An automatic analysis and lesion warning method for fundus images of patients with endocrine diseases

Abstract

The present invention relates to the field of fundus image recognition of patients with endocrine diseases, and specifically to an automatic analysis and lesion warning method for fundus images of patients with endocrine diseases. The present invention obtains all suspected disease areas; obtains the performance indicators of microaneurysms according to the shape, grayscale distribution and position distribution of the area; obtains the performance indicators of small area bleeding in combination with the distance difference between adjacent areas; obtains the performance indicators of hard exudation in combination with the grayscale distribution characteristics in adjacent areas; obtains the comprehensive intensity index of each suspected disease based on the intensity characteristics of the performance indicators of all suspected diseases; obtains the true intensity of the disease based on the case data differences between different patients and the comprehensive intensity index; and then constructs a prediction model; and automatically analyzes the fundus images of patients according to the prediction model. The present invention can construct an accurate prediction model, increase the accuracy of image diagnosis results, and enable medical personnel to grasp the best time to intervene in patients.

Description

Technical Field

[0001] The present invention relates to the field of fundus image recognition of patients with endocrine diseases, and in particular to an automatic analysis and lesion early warning method for fundus images of patients with endocrine diseases. Background Art

[0002] Endocrine diseases mainly refer to diseases caused by abnormal secretion functions of endocrine glands or endocrine tissues in the patient's body, such as diabetes, thyroid disease, adrenal disease, etc. Among them, diabetes is the most common disease that is most likely to cause retinopathy. Retinopathy caused by diabetes is a microvascular disease, which is mainly caused by retinal ischemia, hypoxia, and microcirculation disorders, and then induces retinal microvascular proliferation, exudation, edema, hemorrhage, macular degeneration, and even retinal detachment, eventually leading to blindness. Diabetic retinopathy will not have obvious symptoms in the early stages, and the patient himself cannot directly perceive it. By the time it is discovered, the condition has often worsened. Therefore, it is extremely important to conduct regular fundus examinations and retinopathy warnings for diabetic patients.

[0003] In the prior art, fundus images of patients with endocrine diseases are often used to conveniently obtain information about the fundus of the patients. However, the analysis of the information contained in the fundus images usually relies on the subjective experience of relevant personnel. In addition, fundus endocrine diseases are highly diverse and hidden in the early stages. The reliability of the diagnostic results of fundus images by traditional diagnostic methods is low, and in severe cases, the patients may miss the best time for intervention. Summary of the invention

[0004] In order to solve the technical problems that the analysis of the information contained in the fundus images usually depends on the subjective experience of relevant personnel, and the fundus secretory diseases are highly diverse and hidden in the early stages, the reliability of the diagnostic results of the fundus images by the traditional diagnostic methods is low, and in severe cases, the patients may miss the best time for intervention, the purpose of the present invention is to provide an automatic analysis and lesion warning method for fundus images of patients with endocrine diseases, and the technical scheme adopted is specifically as follows: an automatic analysis and lesion warning method for fundus images of patients with endocrine diseases, the method comprising: obtaining fundus images of each patient with endocrine diseases; using image segmentation to screen out all non-vascular areas in each fundus image; obtaining all suspected disease areas in each fundus image according to the grayscale distribution in all the non-vascular areas; dividing the suspected diseases into microaneurysms, small areas of hemorrhage and hard exudates; selecting any suspected disease area as the target area; obtaining the performance index of the microaneurysm in the target area according to the shape characteristics, grayscale distribution characteristics and position distribution of the target area. ; According to the distance difference between the target area and the adjacent suspected disease area, the performance index of the microaneurysm in each suspected disease area and the grayscale distribution characteristics, the performance index of the small area of ​​bleeding in the target area is obtained; According to the performance index of the microaneurysm in the target area and the adjacent suspected disease area, the performance index of the small area of ​​bleeding and the grayscale distribution characteristics, the performance index of the hard exudation in the target area is obtained; According to the intensity characteristics of the performance index of the suspected disease in all suspected disease areas, the comprehensive intensity index of each suspected disease is obtained; Any endocrine disease patient is selected as the target patient; The case data of the target patient and other endocrine disease patients are obtained; According to the difference in case data between the target patient and other endocrine disease patients and the comprehensive intensity index of each suspected disease, the true intensity of each suspected disease of the target patient is obtained; According to the difference in case data between the target patient and other endocrine disease patients and the comprehensive intensity index of each suspected disease, a prediction model is constructed according to the comprehensive intensity index and the true intensity of each suspected disease of each endocrine patient; The fundus image of each endocrine disease patient is automatically analyzed according to the prediction model.

[0005] Furthermore, the method for obtaining the suspected disease area includes: using an image segmentation algorithm to obtain the vascular area in each fundus image; screening out all vascular areas and treating other areas in the fundus image as non-vascular areas; using the interquartile range method for all pixels in the non-vascular areas to obtain all suspected disease pixels; randomly selecting a suspected disease pixel in the non-vascular area as a pixel to be analyzed; marking all suspected disease pixels within a preset neighborhood of the pixel to be analyzed, and forming the suspected disease area to which the pixel to be analyzed belongs; traversing each suspected disease pixel to obtain all suspected disease areas.

[0006] Furthermore, the method for obtaining the performance index of the microaneurysm includes: obtaining the performance index of the microaneurysm according to a performance index calculation formula of the microaneurysm, and the performance index calculation formula of the microaneurysm is as follows: In the formula, performance indicators indicating microaneurysms in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the length of the minimum enclosing rectangle of the target area; Indicates the width of the minimum bounding rectangle of the target area; Represents the distance between the centroid pixel in the target area and the nearest pixel in the blood vessel area; Represents the mean gray value of pixels in the target area; Represents the gray value of the pixel in the blood vessel region that is closest to the centroid pixel in the target region; Represents an exponential function with a natural constant as base.

[0007] Furthermore, the method for obtaining the performance index of small area bleeding includes: obtaining the performance index of small area bleeding according to a performance index calculation formula of small area bleeding, and the performance index formula of small area bleeding is as follows: In the formula, Performance indicators that indicate small areas of bleeding in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the number of suspected disease areas adjacent to the target area; Indicates the first Performance indicators of microaneurysms in suspected disease areas; Indicates The distance between the centroid pixel of the adjacent suspected disease area and the centroid pixel of the target area; Indicates The average gray value of the pixels in adjacent suspected disease areas; Represents the mean gray value of pixels in the target area; Represents an exponential function with a natural constant as base.

[0008] Furthermore, the method for obtaining the performance index of hard exudation includes: obtaining the performance index of hard exudation according to a performance index calculation formula of hard exudation, and the performance index calculation formula of hard exudation is as follows: In the formula, A performance indicator indicating hard exudate in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the number of suspected disease areas adjacent to the target area; Indicates The distance between the centroid pixel of the adjacent suspected disease area and the centroid pixel of the target area; Indicates the first Performance indicators of microaneurysms in suspected disease areas; Indicates the first performance indicators of small areas of bleeding in suspected disease areas; Represents the mean gray value of pixels in the target area; represents the maximum value function; Represents an exponential function with a natural constant as base.

[0009] Furthermore, the method for obtaining the comprehensive strength index includes: when there is only one performance index of a suspected disease in the suspected disease area that is greater than a preset threshold, the suspected disease area is used as the suspected disease area to which the corresponding suspected disease belongs; when there are no less than two performance indexes in the suspected disease area that are greater than the preset threshold, the maximum value between the performance indexes is calculated, and the suspected disease area is used as the suspected disease area to which the suspected disease corresponding to the maximum performance index belongs; the comprehensive strength index is obtained according to the comprehensive strength index calculation formula, and the comprehensive strength index calculation formula is as follows: In the formula, Indicates the serial number of suspected symptoms, where 1 indicates microaneurysm, 2 indicates small area of ​​hemorrhage, and 3 indicates hard exudate; Indicates A comprehensive intensity index of the suspected symptoms; Indicates The number of suspected disease areas to which the suspected disease belongs; Indicates performance indicators of suspected symptoms; Indicates The first Area of ​​suspected disease area; Represents an exponential function with a natural constant as base.

[0010] Furthermore, the method for obtaining the true intensity of each suspected disease includes: obtaining all case data of each endocrine disease patient in each suspected disease according to a database; selecting any suspected disease as a target disease, converting the case data of the target disease of each endocrine disease patient into a feature vector, calculating the cosine similarity between the feature vectors of the target disease of the target patient and each other endocrine disease patient, and forming a comprehensive intensity index group with the comprehensive intensity indicators of the target disease of a preset number of other endocrine disease patients with the highest cosine similarity who do not suffer from the target disease; performing negative correlation mapping on the ratio between the mean of the comprehensive intensity indicators in the comprehensive intensity index group and the mean of the comprehensive intensity indicators of the target disease of all endocrine disease patients suffering from the target disease to obtain a authenticity coefficient; and taking the product of the comprehensive intensity index of each suspected disease and the authenticity coefficient as the true intensity of each suspected disease of the target patient.

[0011] An automatic analysis and lesion warning system for fundus images of patients with endocrine diseases, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0012] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for automatic analysis of fundus images of patients with endocrine diseases and lesion warning.

[0013] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases are implemented.

[0014] Beneficial effects of the technical solution of the present invention: In order to analyze fundus information, the present invention obtains fundus images of each endocrine disease patient; since the grayscale of the fundus vascular area in the fundus image is obvious and easily confused with the suspected disease area, the vascular area of ​​the fundus image is first screened out to obtain all suspected disease areas; since the embodiment of the present invention mainly conducts a specific analysis on whether the endocrine disease patient suffers from diabetic retinopathy, the suspected diseases are divided into microaneurysms, small-area hemorrhages and hard exudates; since microaneurysms usually appear as small and uniformly colored red circular spots in fundus images, mostly distributed on both sides of the blood vessels, the performance indicators of microaneurysms are obtained according to the shape characteristics, grayscale distribution characteristics and position distribution of the target area; since microaneurysm rupture is the main cause of intraretinal hemorrhage, and the grayscale distribution inside the small-area hemorrhage area is uneven, but the overall grayscale is deeper than the microaneurysm, the performance indicators of the microaneurysms in each suspected disease area and the grayscale distribution characteristics are used to obtain the performance of small-area hemorrhage. Indicators; Since hard exudation usually appears around microaneurysms and small-area bleeding areas, showing yellow-white uneven spots and plaques, and the boundaries are relatively clear, the performance indicators of hard exudation are obtained according to the performance indicators of microaneurysms in the target area and the adjacent suspected disease area, the performance indicators of small-area bleeding, and the grayscale distribution characteristics; In obtaining the performance indicators of each disease in each suspected disease area, it is necessary to judge the specific disease attribution in each suspected disease area, and then analyze the performance intensity of each disease in the fundus area to obtain the comprehensive intensity index of each suspected disease; In order to increase the credibility of the comprehensive intensity index of each suspected disease of an endocrine patient, it is necessary to compare and analyze the disease conditions of the fundus area of ​​other endocrine patients, so according to the difference in case data between the target patient and other endocrine disease patients, the true intensity of the disease of each suspected disease of the target patient is analyzed; Combined with the comprehensive intensity index of each suspected disease of each endocrine patient and the true intensity of the disease, a prediction model is constructed; According to the prediction model, the fundus image of each endocrine disease patient is automatically analyzed. The present invention can construct an accurate prediction model, increase the accuracy of the fundus image diagnosis results, and enable medical personnel to grasp the best intervention time for patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1A flow chart of a method for automatically analyzing fundus images and early warning of lesions of patients with endocrine diseases provided by one embodiment of the present invention; Figure 2 A schematic diagram of a fundus image of a patient with an endocrine disease provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the automatic analysis and lesion warning method of fundus images of patients with endocrine diseases proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0019] The following is a detailed description of a specific scheme of the method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases provided by the present invention in conjunction with the accompanying drawings.

[0020] See also Figure 1 , which shows an automatic analysis and lesion warning method for fundus images of patients with endocrine diseases provided by an embodiment of the present invention. The method comprises: Step S1: Acquire fundus images of each patient with endocrine diseases.

[0021] The embodiment of the present invention is mainly applied to the automatic analysis of fundus information of patients with endocrine diseases. In order to analyze the fundus information, firstly, a fundus image of a patient with endocrine diseases is obtained. Figure 2 As shown, a schematic diagram of a fundus image is provided.

[0022] In one embodiment of the present invention, a fundus camera is used to photograph the fundus to obtain an initial fundus image, which is transmitted to an image analysis system, grayscaled and filtered to remove obvious noise in the image. It should be noted that grayscale processing and filtering are technical means well known to those skilled in the art, and are not limited or elaborated herein.

[0023] Step S2: Use image segmentation to screen out all non-vascular areas in each fundus image; obtain all suspected disease areas in each fundus image according to the grayscale distribution in all non-vascular areas; divide the suspected diseases into microaneurysms, small-area hemorrhages and hard exudates; select any suspected disease area as the target area; obtain the performance index of the microaneurysm in the target area according to the shape characteristics, grayscale distribution characteristics and position distribution of the target area; obtain the performance index of the small-area hemorrhage in the target area according to the distance difference between the target area and the adjacent suspected disease areas, the performance index of the microaneurysm in each suspected disease area and the grayscale distribution characteristics; obtain the performance index of the hard exudate in the target area according to the performance index of the microaneurysm in the target area and the adjacent suspected disease areas, the performance index of the small-area hemorrhage and the grayscale distribution characteristics; obtain the comprehensive intensity index of each suspected disease according to the intensity characteristics of the performance indicators of the suspected diseases in all suspected disease areas.

[0024] In the fundus area, the fundus vascular area is more obvious in the grayscale of the fundus image and is easily confused with the suspected disease area. Therefore, the vascular area of ​​the fundus image is first screened out, and the fundus image that does not contain the vascular area is analyzed for suspected diseases.

[0025] Preferably, in one embodiment of the present invention, the method for obtaining the suspected disease area includes: obtaining the vascular area in each fundus image using an image segmentation algorithm; screening out all vascular areas, and treating other areas in the fundus image as non-vascular areas. In the embodiment of the present invention, the Canny edge detection algorithm is used to extract the vascular area. It should be noted that other edge detection algorithms can also be used for this operation, which is not limited here. The Canny edge detection algorithm is a technical means well known to those skilled in the art and will not be described in detail here.

[0026] The interquartile range method is used for all pixels in the non-vascular area to obtain the upper and lower limits of abnormal grayscale values ​​in the fundus image; all pixels less than the lower limit and greater than the upper limit are regarded as suspected disease pixels.

[0027] In the non-vascular area, any suspected disease pixel is selected as the pixel to be analyzed; all suspected disease pixels within the preset neighborhood of the pixel to be analyzed are marked to form the suspected disease area to which the pixel to be analyzed belongs. In one embodiment of the present invention, the preset neighborhood is set to It should be noted that the preset neighborhood can be set by yourself and is not limited here.

[0028] Each suspected disease pixel is traversed to obtain all suspected disease areas.

[0029] In the embodiment of the present invention, a specific analysis is performed on whether the patient with endocrine disease suffers from diabetic retinopathy. In actual situations, diabetic retinopathy (DP) has clear clinical staging in medicine, among which the older domestic classification divides it into two types and six stages, and the international classification is divided into non-proliferative type (divided into three types of mild, moderate and severe) and proliferative type. Before the first type and three stages of the domestic classification and before the moderate non-proliferative type of the international classification, it belongs to the early stage of the lesion. At this time, the patient himself often does not have much perception. It is the period when the lesion is most easily ignored, but it is also the best period for treatment. During this period, the patient's fundus will not have more obvious symptoms such as new blood vessels and vitreous hemorrhage, but minor symptoms such as microaneurysms, small areas of bleeding and hard exudates. Therefore, in the embodiment of the present invention, the suspected symptoms are divided into microaneurysms, small areas of bleeding and hard exudates, and further analysis is performed.

[0030] Since microaneurysms are a type of retinal capillary hemangioma and are the earliest symptoms of diabetic retinopathy, they can only be discovered through fundus examination. In addition, microaneurysms are caused by local vascular expansion or tumor expansion. In fundus images, they usually appear as small and uniformly colored red circular spots, mostly distributed on both sides of the blood vessels. Therefore, in an embodiment of the present invention, performance indicators of microaneurysms in the target area are obtained based on the shape characteristics, grayscale distribution characteristics, and position distribution of the target area.

[0031] Preferably, in one embodiment of the present invention, the method for obtaining the performance index of a microaneurysm includes: obtaining the performance index of a microaneurysm according to a performance index calculation formula of a microaneurysm, and the performance index calculation formula of a microaneurysm is as follows: In the formula, performance indicators indicating microaneurysms in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the length of the minimum enclosing rectangle of the target area; Indicates the width of the minimum bounding rectangle of the target area; Represents the distance between the centroid pixel in the target area and the nearest pixel in the blood vessel area; Represents the mean gray value of pixels in the target area; Represents the gray value of the pixel in the blood vessel region that is closest to the centroid pixel in the target region; Represents an exponential function with a natural constant as base.

[0032] In the calculation formula of the performance index of microaneurysm, the minimum enclosing rectangle of the target area can be directly obtained by the existing technology, wherein, if the ratio of the length to the width of the minimum enclosing rectangle is The smaller the standard deviation, the closer the contour of the target area is to a circle, that is, the closer it is to the shape characteristics of a microaneurysm. The smaller the standard deviation of the grayscale value of the pixel point in the target area, the more uniform the grayscale distribution in the target area, and the more likely the target area is to belong to a microaneurysm area. The distance between the centroid pixel point in the target area and the pixel point in the nearest blood vessel area The smaller the value, the closer the target area is to the blood vessel area, and the more likely the target area is a microaneurysm area. The more similar the grayscale mean value of the target area is to the grayscale value of the blood vessel area, the more likely the target area is to be a microaneurysm area. The smaller, The larger it is, the greater the performance index of the microaneurysm in the target area.

[0033] There will also be small areas of bleeding in the fundus area, which are caused by ruptured blood vessels, usually located at the venous end of the capillaries. Since microaneurysm rupture is the main cause of intraretinal hemorrhage, small areas of bleeding are often closely related to the location of microaneurysms. And because small areas of bleeding are formed by blood overflow and diffusion, the bleeding area mostly appears as irregular blocks, and in the fundus image, the grayscale distribution inside the small area of ​​bleeding is uneven, but the overall grayscale is deeper than the microaneurysm. Therefore, in the embodiment of the present invention, according to the distance difference between the target area and the adjacent suspected disease area, the performance indicators of the microaneurysm in each suspected disease area and the grayscale distribution characteristics are used to obtain the performance indicators of small area bleeding in the target area.

[0034] Preferably, in one embodiment of the present invention, the method for obtaining the performance index of small area bleeding includes: obtaining the performance index of small area bleeding according to a performance index calculation formula of small area bleeding, and the performance index formula of small area bleeding is as follows: In the formula, Performance indicators that indicate small areas of bleeding in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the number of suspected disease areas adjacent to the target area; Indicates the first Performance indicators of microaneurysms in suspected disease areas; Indicates The distance between the centroid pixel of the adjacent suspected disease area and the centroid pixel of the target area; Indicates The average gray value of the pixels in adjacent suspected disease areas; Represents the mean gray value of pixels in the target area; Represents an exponential function with a natural constant as base.

[0035] In the performance indicator formula for small area bleeding, the standard deviation of the grayscale value of pixels in the target area The larger the value is, the more uneven the grayscale distribution in the target area is, and the more likely it is a small area of ​​bleeding. The larger the performance index of the microaneurysm in the target area is, the closer the relationship between the target area and the microaneurysm is. At this time, if the distance between the centroid pixel point of the adjacent suspected disease area and the centroid pixel point of the target area is smaller, it means that the target area is closer to the microaneurysm area and is more likely to be a small area of ​​bleeding. The difference between the overall grayscale of the target area and the overall grayscale of the adjacent suspected disease area The larger it is, the darker the grayscale of the target area is, and the more likely it is a small bleeding area.

[0036] In actual situations, hard exudation is closely related to retinal vascular leakage. The reason is that after retinal microvascular damage, lipid substances and proteins in the blood leak into the retinal tissue to form lipid deposits, which usually appear around microaneurysms and small areas of bleeding, showing yellow-white uneven spots and plaques with clear boundaries. Therefore, in the embodiment of the present invention, the performance indicators of hard exudation in the target area are obtained based on the performance indicators of microaneurysms in the target area and the adjacent suspected disease area, the performance indicators of small-area bleeding, and the grayscale distribution characteristics.

[0037] Preferably, in one embodiment of the present invention, the method for obtaining the performance index of hard exudation includes: obtaining the performance index of hard exudation according to a performance index calculation formula of hard exudation, and the performance index calculation formula of hard exudation is as follows: In the formula, A performance indicator indicating hard exudate in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the number of suspected disease areas adjacent to the target area; Indicates The distance between the centroid pixel of the adjacent suspected disease area and the centroid pixel of the target area; Indicates the first Performance indicators of microaneurysms in suspected disease areas; Indicates the first performance indicators of small areas of bleeding in suspected disease areas; Represents the mean gray value of pixels in the target area; represents the maximum value function; Represents an exponential function with a natural constant as base.

[0038] In the performance index calculation formula of hard exudation, the standard deviation of the gray value of pixels in the target area is The larger the value, the more uneven the grayscale distribution in the target area is, and the more likely it is to be a hard exudation area; the larger the larger value of the performance indicators of microaneurysms and small-area hemorrhages in adjacent suspected disease areas is, the closer the relationship between the target area and the microaneurysms and small-area hemorrhages is. If the distance between the centroid pixel point of the adjacent suspected disease area and the centroid pixel point of the target area is smaller, it means that the target area is closer to the microaneurysm area or the small-area hemorrhage area, and the more likely it is to be a hard exudation area; the larger the grayscale mean value of the target area is, the more likely the target area is to present yellow-white characteristics, and the more likely the target area is to be a hard exudation area.

[0039] In order to obtain the performance indicators of each symptom in each suspected symptom area, it is necessary to determine the specific symptom affiliation in each suspected symptom area, and then analyze the performance intensity of each symptom in the fundus area. Therefore, in an embodiment of the present invention, a comprehensive intensity index of each suspected symptom is obtained based on the intensity characteristics of the performance indicators of the suspected symptoms in all suspected symptom areas.

[0040] Preferably, in one embodiment of the present invention, the method for obtaining the comprehensive strength index includes: when there is only one performance index of a suspected disease in the suspected disease area that is greater than a preset threshold, the suspected disease area is used as the suspected disease area to which the corresponding suspected disease belongs; when there are no less than two performance indexes in the suspected disease area that are greater than the preset threshold, the maximum value between the performance indexes is calculated, and the suspected disease area is used as the suspected disease area to which the suspected disease corresponding to the maximum value of the performance index belongs. In one embodiment of the present invention, the preset threshold is set to 0.75, and the preset threshold can be set by the implementer, and is not limited here.

[0041] The comprehensive strength index is obtained according to the comprehensive strength index calculation formula. The comprehensive strength index calculation formula is as follows: In the formula, Indicates the serial number of suspected symptoms, where 1 indicates microaneurysm, 2 indicates small area of ​​hemorrhage, and 3 indicates hard exudate; Indicates A comprehensive intensity index of the suspected symptoms; Indicates The number of suspected disease areas to which the suspected disease belongs; Indicates performance indicators of suspected symptoms; Indicates The first Area of ​​suspected disease area; Represents an exponential function with a natural constant as base.

[0042] In the calculation formula of comprehensive strength index, The more suspected disease areas a suspected disease belongs to, the more The stronger the suspected symptom is in the fundus area, the The larger the The greater the comprehensive intensity index of the suspected symptom, the greater the The larger the area and performance index of the suspected disease in the fundus area, the The stronger the suspected symptom is in the fundus area, the The larger the The greater the comprehensive intensity index of the suspected symptoms.

[0043] Step S3: Select any endocrine disease patient as the target patient; obtain case data of the target patient and other endocrine disease patients; obtain the true intensity of each suspected symptom of the target patient based on the difference in case data between the target patient and other endocrine disease patients and the comprehensive intensity index of each suspected symptom; construct a prediction model based on the comprehensive intensity index and the true intensity of each suspected symptom of each endocrine patient; and automatically analyze the fundus image of each endocrine disease patient based on the prediction model.

[0044] In order to increase the credibility of the comprehensive intensity index of each suspected symptom of an endocrine patient, it is necessary to compare and analyze the symptom conditions of the fundus area of ​​other endocrine patients. Therefore, in an embodiment of the present invention, the case data of the target patient and other endocrine disease patients are obtained; based on the differences in the case data between the target patient and other endocrine disease patients and the comprehensive intensity index of each suspected symptom, the true intensity of each suspected symptom of the target patient is obtained.

[0045] Preferably, in one embodiment of the present invention, the method for obtaining the true intensity of each suspected disease includes: obtaining all case data of each endocrine disease patient according to the database. In the embodiment of the present invention, the case data of each endocrine disease patient is collected through the hospital's electronic medical record system (or other public channels) to facilitate subsequent analysis and processing. The specific process is a technical means well known to those skilled in the art and will not be elaborated here.

[0046] Select any suspected disease as the target disease, convert the case data of the target disease of each endocrine disease patient into a feature vector, calculate the cosine similarity between the feature vectors of the target disease of the target patient and each other endocrine disease patient, and form a comprehensive strength index group with the comprehensive strength indexes of the target disease of a preset number of other endocrine disease patients with the highest cosine similarity. In one embodiment of the present invention, the preset number is set to 10. It should be noted that the preset number can be set voluntarily and is not limited here.

[0047] The ratio between the mean value of the comprehensive strength index in the comprehensive strength index group and the mean value of the comprehensive strength index of the target disease of all endocrine disease patients with the target disease is negatively correlated to obtain the authenticity coefficient; the product of the comprehensive strength index of each suspected disease and the authenticity coefficient is used as the disease true intensity of each suspected disease of the target patient. In the embodiment of the present invention, the calculation formula of the disease true intensity is as follows: In the formula, The target patient The true intensity of the suspected disease; Indicates A comprehensive intensity index of the suspected symptoms; Indicates the preset number of patients who have not The comprehensive intensity indexes of the suspected diseases constitute the average value of the comprehensive intensity index group; Indicates that all patients with The mean values ​​of the comprehensive intensity indexes of target symptoms in patients with endocrine diseases with suspected symptoms.

[0048] In the calculation formula of the true intensity of the disease, The higher the comprehensive intensity index of the suspected disease, the more likely the target patient is to have the The higher the true intensity of the suspected disease, the higher the true intensity of the disease; if the patient who is most similar to the target patient does not have the first The proportion of comprehensive intensity index of endocrine patients with suspected diseases The smaller the The suspected disease is not strongly expressed in the normal similar patient group of the target patient. The more reliable the comprehensive intensity index of the suspected disease is, the The greater the true intensity of the suspected symptom.

[0049] According to the comprehensive intensity index of each suspected symptom of each endocrine patient and the true intensity of the symptom, a prediction model is constructed. In one embodiment of the present invention, a logistic regression model is provided, which specifically includes: In the formula, represents the probability that the target patient has diabetic retinopathy; represents the true intensity of the microaneurysm in the target patient; The true intensity of the condition representing a small area of ​​bleeding in the target patient; Indicates the true intensity of the condition of hard exudate in the target patient; It represents the comprehensive intensity index of microaneurysms; It represents the comprehensive intensity index of small area bleeding; It indicates the comprehensive strength index of hard exudate; Represents an exponential function with a natural constant as base.

[0050] The fundus images of each endocrine disease patient are automatically analyzed according to the prediction model. In one embodiment of the present invention, if the probability that the target patient suffers from diabetic retinopathy is greater than 0.4, the patient needs to be informed to take preventive measures in advance; if the probability that the target patient suffers from diabetic retinopathy is greater than 0.8, the patient's disease risk is considered to be too high and further examination is required, and appropriate intervention treatment is received according to the situation.

[0051] At this point, the automatic analysis of fundus images and lesion warning of patients with endocrine diseases have been completed.

[0052] In summary, the method obtains fundus images of each patient with endocrine diseases; uses image segmentation to screen out all suspected disease areas in each fundus image; obtains all suspected disease areas in each fundus image according to the grayscale distribution in all suspected disease areas; divides suspected diseases into microaneurysms, small-area hemorrhages and hard exudates; selects any suspected disease area as the target area; obtains the performance index of the microaneurysm in the target area according to the shape characteristics, grayscale distribution characteristics and position distribution of the target area; obtains the performance index of the small-area hemorrhage in the target area according to the distance difference between the target area and the adjacent suspected disease areas, the performance index of the microaneurysm in each suspected disease area and the grayscale distribution characteristics; obtains the performance index of the small-area hemorrhage in the target area according to the microaneurysm in the target area and the adjacent suspected disease areas. The performance indicators of aneurysm, small area hemorrhage and grayscale distribution characteristics are used to obtain the performance indicators of hard exudates in the target area; the comprehensive intensity index of each suspected symptom is obtained according to the intensity characteristics of the performance indicators of the suspected symptom in all suspected symptom areas; any patient with endocrine disease is selected as the target patient; the case data of the target patient and other patients with endocrine diseases are obtained; the true intensity of each suspected symptom of the target patient is obtained according to the difference in case data between the target patient and other patients with endocrine diseases and the comprehensive intensity index of each suspected symptom; a prediction model is constructed according to the comprehensive intensity index and the true intensity of each suspected symptom of each endocrine patient; the fundus images of each endocrine disease patient are automatically analyzed according to the prediction model.

[0053] An embodiment of the present invention provides an automatic analysis and lesion warning system for fundus images of patients with endocrine diseases, the system comprising a memory, a processor and a computer program, wherein the memory is used to store the corresponding computer program, the processor is used to run the corresponding computer program, and when the computer program runs in the processor, it can implement the method described in steps S1-S3.

[0054] The third objective of an embodiment of the present invention is to provide a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the method described in steps S1-S3 when executing the computer program.

[0055] A fourth objective of an embodiment of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in steps S1-S3 can be implemented.

[0056] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0057] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases, characterized in that: The method comprises: obtaining fundus images of each endocrine disease patient; screening out all non-vascular areas in each fundus image by image segmentation; obtaining all suspected disease areas in each fundus image according to the grayscale distribution in all the non-vascular areas; dividing the suspected diseases into microaneurysms, small-area hemorrhages and hard exudates; selecting any suspected disease area as a target area; obtaining the performance index of the microaneurysm in the target area according to the shape characteristics, grayscale distribution characteristics and position distribution of the target area; obtaining the performance index of the small-area hemorrhage in the target area according to the distance difference between the target area and the adjacent suspected disease areas, the performance index of the microaneurysm in each suspected disease area and the grayscale distribution characteristics; obtaining the performance index of the hard exudate in the target area according to the performance index of the microaneurysm in the target area and the adjacent suspected disease areas, the performance index of the small-area hemorrhage and the grayscale distribution characteristics; obtaining the comprehensive intensity index of each suspected disease according to the intensity characteristics of the performance index of the suspected disease in all suspected disease areas; selecting A patient with an endocrine disease is taken as a target patient; case data of the target patient and other patients with endocrine diseases are obtained; the true intensity of each suspected symptom of the target patient is obtained based on the difference in case data between the target patient and other patients with endocrine diseases and the comprehensive intensity index of each suspected symptom; a prediction model is constructed based on the comprehensive intensity index and the true intensity of each suspected symptom of each endocrine patient; the fundus image of each patient with endocrine diseases is automatically analyzed according to the prediction model; the method for obtaining the comprehensive intensity index includes: when there is only one performance index of a suspected symptom in the suspected symptom area that is greater than a preset threshold, the suspected symptom area is taken as the suspected symptom area to which the corresponding suspected symptom belongs; when there are no less than two performance indicators in the suspected symptom area that are greater than the preset threshold, the maximum value between the performance indicators is calculated, and the suspected symptom area is taken as the suspected symptom area to which the suspected symptom corresponding to the maximum performance index belongs; the comprehensive intensity index is obtained according to the comprehensive intensity index calculation formula, and the comprehensive intensity index calculation formula is as follows: In the formula, Indicates the serial number of suspected symptoms, where 1 indicates microaneurysm, 2 indicates small area of ​​hemorrhage, and 3 indicates hard exudate; Indicates A comprehensive intensity index of the suspected symptoms; Indicates The number of suspected disease areas to which the suspected disease belongs; Indicates performance indicators of suspected symptoms; Indicates The first Area of ​​suspected disease area; Represents an exponential function with a natural constant as the base; the method for obtaining the true intensity of each suspected symptom includes: obtaining all case data of each endocrine disease patient in each suspected symptom according to a database; selecting any suspected symptom as a target symptom, converting the case data of the target symptom of each endocrine disease patient into a feature vector, calculating the cosine similarity between the feature vectors of the target symptom of the target patient and each other endocrine disease patient, and forming a comprehensive intensity index group with the comprehensive intensity indicators of the target symptom of a preset number of other endocrine disease patients with the highest cosine similarity who do not suffer from the target symptom; performing negative correlation mapping on the ratio between the mean value of the comprehensive intensity indicators in the comprehensive intensity index group and the mean value of the comprehensive intensity indicators of the target symptom of all endocrine disease patients suffering from the target symptom to obtain a authenticity coefficient; and taking the product of the comprehensive intensity indicator of each suspected symptom and the authenticity coefficient as the true intensity of each suspected symptom of the target patient.

2. The method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases according to claim 1, characterized in that: The method for obtaining the suspected disease area includes: using an image segmentation algorithm to obtain the vascular area in each fundus image; screening out all vascular areas and treating other areas in the fundus image as non-vascular areas; using the interquartile range method for all pixels in the non-vascular areas to obtain all suspected disease pixels; randomly selecting a suspected disease pixel in the non-vascular area as a pixel to be analyzed; marking all suspected disease pixels within a preset neighborhood of the pixel to be analyzed, and forming the suspected disease area to which the pixel to be analyzed belongs; traversing each suspected disease pixel to obtain all suspected disease areas.

3. The method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases according to claim 1, characterized in that: The method for obtaining the performance index of the microaneurysm includes: obtaining the performance index of the microaneurysm according to a performance index calculation formula of the microaneurysm, and the performance index calculation formula of the microaneurysm is as follows: In the formula, performance indicators indicating microaneurysms in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the length of the minimum enclosing rectangle of the target area; Indicates the width of the minimum bounding rectangle of the target area; Represents the distance between the centroid pixel in the target area and the nearest pixel in the blood vessel area; Represents the mean gray value of pixels in the target area; Represents the gray value of the pixel in the blood vessel region that is closest to the centroid pixel in the target region; Represents an exponential function with a natural constant as base.

4. The method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases according to claim 1, characterized in that: The method for obtaining the performance index of small area bleeding includes: obtaining the performance index of small area bleeding according to a performance index calculation formula of small area bleeding, and the performance index formula of small area bleeding is as follows: In the formula, Performance indicators that indicate small areas of bleeding in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the number of suspected disease areas adjacent to the target area; Indicates the first Performance indicators of microaneurysms in suspected disease areas; Indicates The distance between the centroid pixel of the adjacent suspected disease area and the centroid pixel of the target area; Indicates The average gray value of the pixels in adjacent suspected disease areas; Represents the mean gray value of pixels in the target area; Represents an exponential function with a natural constant as base.

5. The method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases according to claim 1, characterized in that: The method for obtaining the performance index of the hard exudation includes: obtaining the performance index of the hard exudation according to a performance index calculation formula of the hard exudation, and the performance index calculation formula of the hard exudation is as follows: In the formula, A performance indicator indicating hard exudate in the target area; Indicates the standard deviation of the grayscale values ​​of pixels within the target area; Indicates the number of suspected disease areas adjacent to the target area; Indicates The distance between the centroid pixel of the adjacent suspected disease area and the centroid pixel of the target area; Indicates the first Performance indicators of microaneurysms in suspected disease areas; Indicates the first performance indicators of small areas of bleeding in suspected disease areas; Represents the mean gray value of pixels in the target area; represents the maximum value function; Represents an exponential function with a natural constant as base.

6. An automatic analysis and lesion warning system for fundus images of patients with endocrine diseases, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases as described in any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for automatic analysis of fundus images and lesion warning of patients with endocrine diseases as described in any one of claims 1 to 5 are implemented.

Images