furthermore, for training data augmentation of the target retinal image, at least one of mixup, cutmix, randaugment, an enhancing contrast module, and random cropping can be used. Furthermore, focal loss and exponential moving average are used in the training of the first diagnostic model 1100, and the size of the training target retinal image can be set to 384x384.

[0382] Furthermore, the processor of the diagnostic device can input the output value of the first diagnostic model 1100 and the subject's physical information (at least one of age, sex, race, smoking status, blood pressure (e.g., blood pressure value, hypertension status), diabetes status (or blood glucose value), and cholesterol value) into the second diagnostic model 1200. The processor of the diagnostic device can then obtain probability values ​​and / or classifications from the second diagnostic model 1200 regarding the probability of kidney disease-related events occurring within 5 years in the subject. Here, kidney disease-related events within 5 years can include fatal kidney disease, non-fatal kidney disease, and various events related to kidney disease (hospitalization, surgery, procedures, death, etc.) occurring within 5 years from the date of retinal image capture. Furthermore, in certain situations…In this case, the processor of the diagnostic device can obtain probability values ​​regarding the probability of kidney disease-related events occurring within 5 years for the subject from the second diagnostic model 1200, and apply a predetermined cutoff value to the obtained probability values ​​to obtain a classification corresponding to the obtained probability values. Furthermore, although this specification focuses on kidney disease-related events within 5 years, it is not limited thereto. According to the embodiments, kidney disease-related events within various periods such as 3 years and 10 years can certainly be applied to the description in this specification.

[0383] Therefore, the processor of the diagnostic device can output probability values ​​(scores) and / or classifications regarding the probability of kidney disease-related events occurring within 5 years for the subject as kidney disease diagnostic information.

[0384] Furthermore, according to the embodiments, the processor of the diagnostic device can obtain probability values ​​and / or classifications regarding the current risk of kidney disease from the second diagnostic model 1200. Furthermore, according to an embodiment, the processor of the diagnostic device can obtain probability values ​​and / or classifications of the probability of occurrence of kidney disease-related events in the subject within 5 years from the second diagnostic model 1200, and based on the obtained probability values ​​and / or classifications of the probability of occurrence of kidney disease-related events in the subject within 5 years, obtain probability values ​​and / or classifications of the current risk of kidney disease. For example, the processor of the diagnostic device can compare the probability values ​​and / or classifications of the probability of occurrence of kidney disease-related events in the subject within 5 years obtained from the second diagnostic model 1200 with at least one predetermined baseline probability value and / or classification, and based on the comparison result, obtain probability values ​​and / or classifications of the current risk of kidney disease.

[0385] For example, a first score and / or a first classification can be predetermined corresponding to kidney disease diagnostic information of high risk of existing biomarkers (e.g., eGFR value less than or equal to 44 mL / min / 1.73m2, albuminuria value greater than or equal to 300 mg / g and / or KDIGO classification high risk group or higher). Then, the processor of the diagnostic device can set the first score and / or the first grade to a first threshold and / or a first threshold grade. When the probability value and / or grade of the probability of occurrence of kidney disease-related events in the subject within 5 years, obtained from the second diagnostic model 1200, is greater than or equal to a predetermined first threshold and / or first threshold grade, the processor of the diagnostic device can determine that the current risk of kidney disease is high. Furthermore, although in the above example, one threshold score and / or threshold grade is described as one, it is not limited thereto, and multiple or more threshold scores and / or threshold grades can be set. For example, the threshold score and / or threshold grade may include a first threshold and / or first threshold grade corresponding to a high risk of the existing biomarker, and a second threshold and / or second threshold grade corresponding to a moderate risk of the existing biomarker.

[0386] Furthermore, in one embodiment, the second diagnostic model 1200 can be trained based on training target data. Here, the training of the second diagnostic model 1200 can include the meaning of the fitting of the second diagnostic model 1200. In addition, the second diagnostic model 1200 can be constructed using a Cox proportional hazards model based on linear regression.

[0387] For example, the training target data can include: the physical information of the subject in the training target retinal image of the first diagnostic model 1100, the tracking observation results of kidney disease events of the subject within 5 years, and the probability value of the first diagnostic model 1100 corresponding to the training target retinal image. In addition, according to the embodiment, the subject in the training target retinal image of the first diagnostic model 1100 and the subject in the training target data of the second diagnostic model 1200 can be the same or at least partially different. For example, in the training target data of the subject in the training target retinal image of the first diagnostic model 1100, the training target data of subjects who have had kidney disease events or have kidney disease at the baseline time point can be excluded from the training of the second diagnostic model 1200. This is because by training the second diagnostic model 1200 using training target data from healthy individuals who have not experienced kidney disease events and do not suffer from kidney disease, the accuracy of the 5-year kidney disease event probability output by the second diagnostic model 1200 can be improved.

[0388] Furthermore, in one embodiment, during the training process of the second diagnostic model 1200, the coefficients of the covariates of the Cox proportional hazards model can be set by fitting the Cox proportional hazards model to the UK Biobank cohort. And, in the UK Biobank cohort, the 5-year survival probability of kidney disease can be 0.9980896. And, the second diagnostic model 1200 can be modeled as the 5-year failure probability. Example 1 uses the UK Biobank cohort as an example for illustration, but the description in this specification is not limited thereto; the second diagnostic model can be fitted or calibrated based on cohorts other than the UK Biobank cohort.

[0389] Furthermore, in one embodiment, the second diagnostic model 1200 may include age, sex, presence or absence of hypertension, presence or absence of diabetes, and probability values ​​of the current risk of kidney disease output from the first diagnostic model 1100 as variables. Since these variables can be obtained from the subject using non-invasive methods, they can be used in the second diagnostic model 1200.

[0390] For example, the second diagnostic model 1200 may obtain probability values ​​of the probability of kidney disease-related events occurring in the subject within 5 years based on the following mathematical formula 1. For example, the Cox proportional hazards model of the second diagnostic model 1200 may be constructed based on mathematical formulas 1 and 2.

[0391] [Mathematical Formula 1]

[0392] Predicted risk = 1 - SP^(Exp[LP])

[0393] Wherein, Predicted risk is the probability value of the occurrence of kidney disease-related events within 5 years, and SP (Survival Probability) is the survival probability of kidney disease within 5 years. For example, in the UK Biobank example above, page 28 / 43, 31 CN 121586537 A, it can be 0.9980896. And, LP (Linear Predictor) will be explained in the following Mathematical Formula 2.

[0394] [Mathematical Formula 2]

[0395] LP = a1*RPS*100 + a2*Age + a3*Female + a4*Hypertension + a5*Diabetes

[0396] Wherein, RPS (Retinal photograph-based Prediction Score) is the probability value of the current kidney disease risk output from the first diagnostic model 1100, Age is age, and Female represents the sex of the examinee. For example, Female can be an indicator function that reflects a value of 1 when the subject is female and a value of 0 when the subject is male. Hypertension reflects a value of 1 when the subject has hypertension and a value of 0 when the subject does not have hypertension. For example, for a user taking antihypertensive medication, it can be set to have hypertension. Furthermore, Diabetes reflects a value of 1 when the subject is diabetic (or prediabetic) and a value of 0 when the subject is not diabetic (or prediabetic).

[0397] Furthermore, a1 to a5 are the coefficients of the covariates of each variable, for example, by fitting the second diagnostic model 1200, a1 to a5 can be set to 0.0546351, 0.0357466, -0.3317242, -0.1207903, and 0.4679712, respectively. Furthermore, the processor of the diagnostic device can, based on (or applying) the mathematical formulas 1 and 2, give the risk of kidney disease over 3 years, 5 years, or 10 years. For example, the processor of the diagnostic device can add or replace race, smoking status, cholesterol level, etc., in the covariates of mathematical formula 2, and set the coefficients of each covariate by fitting the second diagnostic model 1200.

[0398] In addition, the processor of the diagnostic device can obtain the probability value of the occurrence probability of kidney disease-related events within 5 years from the second diagnostic model 1200, and apply a predetermined cutoff value to the probability value to obtain the classification corresponding to the probability value. Of course, in some cases, the predetermined cutoff value can also be applied to the second diagnostic model 1200 in advance to obtain the classification corresponding to the probability value.The second diagnostic model 1200 obtains a classification corresponding to the probability value. At this time, there can be one or more cutoff values. For example, if there is one cutoff value, there can be two classifications corresponding to the probability value; if there are two cutoff values, there can be three classifications corresponding to the probability value. The description in this specification can be applied to various numbers of cutoff values ​​and various cutoff values.

[0399] In one embodiment, the processor of the diagnostic device can set a predetermined cutoff value based on the population distribution (i.e., the incidence rate of each group) of multiple classifications, according to the tracking observation results of a predetermined population. For example, the cutoff value can be determined based on the results of tracking observation of a predetermined population. For example, if, based on the entire tracking observation subjects, tracking observation results are obtained showing 0~n1% as the low-risk group, n1~n2% as the medium-risk group, n2~n3% as the high-risk group, and n3~100% as the highest-risk group, then n1%, n2%, and n3% can be set as cutoff values ​​respectively. Furthermore, even if n1%, n2%, and n3% are not cutoff values, cutoff values ​​can be set to make the population distribution of each risk group in the kidney disease diagnostic information similar to the follow-up observation results of the predetermined population.

[0400] For example, if the predetermined population is the UK Biobank cohort and the Korean Diabetic cohort, cutoff values ​​can be set according to the population distribution of the low-risk, intermediate-risk, high-risk, and highest-risk groups in the UK Biobank cohort and the Korean Diabetic cohort, so that the population distribution of each risk group in the kidney disease diagnostic information is similar to the population distribution of each risk group in the UK Biobank cohort and the Korean Diabetic cohort.

[0401] In addition, the processor of the diagnostic device can set cutoff values ​​based on the incidence rate of kidney disease biomarkers. For example, based on the eGFR value, the risk of kidney disease can be categorized into low risk (90 mL / min / 1.73m² or higher), intermediate risk (60–89 mL / min / 1.73m²), high-intermediate risk (45–59 mL / min / 1.73m²), very high risk (30–44 mL / min / 1.73m²), and highest risk (29 mL / min / 1.73m² or lower). Furthermore, if the incidence rate at the eGFR value, corresponding to the proportions of the population at low, intermediate, high-intermediate, very high, and highest risk, are x1%, x2%, x3%, x4%, and x5%, respectively, then a cutoff value can be set so that the proportions of the population included in each category of the kidney disease diagnostic information output by the processor of the diagnostic device match the stated x1%, x2%, x3%, x4%, and x5%. This can be applied not only to eGFR values ​​but also to the information on pages 29 / 43 of the instruction manual (CN 121586537 A).Biomarkers such as albuminuria level, cystatin C level, and KDIGO classification.

[0402] In addition, the incidence rate according to the Kidney Failure Risk Equation and the incidence rate according to the Risk Prediction Equation can also be used to set the above cutoff value.

[0403] In addition, the cutoff value can be an optimized value to most effectively detect the group corresponding to the high risk of each biomarker.

[0404] Specifically, the processor of the diagnostic device can set the cutoff value based on the proportion of high-risk and low-risk groups classified by the 5-year kidney risk calculator, according to the 5-year kidney risk calculation method based on race or applicable clinical guidelines. Therefore, the predictive performance of kidney disease diagnostic information for high-risk groups can be improved. For example, in order to maximize the diagnostic performance of kidney disease diagnostic information for high-risk groups in eGFR values, the cutoff value of kidney disease diagnostic information can be set such that the proportion of all subjects in the low-risk group and the proportion of all subjects in the high-risk group of kidney disease diagnostic information is similar to the proportion of subjects in the low-risk group and the proportion of subjects in the high-risk group of eGFR values. For example, the processor of the diagnostic device can set a cutoff value for kidney disease diagnostic information such that the proportion of all subjects in the low-risk group and the proportion of all subjects in the high-risk group of kidney disease diagnostic information are included within a predetermined range of the proportion of subjects in the low-risk group and the proportion of subjects in the high-risk group of eGFR values. In this case, the processor of the diagnostic device can accurately determine the subjects corresponding to the low-risk group and the high-risk group of eGFR values.

[0405] Furthermore, the biomarkers used to train the first diagnostic model 1100 and the biomarkers used to set the cutoff values ​​of the second diagnostic model 1200 can be the same or different.

[0406] Furthermore, when setting cutoff values ​​based on each biomarker, the classification of kidney disease diagnostic information output by the processor of the diagnostic device can be similar to the actual results of the biomarkers with the cutoff values ​​set. For example, if the cutoff value corresponding to the eGFR biomarker is applied to the probability value of the occurrence of kidney disease-related events within 5 years output from the second diagnostic model 1200, and the processor of the diagnostic device outputs the grade of the subject, then the grade of the subject output by the processor of the diagnostic device can match the grade determined after the subject actually performs eGFR-related tests. If tests such as eGFR, albumin urine value, and cystatin C are actually performed, the subject may experience inconveniences such as blood collection or urine tests. However, according to the kidney disease diagnosis method of this specification, since highly accurate kidney disease diagnosis information can be obtained using only the retinal image of the subject, user convenience can be improved.

[0407] Furthermore, in one embodiment, when kidney disease diagnosis information is applied to the scores of existing biomarkers and / orWhen grading, the scores and / or grading of existing biomarkers, and the incidence of kidney disease-related events corresponding to those scores and / or gradings, can be differentiated or stratified according to the groups of kidney disease diagnostic information (e.g., low-risk group, intermediate-risk group, high-risk group, and highest-risk group). Furthermore, the cutoff value for kidney disease diagnostic information can be set such that when kidney disease diagnostic information is applied to the scores and / or gradings of existing biomarkers, the incidence of kidney disease-related events corresponding to those scores and / or gradings is explicitly stratified according to the groups of kidney disease diagnostic information.

[0408] Embodiments of the kidney disease diagnostic method according to this specification will be described in detail below.

[0409] 2.1.3. Example 1

[0410] Example 1 illustrates experimental results of the kidney disease diagnostic method according to this specification using clinical data and retinal images from the UK Biobank cohort and the Korean diabetes cohort.

[0411] First, in Example 1, the first diagnostic model 1100 can be trained based on 158,216 retinal images (79,108 people) from a Korean health checkup center and clinical data (whether the eGFR value of each retinal image is less than or equal to 60 mL / min / 1.73 m2 and / or whether albuminuria is present in the urine of the examinee).

[0412] Furthermore, in Example 1, the second diagnostic model 1200 can be trained (or fitted) using the UK Biobank cohort (30,477 people) and the Korean diabetes cohort (5,014 people). Among them, data from participants with chronic kidney disease (CKD), participants with eGFR values ​​less than 90 mL / min / 1.73 m2, or participants with detected albuminuria (or albuminuria values ​​greater than or equal to 30 mg / Cr) can be excluded from the training of the second diagnostic model 1200. In this case, the processor of the diagnostic device can provide a probability value and / or classification of the probability of occurrence of kidney disease-related events within 5 years with high accuracy for normal individuals who do not currently have kidney disease.

[0413] In addition, the content illustrated in FIG20 can be applied to the first diagnostic model 1100 and the second diagnostic model 1200 in Embodiment 1.

[0414] The results of Embodiment 1 will be described below. In the following figures, the Reti-CKD score represents the probability value of occurrence of kidney disease-related events within 5 years as kidney disease diagnostic information, and the various classifications of the Reti-CKD score can be used as kidney disease diagnostic information to represent the classification corresponding to the respective probability values.

[0415] FIG21 shows the clinical characteristics of the subjects corresponding to the kidney disease diagnostic information according to Embodiment 1.

[0416] Referring to FIG21, the Korean health checkup data (Health) used to train the first diagnostic model 1100The screening data (for 79,108 participants with a mean age of 49.5 years (SD: 11.8) and a mean eGFR of 100.3 mL / min / 1.73 m² (SD: 14.3) were used to train the second diagnostic model 1200. Additionally, the UK Biobank cohort data (for 30,477 participants) included 720 (2.4%) participants who were diagnosed with chronic kidney disease during a mean follow-up period of 10.8 years (interquartile range (IQR): 10.7–11.0). Furthermore, the Korean diabetes cohort data (for 5,014 participants) included 206 (4.1%) participants who were diagnosed with chronic kidney disease during a mean follow-up period of 6.1 years (IQR: 4.0–8.4). Furthermore, the mean eGFR value of the UK Biobank cohort can be 99.4 (standard deviation (SD): 6.6) mL / min / 1.73 m2, and the mean eGFR value of the Korean diabetes cohort can be 102.5 (standard deviation (SD): 9.1) mL / min / 1.73 m2.

[0417] Figures 22a and 22b are graphs illustrating the incidence of kidney disease-related events based on kidney disease diagnostic information according to Example 1.

[0418] Referring to Figures 22a and 22b, in the graphs of Figures 22a and 22b, the x-axis represents time (years), and the y-axis represents the incidence of chronic kidney disease-related events.

[0419] Additionally, Figure 22a shows the Kaplan-Meier curves of the results of Kaplan-Meier survival analysis performed on four groups of subjects in the UK Biobank cohort, divided into four categories based on kidney disease diagnostic information; Figure 22b shows the Kaplan-Meier curves of the results of Kaplan-Meier survival analysis performed on four groups of subjects in the Korean diabetes cohort, divided into four categories based on kidney disease diagnostic information. In the UK Biobank cohort of Figure 22a, 321,317 person-years were investigated over a mean follow-up period of 10.8 years, and in the Korean diabetes cohort of Figure 22b, 30,122 person-years were investigated over a mean follow-up period of 6.1 years.

[0420] As shown in Figures 22a and 22b, kidney disease diagnostic information can be used to clearly distinguish kidney disease risk in both the UK Biobank cohort and the Korean diabetes cohort based on four categories. In Figures 22a and 22b, the hazard ratio (HR) for the development of chronic kidney disease can be shown as a dose-dependent association according to four levels.The adjusted HRs per 1 SD increment of kidney disease diagnostic information can be 1.34 (95% confidence interval: 1.27–1.41) in the UK Biobank cohort in Figure 22a and 1.94 (95% confidence interval: 1.63–2.31) in the Korean diabetes cohort in Figure 22b.

[0421] Furthermore, kidney disease diagnostic information can even clearly distinguish the risk of kidney disease with significant hazard ratios in subgroups divided according to sex, age, presence or absence of hypertension, and presence or absence of diabetes. For example, in the UK Biobank cohort in Figure 22a, the trend of the hazard ratio for chronic kidney disease can be: overall 1.52 (95% confidence interval: 1.42–1.63), male 1.47. (Instructions 31 / 43 pages 34 CN 121586537 A) (95% confidence interval: 1.31–1.65), Female_1.51 (95% confidence interval: 1.36–1.67), Under 55_1.35 (95% confidence interval: 1.20–1.51), Over 55_1.46 (95% confidence interval: 1.29–1.65), No hypertension_1.59 (95% confidence interval: 1.47–1.74), With hypertension_1.32 (9 5% confidence interval: 1.12–1.56), no diabetes: 1.50 (95% confidence interval: 1.39–1.61), with diabetes: 1.35 (95% confidence interval: 1.07–1.72), eGFR greater than 100: 1.53 (95% confidence interval: 1.34–1.73), eGFR less than or equal to 100: 1.41 (95% confidence interval: 1.30–1.53). Furthermore, in the Korean diabetes cohort in (b), the trends in hazard ratios for chronic kidney disease could be: overall 1.99 (95% confidence interval: 1.72–2.31), males 2.17 (95% confidence interval: 1.73–2.70), females 1.85 (95% confidence interval: 1.51–2.28), under 55 years old 1.75 (95% confidence interval: 1.28–2.39), over 55 years old 1.97 (95% confidence interval: 1.58–2.46), no hypertension 2.13 (95% confidence interval: 1.76–2.57), with hypertension 1.71 (95% confidence interval: 1.37–2.12), eGFR greater than 100 2.10 (95% confidence interval: 1.62–2.72), eGFR less than or equal to 100 1 .49 (95% confidence interval: 1.28–1.78).

[0422] Figure 23 is used to illustrate the correlation between kidney disease diagnostic information and kidney disease-related events within 5 years according to Example 1.A graph showing the predictive performance of the kidney disease diagnosis information (Reti-CKD) and the eGFR-based kidney disease score (eGFR-CKD score) in the UK Biobank cohort and the Korean diabetes cohort. The eGFR-based kidney disease score can be used as a biomarker to assess the risk of existing kidney disease.

[0424] In the table of Figure 23, the NRI (Net reclassification Index) can be used to represent an indicator of how much better the new model is than the old model. For the NRI, it is 0.109 in the UK Biobank cohort (95% confidence interval (CI): 0.44–0.156) and 0.179 in the Korean diabetes cohort (95% confidence interval: 0.017–0.292), showing similar results. Furthermore, when comparing scores corresponding to kidney disease diagnostic information and eGFR-based kidney disease scores, the difference in C-statistics was 0.020 (95% confidence interval (CI): 0.011–0.029) in the UK Biobank cohort and 0.024 (95% confidence interval: 0.002–0.046) in the Korean diabetes cohort. Scores corresponding to kidney disease diagnostic information demonstrated significantly higher performance than eGFR-based kidney disease scores. Therefore, scores corresponding to kidney disease diagnostic information exhibit superior accuracy when compared to existing eGFR-based kidney disease scores.

[0425] Furthermore, although not shown in the table of Figure 23, if a sensitivity analysis is performed, including all subjects not limited to those with eGFR values ​​greater than or equal to 90 mL / min / 1.73 m², the difference in C-statistics between scores corresponding to kidney disease diagnostic information and eGFR-based kidney disease scores was 0.008 (95% confidence interval (CI): 0.001–0.016) in the UK Biobank cohort and 0.057 (95% confidence interval: 0.048–0.067) in the Korean diabetes cohort. Scores corresponding to kidney disease diagnostic information showed significantly higher performance than eGFR-based kidney disease scores. Therefore, scores corresponding to kidney disease diagnostic information exhibit superior performance with higher sensitivity when compared to existing eGFR-based kidney disease scores. Therefore, even without eGFR testing via blood tests (e.g., even if it is unknown whether the eGFR value is greater than or equal to 90 mL / min / 1.73 m2), kidney disease diagnostic information can stratify the future risk of kidney disease in the general population and patients with diabetes better than eGFR.

[0426] Furthermore, although not shown in the table of Figure 23, the sensitivity analysis can be performed in groups with hypertension and / or diabetes and groups without hypertension and diabetes.

[0427] In the group without hypertension and diabetes, when comparing the score corresponding to the kidney disease diagnostic information and the eGFR-based kidney disease score, the difference in the C-statistic was 0.025 (95% confidence interval (CI): 0.002–0.048) in the UK Biobank cohort, and the score corresponding to the kidney disease diagnostic information showed significantly higher performance than the eGFR-based kidney disease score. Therefore, the score corresponding to the kidney disease diagnostic information shows superior performance with higher sensitivity than the existing eGFR-based kidney disease score for normal individuals without hypertension and diabetes.

[0428] Furthermore, in the group with hypertension and / or diabetes, when comparing the score corresponding to the kidney disease diagnostic information and the eGFR-based kidney disease score, the difference in the C-statistic was 0.019 (95% confidence interval (CI): 0.008–0.030) in the UK Biobank cohort, and the score corresponding to the kidney disease diagnostic information showed significantly higher performance than the eGFR-based kidney disease score. Therefore, the score corresponding to the kidney disease diagnostic information showed superior performance with higher sensitivity than the existing eGFR-based kidney disease score, regardless of whether hypertension and / or diabetes were present. Therefore, it can be seen that it is applicable not only to the population with hypertension and diabetes, which are the most important risk factors for kidney disease, but also to kidney disease caused by problems of the kidney itself that are unrelated to hypertension / diabetes.

[0429] 2.1.4. Obtaining information on kidney disease risk based on existing biomarkers

[0430] Figure 24 is a diagram illustrating a kidney disease diagnostic method according to another embodiment.

[0431] Referring to FIG24, a kidney disease diagnosis method according to another embodiment may include: a step S300 of acquiring a retinal image and a step S400 of acquiring kidney disease risk information based on existing biomarkers as kidney disease diagnostic information.

[0432] In step S300, the processor of the diagnostic device may acquire a retinal image. Since FIG19 and the above description are applicable here, a detailed description is omitted.

[0433] Furthermore, in step S400, the processor of the diagnostic device may acquire kidney disease risk information based on existing biomarkers as kidney disease diagnostic information. The above description regarding the diagnostic model is applicable to step S400.

[0434] Here, information on kidney disease risk based on existing biomarkers refers to the kidney disease risk calculated based on existing biomarkers. The processor of the diagnostic device can use retinal images to acquire information on the kidney disease risk calculated based on existing biomarkers. For example, according to step S400, the processor of the diagnostic device can use a diagnostic model based on retinal images to acquire information about eGFR values ​​and / or the corresponding grades of eGFR values, albuminuria values ​​and / or the corresponding grades of albuminuria values, cystatin C values ​​and / or the corresponding grades of cystatin C values, and output the acquired information, or output information about the final score and / or the corresponding grades of the final score based on the acquired information.

[0435] In one embodiment, the processor of the diagnostic device can use a diagnostic model to acquire information on kidney disease risk based on existing biomarkers as kidney disease diagnostic information.

[0436] For example, learning can be performed using retinal images and information on kidney disease risk based on existing biomarkers labeled on the retinal images. For example, at least one of the following can be labeled on each retinal image: eGFR value (and / or corresponding grade), albuminuria value (and / or corresponding grade), or cystatin C value (and / or corresponding grade). The diagnostic model can learn using the retinal images and their corresponding labels. Furthermore, the labels on the retinal images can include information about at least one of the following: age, sex, ethnicity, smoking status, blood pressure (hypertension or not), and diabetes status of the subject of the retinal image.

[0437] The processor of the diagnostic device can obtain information on the risk of kidney disease based on existing biomarkers by inputting retinal images into a single, learned diagnostic model. For example, if the diagnostic model is learned based on eGFR values, the processor of the diagnostic device can input retinal images into the diagnostic model and obtain information on eGFR values ​​and / or grades based on eGFR values ​​from the diagnostic model.

[0438] Furthermore, according to an embodiment, the processor of the diagnostic device can also input information about at least one of the subject's age, sex, ethnicity, smoking status, blood pressure (hypertension), diabetes status, or cholesterol level, along with a retinal image, into the diagnostic model to obtain eGFR values ​​and / or information on the classification based on eGFR values ​​from the diagnostic model. Specification 33 / 43 pages 36 CN 121586537 A

[0439] Furthermore, in one embodiment, the diagnostic models can be constructed in parallel. For example, the description of the diagnostic models in 1.4.1.1 to 1.4.1.3 can be applied to the diagnostic models.

[0440] Specifically, the diagnostic model can include multiple diagnostic models. For example, the diagnostic model can include a first diagnostic model and a second diagnostic model. The first and second diagnostic models can utilize kidneys corresponding to different biomarkers.Disease risk information is learned. For example, a first diagnostic model can learn using retinal images labeled with eGFR values ​​(and / or corresponding grades), and a second diagnostic model can learn using retinal images labeled with albuminuria values ​​(and / or corresponding grades). Of course, this is not limited to this; the diagnostic model may include additional diagnostic models, such as a third / fourth diagnostic model together with the first / second diagnostic model. In addition, the labels on the retinal images may include information about at least one of the following: age, sex, race, smoking status, blood pressure (hypertension or not), and diabetes status of the subject in the retinal image.

[0441] Furthermore, the processor of the diagnostic device can obtain kidney disease risk information corresponding to different biomarkers by inputting retinal images into the first and second diagnostic models. For example, the processor of the diagnostic device can obtain the subject's eGFR value (and / or corresponding grade) from the first diagnostic model, obtain the subject's albuminuria value (and / or corresponding grade) from the second diagnostic model, and output the obtained information. In addition, the processor of the diagnostic device can output information about the final score and / or the grade corresponding to the final score based on the information obtained from each diagnostic model. For example, information about the final score and / or the corresponding grade can be obtained and output based on the eGFR value (and / or corresponding grade) of the subject obtained from the first diagnostic model and the albuminuria value (and / or corresponding grade) of the subject obtained from the second diagnostic model. Therefore, the processor of the diagnostic device can non-invasively acquire information on the risk of kidney disease based on existing biomarkers, serving as diagnostic information for kidney disease, using retinal images without the need for invasive methods such as blood sampling.

[0442] Furthermore, in one embodiment, the diagnostic models can be constructed sequentially. For example, the descriptions of the diagnostic models in 1.4.2.1, 1.4.2.2, and 2.1.2 can be applied to the diagnostic models.

[0443] For example, the diagnostic model can include a first diagnostic model and a second diagnostic model, as shown in FIG20. For example, the processor of the diagnostic device can input retinal images into a first diagnostic model and obtain from the first diagnostic model the probability value that the subject is currently at high risk of kidney disease (the probability that the subject's eGFR value is less than or equal to 60 mL / min / 1.73 m2 and / or the probability that albuminuria is present in the subject's urine). Furthermore, the processor of the diagnostic device can input the output value of the first diagnostic model and the subject's physical information (at least one of age, sex, race, smoking status, blood pressure (e.g., blood pressure value, hypertension status), diabetes status (or blood glucose value), and cholesterol value). At this time, a second diagnostic model...Cutoff values ​​based on various biomarkers (such as eGFR value, albuminuria value, cystatin C value) can be set. For example, the cutoff value of the second diagnostic model can be set to a population proportion similar to the high-risk group (45-59 mL / min / 1.73m2) in eGFR value, and / or a population proportion similar to the intermediate-risk group (30-300 mg / g) in albuminuria value. Therefore, kidney disease diagnostic information can predict the high-risk group (or intermediate-risk group) corresponding to various biomarkers with high accuracy.

[0444] 2.1.5. Providing guidance information on kidney disease diagnostic information

[0445] FIG25 is a diagram illustrating a method for providing guidance information on kidney disease diagnostic information according to one embodiment.

[0446] Referring to FIG25, the method for providing guidance information according to one embodiment may include: step S500 of acquiring kidney disease diagnostic information and step S600 of providing guidance information on diagnostic information.

[0447] In step S500, the processor of the diagnostic device may acquire diagnostic information. Since the above description can be applied to step S500, detailed description is omitted. Specification 34 / 43 pages 37 CN 121586537 A

[0448] Furthermore, in step S600, the processor of the diagnostic device can provide guidance information regarding the diagnostic information. For ease of explanation, steps S500 and S600 are described with a focus on kidney disease, but are not limited thereto. The guidance information in this specification can also be applied to various diseases such as ophthalmic diseases and cardiovascular diseases. Therefore, steps S500 and S600 will be described with a focus on the guidance information corresponding to the kidney diagnostic information.

[0449] In one embodiment, the guidance information may refer to medical / non-medical treatment information recommended to the examinee based on the kidney disease diagnostic information. For example, the guidance information may include prescription information, treatment information, and management information.

[0450] Prescription information may refer to information on medications (e.g., prescription drugs) recommended to the examinee to maintain or improve the kidney disease risk corresponding to the kidney disease diagnostic information. Furthermore, the prescription information may include information about the prescription drug, the time of administration, and the dosage. For example, prescription information may include information about prescription ACE inhibitors (e.g., lisinopril, enalapril), hydrochlorothiazide, cholesterol-lowering drugs (e.g., statins, atorvastatin, rosuvastatin), and phosphate binders (e.g., calcitriol).Information on one or more of the following: Calcitriol, Sevelamer, and anticoagulants (e.g., warfarin). Additionally, the prescription information may include SGLT2 inhibitors (Sodium-Glucose Co-Transporter 2 Inhibitors) (e.g., Dapagliflozin, Empagliflozin, Canagliflozin) as a combination drug for diabetes prescriptions, ARBs (Angiotensin II Receptor Blockers) (e.g., Losartan, Valsartan) as a combination drug for hypertension prescriptions, and GLP1 (Glucagon-Like Peptide-1) agonists and / or diuretics (e.g., furosemide) as a combination drug for diabetes and obesity prescriptions.

[0451] Furthermore, the processor of the diagnostic device may provide prescription information considering medications that require attention when administering medication to patients with kidney disease. For example, medications requiring caution when administering to patients with the aforementioned diseases include analgesics and antipyretics (nonsteroidal anti-inflammatory drugs (NSAIDs), high-dose aspirin), antibiotics (aminoglycosides, amphotericin B, cephalosporins, penicillins, beta-lactamase inhibitors, quinolones, rifampin, sulfonamides, vancomycin), antiviral drugs (acyclovir, adefovir, ganciclovir, atazanavir, indinavir, tenofovir), and bisphosphonates (pamidronate, zoledronic acid). (acid), calcineurin inhibitors (cyclosporine, tacrolimus), anticancer drugs (alkylating agents, cisplatin)Cisplatin, methotrexate, mitomycin, interferon-alpha, proteasome inhibitors, vascular endothelial growth factor (VEGF) inhibitors, checkpoint inhibitors, iodine contrast agents used in CT and angiography, diuretics (loop diuretics, thiazides, triamterene), proton pump inhibitors (dexlansoprazole, esomeprazole, lansoprazole, omeprazole, pantoprazole, rabeprazole), and other drugs (allopurinol, gold sodium thiomalate). thiomalate), lithium, quinine, sodium phosphate, and traditional Chinese medicine (Aristolochic acid, Cat's claw, licorice root).

[0452] In addition, treatment information may refer to subsequent treatment information recommended to the examinee in order to maintain or improve the risk of kidney disease corresponding to the kidney disease diagnosis information. For example, additional examination information may include information about secondary diagnosis or medical treatment for the examinee. For example, additional examination information may include additional required examinations, information about hospitals / medical personnel that can perform additional examinations, and information about recommended procedures / surgeries.

[0453] In addition, management information may include non-medical treatment information recommended to the examinee in order to maintain or improve the risk of kidney disease corresponding to the kidney disease diagnosis information. For example, management information may include information on lifestyle habits, dietary habits, exercise, and non-prescription drugs such as nutritional supplements to reduce the risk of kidney disease.

[0454] Furthermore, in one embodiment, the diagnostic device may be linked with an external monitoring device. Here, the monitoring device may refer to a device that monitors the subject's lifestyle habits or behaviors. For example, the monitoring device may include portable devices, wearable devices, etc.Equipment, health measurement devices, etc. Furthermore, the monitoring device can be the aforementioned client device. For example, the monitoring device may include an imaging unit, which captures images of the inside and outside of the eyeball to obtain images of the inside and outside of the eyeball.

[0455] 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 glucose level, body water content, oxygen level, body temperature, oxygen saturation, pulse wave, medical visit status, whether an examination was performed, whether surgery was performed, eyeball images, eGFR value, albumin urine value, cystatin C value, etc.

[0456] The processor of the diagnostic device can communicate with the monitoring device via a wired or wireless communication module.

[0457] 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 can obtain management information (e.g., lifestyle information, dietary information, exercise information) from the diagnostic device as guidance information, compare the management information with the monitored information, determine whether the monitored information matches the management information, and provide the judgment result and / or corresponding additional information.

[0458] For example, if the monitored exercise time is less than the exercise time in the management information, the monitoring device can provide information to the subject, requiring them to exercise according to the management information. In addition, if the food intake in the monitored information corresponds to the food intake in the management information, the monitoring device can provide information to the subject, informing them that they are eating well according to the management information.

[0459] In addition, the processor of the diagnostic device can obtain the monitored information from the monitoring device. The processor of the diagnostic device can provide various information to the subject based on the guidance information and the monitored information. For example, the processor of the diagnostic device can compare the monitored information with the guidance information, determine whether the monitored information matches the management information, and provide the judgment result and / or corresponding additional information. The above examples of the monitoring device can be applied to the operation of the processor of the diagnostic device.

[0460] In addition, the processor of the diagnostic device can generate guidance information based on the monitoring information received from the monitoring device. For example, the processor of the diagnostic device can acquire the subject's status information (exercise status, lifestyle status, dietary habits, etc.) based on monitoring information, and modify the guidance information identified as kidney disease diagnostic information based on the subject's status information to suit the subject.

[0461] In one embodiment, the processor of the diagnostic device can provide guidance information using a predetermined database. For example, the diagnostic device may include a database that matches the scores and / or grades of kidney disease diagnostic information with guidance information. For example, if the kidney disease diagnostic information is represented by 3 grades, the database may include: guidance information matched with the low-risk grade.Information (e.g., prescription information - none, treatment information - information about the next appointment time, management information - providing dietary information, providing exercise information), guidance information matched to the medium-risk stratification (e.g., prescription information - none, treatment information - providing additional examination information, management information - providing dietary information, providing exercise information, providing over-the-counter drug information), and guidance information matched to the high-risk stratification (e.g., prescription information - providing statin prescription information, treatment information - additional examination information, providing recommended treatment / surgery information, management information - providing dietary information, providing exercise information, providing over-the-counter drug information). The processor of the diagnostic device can provide guidance information matched to the kidney disease diagnostic information based on a database.

[0462] Furthermore, 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 scores and / or stratifications of the kidney disease diagnostic information and the guidance information. In addition, the guidance information model can also be learned along with at least one of the following: height, weight, sex, race, smoking status, blood pressure (e.g., blood pressure value, hypertension status), diabetes status (or blood glucose value), and cholesterol value. Therefore, the processor of the diagnostic device can input the score and / or classification of the kidney disease diagnostic information and the subject's physical information into the guidance information model to obtain guidance information about the subject. Furthermore, according to embodiments, the guidance information model can be included in the diagnostic model or constructed independently of the diagnostic model.

[0463] 2.1.6. Providing guidance information using existing prescription information and kidney disease auxiliary information

[0464] FIG26 is a diagram illustrating a method for providing guidance information using existing prescription information and kidney disease diagnostic information according to one embodiment.

[0465] Referring to FIG26, the method for providing guidance information according to one embodiment may include: step S710 of obtaining existing prescription information of the subject, step S720 of obtaining kidney disease diagnostic information, and step S730 of providing guidance information based on existing prescription information and kidney disease diagnostic information.

[0466] In some cases, kidney disease often progresses alongside underlying conditions such as diabetes, hypertension, and / or obesity. Furthermore, some prescription medications for diabetes, hypertension, and / or obesity are also effective in treating kidney disease. For example, SGLT2 inhibitors are prescription medications for diabetes but are also combination drugs that are effective in treating kidney disease; ARBs are prescription medications for hypertension but are also combination drugs that are effective in treating kidney disease. Additionally, GLP1 (Glucagon-Like Peptide-1)...Agonists are prescription drugs for diabetes and / or obesity, but also combination drugs that are effective in treating kidney disease.

[0467] If the subject is taking such combination drugs, depending on the subject's risk of kidney disease, it may not be necessary to prescribe additional kidney disease medications, or it may be necessary to exclude medications that would reduce the subject's treatment effectiveness when taken with the combination drugs, and to provide the subject with prescription information. For example, there are also studies indicating that it is not recommended to prescribe ARBs (prescription drugs for hypertension) and ACE inhibitors (prescription drugs for kidney disease) in combination. In this way, when providing guidance information, especially prescription information, to the subject, the medications that the subject is taking must be taken into account in order to provide accurate prescription information.

[0468] Therefore, the processor of the diagnostic device can provide guidance information, especially prescription information, based on the subject's existing prescription information and kidney disease diagnosis information.

[0469] According to step S710, the processor of the diagnostic device can obtain the subject's existing prescription information. For example, the processor of the diagnostic device can obtain existing prescription information from an external device or the input module of the diagnostic device. Furthermore, the processor of the diagnostic device can acquire information about the patient's underlying diseases (e.g., hypertension, blood pressure, diabetes, blood sugar, obesity, BMI, cholesterol, etc.) instead of or together with existing prescription information.

[0470] Furthermore, in step S720, the processor of the diagnostic device can acquire kidney disease diagnostic information. Here, kidney disease diagnostic information may refer to kidney disease diagnostic information based on retinal images of the same patient as the patient acquired in step S710 with existing prescription information (and / or underlying disease information). Since the above description can be applied to step S720, detailed description is omitted. Specification 37 / 43 pages 40 CN 121586537 A

[0471] Furthermore, in step S730, the processor of the diagnostic device can provide guidance information based on existing prescription information and kidney disease diagnostic information.

[0472] The processor of the diagnostic device can provide guidance information based on kidney disease diagnostic information, as in 2.1.5 above. However, in step S730, not only kidney disease diagnostic information but also existing prescription information (and / or underlying disease information) are considered to provide guidance information.

[0473] In one embodiment, if the score and / or grade of the kidney disease diagnostic information belongs to the low-risk group, the processor of the diagnostic device may not provide prescription information in the guidance information. Furthermore, the processor of the diagnostic device can know from existing prescription information (and / or underlying disease information) that the examinee is receiving a prescription for treatment of an underlying disease, or if the underlying disease risk is high, and can provide information that is helpful for the treatment and even management of the underlying disease when providing treatment information and / or management information.

[0474] In addition, if the score and / or grade of the kidney disease diagnostic information belongs to the medium-risk group and / or high-risk group,The processor of the diagnostic device can provide prescription information from the guidance information. If the existing prescription information includes combination drugs that can treat both the underlying disease and kidney disease, such as SGLT2 inhibitors, ARBs, and GLP1 agonists, the risk of the kidney disease diagnostic information can be taken into account when providing prescription information. For example, if the risk of kidney disease diagnosis assistance is moderate, the processor of the diagnostic device can provide prescription information indicating that no new kidney disease treatment drug should be prescribed, but rather the existing combination drug should be maintained. In addition, if the risk of kidney disease diagnosis assistance is high, the processor of the diagnostic device can provide prescription information about kidney disease treatment drugs that can be taken with the combination drug in the existing prescription information.

[0475] In addition, if the existing prescription information does not include treatment drugs for the underlying disease, such as combination drugs, or if the risk of the underlying disease is low, and the risk of kidney disease diagnosis assistance is moderate or high, the processor of the diagnostic device can provide prescription information about kidney disease treatment drugs (or combination drugs).

[0476] However, these examples are only one example among various embodiments of this specification. Not limited to these examples, the processor of the diagnostic device can provide guidance information using a predetermined database and / or guidance information model, as described in 2.1.5. At this time, the database can store guidance information corresponding to existing prescription information and kidney disease diagnosis information, and the guidance information model can also learn based on the guidance information corresponding to existing prescription information and kidney disease diagnosis information. Therefore, the processor of the diagnostic device can obtain guidance information from the database using existing prescription information and kidney disease diagnosis information, or can input existing prescription information (or existing prescription information and kidney disease diagnosis information) into the guidance information model to obtain guidance information.

[0477] 2.1.7. Predicting the rate of kidney disease progression using kidney disease diagnosis information

[0478] FIG27 is a diagram illustrating a method for predicting the rate of kidney disease progression using kidney disease diagnosis information according to an embodiment.

[0479] Referring to FIG27, the method for predicting the rate of kidney disease progression according to an embodiment may include: step S810 of obtaining the result value corresponding to a biomarker, step S820 of obtaining kidney disease diagnosis information, and step S830 of determining the rate of kidney disease progression using the result value corresponding to the biomarker and the kidney disease diagnosis information.

[0480] In step S810, the processor of the diagnostic device can obtain the result value corresponding to the biomarker. For example, the processor of the diagnostic device can obtain scores and / or classifications of biomarkers from external devices or the input module of the diagnostic device. Here, biomarkers may include the aforementioned eGFR values ​​(and / or corresponding classifications), albuminuria values ​​(and / or corresponding classifications), or cystatin C values ​​(and / or corresponding classifications), etc. In addition, the biomarkers may include other kidney disease biomarkers or kidney disease risk assessment tools.

[0481] Furthermore, in step S820, the processor of the diagnostic device can acquire kidney disease diagnostic information. Here, kidney disease diagnostic information may refer to kidney disease diagnostic information based on retinal images of the same subject as the subject whose biomarker corresponding result value was acquired in step S810. Since the above description can be applied to step S820, detailed description is omitted.

[0482] Furthermore, in step S830, the rate of kidney disease progression can be determined using the result value corresponding to the biomarker and the kidney disease diagnostic information.

[0483] Specifically, even if the result value corresponding to the biomarker is the same, the rate of kidney disease progression may be different. For example, even if the eGFR value is 50 mL / min / 1.73m2, belonging to the intermediate-high risk group, the score of the kidney disease diagnostic information may be different. However, even if the eGFR value is the same, the rate of kidney disease progression in people with a high kidney disease diagnostic information score may be faster than that in people with a low kidney disease diagnostic information score.

[0484] This will be explained in detail below with reference to Example 2.

[0485] 2.1.7.1. Example 2

[0486] Example 2 is to predict the rate of progression of kidney disease using KDIGO classification and kidney disease diagnostic information.

[0487] In Example 2, data from 5,346 diabetic patients from two tertiary hospitals in South Korea can be used. This may include data from people with eGFR <90 ml / min / 1.73m2 or those with albuminuria. Furthermore, data from people lacking retinal images, serum creatinine, or albuminuria information can be excluded.

[0488] Figure 28 is a diagram illustrating the clinical characteristics of the subjects according to Example 2.

[0489] Referring to Figure 28, KDIGO classification can be divided into three groups (low-risk group, intermediate-risk group, and high-risk group) based on eGFR value (or classification) and albuminuria value (or classification). In the table of Figure 28, the risk of kidney disease can be divided into low-risk group (A1, less than 30 mg / g), intermediate-risk group (A2, 30-300 mg / g), and high-risk group (A3, 300 mg / g or above) according to albuminuria value. Furthermore, according to eGFR value, the risk of kidney disease can be divided into low-risk group (G1, 90 mL / min / 1.73m2 or above), intermediate-risk group (G2, 60-89 mL / min / 1.73m2), intermediate-high-risk group (G3a, 45-59 mL / min / 1.73m2), very high-risk group (G3b, 30-44 mL / min / 1.73m2), and highest-risk group (G4, 29 mL / min / 1.73m2 or below).

[0490] In the KIDGO classification, the low-risk group includes 3...135 individuals were included, comprising a low-risk group based on albuminuria values, a low-risk group based on eGFR values, and a portion of the intermediate-risk group; the intermediate-risk group comprised 1,814 individuals, comprising low-risk and intermediate-risk groups based on albuminuria values, low-risk groups based on eGFR values, intermediate-risk groups, and a portion of the intermediate-high-risk group; the high-risk group comprised 397 individuals, comprising all risk groups based on albuminuria values ​​and a portion of all risk groups based on eGFR values.

[0491] Furthermore, the mean age of the subjects was 62.4 (+ / - 11.4) years, and 60.6% were male. The mean eGFR of the subjects was 86.6 (+ / - 15.3 mL / min per 1.73 m2), and 46.9% had albuminuria. During the 5.0-year (interquartile range: 2.5–7.8) follow-up period, 1,379 (25.8%) subjects experienced kidney disease-related events.

[0492] Figures 29a and 29b are graphs illustrating the incidence of kidney disease-related events based on KDIGO classification and kidney disease diagnostic information according to Example 2.

[0493] Referring to Figures 29a and 29b, in the graphs of Figures 29a and 29b, the x-axis represents time (years), and the y-axis represents the incidence of chronic kidney disease-related events.

[0494] The graph in Figure 29a shows the incidence of kidney disease-related events according to KDIGO classification (low-risk group a1, intermediate-risk group a2, high-risk group a3), with the incidence of kidney-related events likely being higher in the high-risk group a3.

[0495] The graph in Figure 29b shows the incidence of kidney disease-related events when the scores corresponding to kidney disease diagnostic information are reflected in the KDIGO classification.

[0496] The first group b1 is the group of subjects whose scores corresponding to the kidney disease diagnosis information in the KDIGO low-risk group are less than 20; the second group b2 is the group of subjects whose scores corresponding to the kidney disease diagnosis information in the KDIGO low-risk group are greater than or equal to 20; the third group b3 is the group of subjects whose scores corresponding to the kidney disease diagnosis information in the KDIGO intermediate-risk group are less than 20; the fourth group b4 is the group of subjects whose scores corresponding to the kidney disease diagnosis information in the KDIGO intermediate-risk group are greater than or equal to 20; the fifth group b1 can represent the KDIGO high-risk group.

[0497] Specifically, the incidence of kidney disease-related events in the low-risk group a1 in the graph of Figure 29a can be distinguished or stratified in the same way as the incidence of kidney disease-related events in the first group b1 and the second group b2 in (b). Here, in (b), the incidence of kidney disease-related events in group 1 (b1) is lower than that in low-risk group a1 in Figure 29a, and the incidence of kidney disease-related events in group 2 (b2) in Figure 29b may be higher than that in low-risk group a1 in Figure 29a.The incidence of kidney disease-related events in group a1. That is, even among subjects belonging to the same low-risk group in the KDIGO classification, the incidence of kidney disease-related events may be differentiated or stratified based on the scores of kidney disease diagnostic information. In other words, even among subjects belonging to the same low-risk group in the KDIGO classification, those with higher scores in kidney disease diagnostic information may have a higher probability of developing kidney disease-related events than those with lower scores. This may mean that even among subjects belonging to the same low-risk group in the KDIGO classification, those with higher scores in kidney disease diagnostic information may experience faster progression of kidney disease than those with lower scores.

[0498] Furthermore, the incidence of kidney disease-related events in the intermediate-risk group a2 in the graph of Figure 29a can be differentiated or stratified in the same way as the incidence of kidney disease-related events in groups b3 and b4 in the graph of Figure 29b. Here, the incidence of kidney disease-related events in group b3 of the third group in Figure 29b is lower than that in group a2 of the intermediate risk group in Figure 29a, and the incidence of kidney disease-related events in group b4 of the fourth group in Figure 29b may be higher than that in group a2 of the intermediate risk group in Figure 29a. Based on these results, as mentioned above, even among subjects belonging to the intermediate risk group of the KDIGO classification, those with higher kidney disease diagnostic information scores may experience faster kidney disease progression than those with lower scores.

[0499] Furthermore, as shown in the figure of Figure 29a, if only the results corresponding to existing biomarkers are considered, the incidence of kidney disease-related events in group a1 of the low risk group in the KDIGO classification may be higher than that in group a2 of the intermediate risk group in the KDIGO classification.

[0500] However, as shown in the graph of Figure 29b, the incidence of kidney disease-related events in Group 2b2, subjects with a score greater than or equal to 20 for kidney disease diagnostic information in the low-risk KDIGO classification group, may be higher than the incidence of kidney disease-related events in Group 3b3, subjects with a score less than 20 for kidney disease diagnostic information in the intermediate-risk KDIGO classification group. Therefore, even among subjects with the same low KDIGO classification, a higher score for kidney disease diagnostic information compared to subjects with a high KDIGO classification may indicate a higher risk of kidney disease and a faster progression of kidney disease.

[0501] Furthermore, the C-statistic when reflecting the score for kidney disease diagnostic information to the KDIGO classification is 0.04 higher (95% confidence interval (CI): 0.02–0.04) than the C-statistic when using only the KDIGO classification. This means that reflecting the score for kidney disease diagnostic information to the KDIGO classification can predict the risk of kidney disease more accurately than using only the KDIGO classification.

[0502] Referring again to Figure 27, in one embodiment, the processor of the diagnostic device can determine whether the score of the kidney disease diagnostic information is greater than or equal to a predetermined threshold.

[0503] Furthermore, if the score of the kidney disease diagnostic information is greater than or equal to the predetermined threshold, the processor of the diagnostic device can determine that the kidney disease is progressing rapidly; if the score of the kidney disease diagnostic information is less than the predetermined threshold, it can determine that the kidney disease is progressing slowly.

[0504] For example, as shown in Embodiment 2, even if the first subject and the second subject have the same KDIGO classification, eGFR value, albuminuria value, and cystatin C value, if the kidney disease diagnostic information score of the first subject is greater than or equal to the predetermined threshold (e.g., kidney disease diagnostic information score 20), while the kidney disease diagnostic information score of the second subject is less than the predetermined threshold, the processor of the diagnostic device can determine that the kidney disease of the first subject is progressing faster than that of the second subject.

[0505] In addition, in one embodiment, in step S830, the processor of the diagnostic device can predict acute kidney injury (AKI) after cardiac surgery. Specifically, after cardiac surgery, kidney injury may or may not occur depending on the patient. For example, after cardiac surgery, 5% of patients develop chronic kidney disease, and 1-2% of patients experience complete kidney function impairment, potentially requiring dialysis or kidney transplantation.

[0506] While the aforementioned existing biomarkers can be used to predict acute kidney injury after cardiac surgery, their accuracy may be limited. Since the kidney disease diagnostic information in this specification can predict the risk of kidney injury more accurately than existing biomarkers, it can also predict the risk of acute kidney injury after cardiac surgery that existing biomarkers cannot predict.

[0507] For example, the processor of the diagnostic device can determine whether the score of the kidney disease diagnostic information is greater than or equal to a predetermined threshold. The predetermined threshold may be the same as or different from the threshold used to predict the rate of progression of kidney disease. Furthermore, if the score of the kidney disease diagnostic information is greater than or equal to the predetermined threshold, the processor of the diagnostic device can determine that the likelihood of acute kidney injury after cardiac surgery is high; if the score of the kidney disease diagnostic information is less than the predetermined threshold, the likelihood of acute kidney injury after cardiac surgery is low.

[0508] Furthermore, the processor of the diagnostic device can provide information about the rate of progression of kidney disease (or information about acute kidney injury after cardiac surgery) and / or corresponding guidance information. In this case, because the rate of progression of kidney disease (or the likelihood of acute kidney injury after cardiac surgery) differs between the first and second subjects, the processor of the diagnostic device can provide different guidance information to the first and second subjects.

[0509] For example, even if the first and second subjects belong to the same risk group based on existing biomarkers and / or kidney disease diagnostic information, different guidance information can be provided based on the rate of kidney disease progression in the first and second subjects. For example, if the rate of kidney disease progression in the first subject is predicted to be rapid, and the first subject belongs to a low-risk group, the processor of the diagnostic device can provide guidance information aimed at strengthening the management of blood pressure, diabetes, etc. For example, the processor of the diagnostic device can provide treatment information including additional examination information (e.g., instructions for additional examinations, dates of additional examinations, information on hospitals / medical personnel where additional examinations can be performed, etc.) and / or management information including recommended lifestyle modification goals, recommended dietary information, etc., as guidance information.

[0510] Furthermore, if the rate of kidney disease progression in the first subject is predicted to be rapid, and the first subject belongs to a high-risk group, the processor of the diagnostic device can increase the prescription dose for the first subject as prescription information.

[0511] Furthermore, the processor of the diagnostic device can communicate with the aforementioned monitoring device. The processor of the diagnostic device can provide guidance information determined based on the rate of kidney disease progression to the monitoring device. For example, if the processor of the diagnostic device determines that the kidney disease of the first subject in the low-risk group of kidney disease diagnosis information (without symptoms and kidney risk factors) is progressing rapidly, it can determine the corresponding treatment information (e.g., additional examination information, etc.) and / or management information (e.g., controlling food intake, target exercise, etc.) as guidance information and provide the guidance information to the monitoring device. The monitoring device can provide the guidance information obtained from the diagnostic device to the subject and determine whether the monitored information matches the guidance information. If the determination result is a match, the monitoring device can provide information to inform the subject that the guidance information has been well followed; if the result is a mismatch, the monitoring device can provide a warning to the subject, requiring them to follow the guidance information.

[0512] In addition, the processor of the diagnostic device can provide guidance information based on the possibility of acute kidney injury after cardiac surgery. For example, if it is determined that the possibility of acute kidney injury after cardiac surgery is high, the processor of the diagnostic device can provide recommended guidance information to reduce the possibility of acute kidney injury after cardiac surgery. Instruction manual, pages 41 / 43, CN 121586537 A

[0513] Furthermore, the processor of the diagnostic device can provide guidance information using a predetermined database and / or guidance information model, as described in 2.1.5. In this case, the database can store guidance information matching kidney disease diagnostic information and the rate of kidney disease progression (or information on acute kidney injury after cardiac surgery), and the guidance information model can also provide guidance information based on the kidney disease diagnostic information and the rate of kidney disease progression (or the likelihood of acute kidney injury after cardiac surgery).The processor of the diagnostic device can learn from the database and / or guidance information model using kidney disease diagnostic information and the rate of kidney disease progression (or the likelihood of acute kidney injury after cardiac surgery).

[0514] For example, if the eGFR value is in the low-risk group and the score of the kidney disease diagnostic information is greater than or equal to a predetermined threshold, the processor of the diagnostic device can determine that the subject has a low kidney risk, but the rate of kidney disease progression is fast compared to other low-risk groups. Therefore, the processor of the diagnostic device can provide information on the rapid rate of kidney disease progression and / or corresponding guidance information (e.g., guidance information provided when the kidney disease diagnostic information is determined to be of intermediate risk).

[0515] Furthermore, if the eGFR value is in the low-risk group and the score of the kidney disease diagnostic information is less than a predetermined threshold, the processor of the diagnostic device can determine that the subject also has a low kidney risk, but the rate of kidney disease progression is slow compared to other low-risk groups. Therefore, the processor of the diagnostic device can provide information on the slow rate of kidney disease progression and / or corresponding guidance information (e.g., guidance information provided when the kidney disease diagnostic information is determined to be of low risk).

[0516] Furthermore, according to embodiments, different thresholds can be applied to kidney disease diagnostic information based on the result values ​​corresponding to biomarkers. For example, different thresholds can be set in the diagnostic device when the KDIGO classification is low-risk, intermediate-risk, or high-risk.

[0517] Furthermore, according to embodiments, the thresholds can be set based on various benchmarks. For example, the thresholds can be set such that the incidence of kidney disease events is clearly stratified when kidney disease diagnostic information is applied to the result values ​​corresponding to biomarkers.

[0518] Various embodiments of this specification can be implemented as software, including instructions recorded on a machine-readable storage media (e.g., computer). A machine is an apparatus that can invoke stored instructions from a storage medium and operate according to the invoked instructions, and may include an electronic device according to the disclosed embodiments. When the instructions are executed by a processor, the processor may perform the function corresponding to the instructions directly or under the control of the processor using other constituent elements. The instructions may include code generated or executed by a compiler or interpreter. The machine-readable storage medium may be provided in the form of a non-transitory storage medium. Here, "non-transitory storage medium" means that it does not include signals and is tangible, but does not distinguish whether data is permanently or temporarily stored in the storage medium. For example, "non-transitory storage medium" may include buffers for temporarily storing data.

[0519] According to one embodiment, the methods of various embodiments disclosed in this specification may be included in a computer program product.Provided in (Computer Program Product). The computer program product can be traded as a commodity between sellers and buyers. The computer program product can be in the form of a machine-readable storage medium (e.g., read-only optical disc, 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 (e.g., a downloadable application) can be at least temporarily stored or temporarily generated in a storage medium such as the memory of the manufacturer's server, the app store's server, or a relay server.

[0520] As described above, although embodiments have been described by way of limited embodiments and drawings, those skilled in the art can make various modifications and variations from the above description. For example, even if the described techniques are performed in a different order than the described methods, and / or the constituent elements of the described systems, structures, devices, circuits, etc. are combined or combined in a different form than the described methods, or are replaced or substituted by other constituent elements or equivalents, suitable results can still be achieved.

[0521] Therefore, other implementations, other embodiments and things equivalent to the scope of the claims also fall within the scope of the claims. Instruction Manual Page 43 / 43 46 CN 121586537 A Figure 1 Figure 2 Figure 3 Instruction Manual Appendix 1 / 18 Page 47 CN 121586537 A Figure 4 Figure 5 Instruction Manual Appendix 2 / 18 Page 48 CN 121586537 A Figure 6 Figure 7 Instruction Manual Appendix 3 / 18 Page 49 CN 121586537 A Figure 8 Figure 9 Instruction Manual Appendix 4 / 18 Page 50 CN 121586537 A Figure 10 Figure 11 Figure 12 Instruction Manual Appendix 5 / 18 Page 51 CN 121586537 A Figure 13 Figure 14 Instruction Manual Appendix 6 / 18 Page 52 CN 121586537 A Figure 15 Figure 16 Instruction Manual Appendix 7 / 18 Page 53 CN 121586537 A Figure 17 Figure 18 Instruction Manual Appendix 8 / 18 Page 54 CN 121586537 A Figure 19 Figure 20: Appendix to the Instruction Manual, page 9 / 18, 55 CN 121586537 A Figure 21: Appendix to the Instruction Manual, page 10 / 18, 56 CN 121586537 A Figure 22a: Appendix to the Instruction Manual, page 11 / 18, 57 CN 121586537 A Figure 22b: Appendix to the Instruction Manual, page 12 / 1858 CN 121586537 A Figure 23 Appendix to the Instruction Manual, page 13 / 18 59 CN 121586537 A Figure 24 Figure 25 Appendix to the Instruction Manual, page 14 / 18 60 CN 121586537 A Figure 26 Figure 27 Appendix to the Instruction Manual, page 15 / 18 61 CN 121586537 A Figure 28 Appendix to the Instruction Manual, page 16 / 18 62 CN 121586537 A Figure 29a Appendix to the Instruction Manual, page 17 / 18 63 CN 121586537 A Figure 29b Appendix to the Instruction Manual, page 18 / 18 64 CN 121586537 A

Claims

1. A control method for a diagnostic device, characterized in that, include: Steps for acquiring retinal images of a subject; as well as The step of obtaining diagnostic information about kidney disease in the subject using a machine learning model based on the retinal image. The machine learning model includes a first model and a second model, wherein the first model is a neural network model and the second model is a regression-based machine learning model.

2. The control method for the diagnostic device according to claim 1, characterized in that, The first model is trained based on first training data and corresponding results from a first biomarker. The first training data includes multiple retinal images. In the step of obtaining diagnostic information about kidney disease in the subject, The retinal image is input into the first model, and a first result value is obtained from the first model. The first score and the physical information of the subject are input into the second model, and the second result value is obtained from the second model. The kidney disease diagnostic information is obtained based on the second result value.

3. The control method for the diagnostic device according to claim 2, characterized in that, The probability that the result of the first biomarker is equal to or less than the first value of eGFR (Estimated Glomerular Filtration Rate), or information about the presence of albuminuria in the urine of the subject, is at least one of the following: The first result value represents at least one of the following: the probability that the eGFR value of the subject is equal to or less than the first value, or the probability that albuminuria is present in the urine of the subject.

4. The control method for the diagnostic device according to claim 3, characterized in that, The second machine learning model is trained based on second training data, which includes kidney disease event tracking results of objects in a predetermined population.

5. The control method for the diagnostic device according to claim 4, characterized in that, In the predetermined population of objects, data concerning objects whose eGFR values ​​are less than a second value are excluded from the second training data, where the second value is higher than the first value.

6. The control method for the diagnostic device according to claim 1, characterized in that, The first result value includes information about the subject's current risk of kidney disease. The second result value includes information about the probability of future kidney disease-related events occurring in the subject.

7. The control method for the diagnostic device according to claim 6, characterized in that, The second result value includes information about the probability of the subject experiencing kidney disease-related events within 5 years.

8. The control method for the diagnostic device according to claim 1, characterized in that, In the step of obtaining diagnostic information about kidney disease in the subject, A predetermined cutoff value is applied to the second result value as the diagnostic information for the kidney disease, and any grade corresponding to the subject is obtained from multiple grades.

9. The control method for the diagnostic device according to claim 8, characterized in that, The cutoff value is set based on the population distribution in multiple hierarchical groups divided according to the tracking and observation results of a predetermined population.

10. The control method for the diagnostic device according to claim 1, characterized in that, It also includes the step of providing guidance information to the subject based on the kidney disease diagnosis information.

11. The control method for the diagnostic device according to claim 10, characterized in that, In the step of providing guidance information to the subject based on the kidney disease diagnostic information Using a pre-stored database or machine learning model, guidance information corresponding to the diagnostic information of the kidney disease can be provided.

12. The control method for the diagnostic device according to claim 10, characterized in that, It also includes the step of determining the rate of progression of the subject's kidney disease based on the kidney disease diagnostic information. In the step of determining the rate of progression of kidney disease in the subject based on the aforementioned kidney disease diagnostic information... Obtain the biomarker results of the subject. The rate of kidney disease progression in the subject is determined by comparing the results of the biomarkers with the second result value.

13. The control method for the diagnostic device according to claim 12, characterized in that, When the second result value is greater than or equal to the predetermined value, it is determined that the rate of progression of kidney disease is faster than the rate of progression of kidney disease expected based on the result value of the biomarker. When the second result value is less than a predetermined value, it is determined that the progression of kidney disease is slower than the rate of kidney disease progression expected based on the result value of the biomarker.

14. The control method for the diagnostic device according to claim 12, characterized in that, In the step of providing guidance information to the subject based on the kidney disease diagnostic information The subject is given guidance information based on the rate of progression of kidney disease and the diagnostic information of the kidney disease.

15. A recording medium having a program for performing the method according to any one of claims 1 to 14.

16. A diagnostic device, characterized in that, include: Storage module; as well as At least one processor, The at least one processor is used for: Acquire retinal images of the subject; Based on the retinal image, the machine learning model stored in the storage module is used to obtain diagnostic information about the kidney disease of the subject. The machine learning model includes a first model and a second model. The first model is a neural network model. The second model is a regression-based machine learning model.

Images