Eye sign recognition method and equipment for thyroid-related eye diseases

By performing partitioning and multi-information fusion on ocular images of thyroid-associated eye diseases, and combining edge and center information of abnormal signs, the recognition strategy is dynamically adjusted, solving the problem of insufficient recognition accuracy in existing technologies and achieving higher recognition accuracy and reliability.

CN119006364APending Publication Date: 2024-11-22THE NAVAL MEDICAL UNIV OF PLA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410934019.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing technologies for identifying ocular signs in thyroid-associated ophthalmopathy have limitations in accuracy and coverage, making it difficult to identify all ocular features. In particular, they are prone to misdiagnosis or omission, especially given the complexity of the disease and individual differences.

Method used

The system employs a partitioning and multi-information fusion mechanism. It partitions eye images using an image recognition model, identifies abnormal signs, fuses texture information, and outputs recognition results by combining the first and second abnormal information. The recognition strategy is dynamically adjusted to improve accuracy.

Benefits of technology

It improves the accuracy and reliability of identifying ocular signs of thyroid-associated ophthalmopathy, reduces the chance of misdiagnosis or missed diagnosis, and provides more comprehensive identification results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006364A_ABST
    Figure CN119006364A_ABST
Patent Text Reader

Abstract

The invention relates to the field of medical image recognition, in particular to an eye sign recognition method and device for thyroid-related eye diseases, and the method comprises the following steps: S1, carrying out the partitioning processing of an eye image; s2, abnormal signs in the range of each sub-eye image are recognized through an image recognition model, and texture information recognition is carried out at the same time; s3, performing fusion processing on the sub-eye images based on the texture information; s4, when the validity of the identification result is smaller than a threshold value, the S2 and the S3 are performed again; when the validity of the identification result is greater than or equal to a threshold value, performing S5; s5, abnormal signs in the sign auxiliary identification image range are identified through the image identification model; and S6, outputting an eye sign recognition result of the thyroid-related eye disease. According to the method, the recognition complexity can be simplified, and the recognition accuracy and reliability are improved; and in combination with multi-level and multi-angle eye sign identification of thyroid-related eye diseases, a more comprehensive identification result is provided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical image recognition, and in particular to a method and device for recognizing eye signs of thyroid-related eye diseases. Background Art

[0002] Thyroid-associated ophthalmopathy (TAO) is an organ-specific autoimmune disease that is common in patients with Graves' disease (GD), so it is also called Graves' ophthalmopathy (GO). Most patients may have clinical or laboratory test manifestations of abnormal thyroid function; thyroid-associated ophthalmopathy may also occur when the thyroid function is normal. The ocular signs of thyroid-associated ophthalmopathy include eyelid retraction, proptosis, conjunctival congestion, etc. By identifying these ocular signs of thyroid-associated ophthalmopathy, thyroid-associated ophthalmopathy can be detected early, thereby allowing early intervention and treatment.

[0003] Identification of eye signs of thyroid-related eye diseases is the key to diagnosis, intervention and treatment of thyroid-related eye diseases. The existing eye sign identification technology of thyroid-related eye diseases has the following main defects:

[0004] (1) Identification accuracy: Existing identification technologies have made some progress, but in some cases, factors such as the complexity of the disease, individual differences, and the limitations of the technology itself may lead to misjudgment or missed diagnosis;

[0005] (2) Technical limitations: Existing recognition technologies may not be able to fully cover all ocular features of thyroid-related eye diseases. For example, it may be difficult to accurately identify certain subtle lesions or early lesions.

[0006] In summary, there is an urgent need to provide a method and equipment that can accurately identify the ocular signs of thyroid-related eye diseases in light of the complexity of the disease and individual differences, so as to reduce the chances of misdiagnosis or missed diagnosis of thyroid-related eye diseases. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a method and device for identifying eye signs of thyroid-related eye diseases, which are used to reduce the probability of misdiagnosis or missed diagnosis of thyroid-related eye diseases.

[0008] In order to achieve the above object, the technical solution of the present invention is as follows: A method for identifying eye signs of thyroid-related eye diseases comprises the following steps:

[0009] S1: collecting the patient's eye image, performing partition processing on the eye image, obtaining a number of sub-eye images, and dividing the edge area of ​​the sub-eye images;

[0010] S2: Identify abnormal signs within the range of each sub-eye image through the image recognition model, and output first abnormal information; at the same time, identify texture information of the edge area divided by each sub-eye image through the image recognition model;

[0011] S3: Perform fusion processing on the sub-eye images based on texture information to obtain a verification image, and compare the verification image with the eye image before the partition processing to evaluate the effectiveness of the recognition result;

[0012] S4: When the validity of the recognition result is less than the threshold, S2 and S3 are repeated; when the validity of the recognition result is greater than or equal to the threshold, S5 is performed;

[0013] S5: taking the verification image corresponding to the recognition result being greater than or equal to the threshold as the physical sign auxiliary recognition image, identifying the abnormal physical signs within the range of the physical sign auxiliary recognition image through the image recognition model, and outputting the second abnormal information;

[0014] S6: Outputting the eye sign recognition result of thyroid-related eye disease by combining the first abnormal information and the second abnormal information.

[0015] Further, in S1, the eye images of the patient are collected as images of gaze in different directions, including: a front gaze image, an upper right gaze image, a right gaze image, a lower right gaze image, an upper left gaze image, a left gaze image, and a lower left gaze image;

[0016] In S1, before the eye images are partitioned, the images of different gaze directions are preprocessed, including: cropping, resizing, graying and denoising.

[0017] Furthermore, in S1, the eye image is partitioned, including:

[0018] S101: Preset the sub-eye image size and the starting point of the partition processing;

[0019] S102: performing partition processing on the eye image according to the sub-eye image size and the partition processing starting point;

[0020] S103: performing denoising, contrast enhancement and edge detection processing on the sub-eye image in sequence;

[0021] S104: Divide the edge regions into sub-eye images based on the edge detection processing result.

[0022] Further, in S2, the abnormal signs within the range of each sub-eye image are identified by the image recognition model, and the first abnormal information is output, including:

[0023] The edge of abnormal signs within the range of each sub-eye image is identified through an image recognition model to obtain a first abnormal sign edge; and the center of abnormal signs within the range of each sub-eye image is identified based on the first abnormal sign edge to obtain the first abnormal sign center; the first abnormal sign edge and the first abnormal sign center are combined to integrate and output the abnormal signs within the range of each sub-eye image, and integrate them into the first abnormal information.

[0024] Further, in S2, texture information recognition is performed on the preset edge area of ​​each sub-eye image by using an image recognition model, including:

[0025] The gradient direction and gradient magnitude of each pixel in the edge area divided by each sub-eye image are calculated, and the gradient direction and gradient magnitude of each pixel in each sub-eye image are formed into a directional gradient histogram; and the directional gradient histogram is converted into texture information of the corresponding sub-eye image.

[0026] Further, in S3, the sub-eye images are fused based on the texture information to obtain a verification image, including:

[0027] S301: Obtaining overall brightness information and contrast information of each sub-eye image according to texture information of each sub-eye image;

[0028] S302: splicing the sub-eye images according to the overall brightness information and contrast information of the images to generate a verification image.

[0029] Further, in S3, the verification image is compared with the eye image before the partitioning process to evaluate the validity of the recognition result, including: comparing the verification image with the eye image before the partitioning process, evaluating the similarity between the verification image and the eye image before the partitioning process through a convolutional neural network, and taking the similarity as the validity of the recognition result.

[0030] Further, in S5, the abnormal physical signs within the physical sign auxiliary recognition image are identified by the image recognition model, and the second abnormality information is output, including:

[0031] The edge of abnormal physical signs within the range of the physical sign assisted identification image is identified through the image recognition model to obtain the second abnormal physical sign edge; and the center of abnormal physical signs within the range of the physical sign assisted identification image is identified based on the second abnormal physical sign edge to obtain the second abnormal physical sign center; the abnormal physical signs within the range of the physical sign assisted identification image are integrated and output in combination with the second abnormal physical sign edge and the second abnormal sign center to integrate them into the second abnormal information.

[0032] Further, S6: combining the first abnormal information and the second abnormal information to output an eye sign recognition result of thyroid-related eye disease, including:

[0033] Locate each abnormal sign area based on the first abnormal information and the second abnormal information, compare the first abnormal sign edge and the second abnormal sign edge of the same abnormal sign, if the overlap degree of the first abnormal sign edge and the second abnormal sign edge is greater than or equal to a threshold, and if the abnormal signs of the corresponding area recorded in the first abnormal information and the second abnormal information are the same, directly output the eye sign recognition result of thyroid-related eye disease in the area;

[0034] If the overlap between the first abnormal physical sign edge and the second abnormal physical sign edge is less than a threshold, the number of the first abnormal physical sign centers within the second abnormal physical sign edge is obtained;

[0035] If the number of first abnormal sign centers within the edge of the second abnormal sign is 1, the abnormal signs of the corresponding area recorded in the first abnormal information are output as the eye sign recognition result of thyroid-related eye diseases in the area; if the number of first abnormal sign centers within the edge of the second abnormal sign is greater than 1, the second abnormal sign centers are redefined based on the first abnormal information, and the abnormal signs of the corresponding abnormal sign area are recognized in combination with the redefined second abnormal sign centers, and the eye sign recognition result of thyroid-related eye diseases in the area is output.

[0036] A device for identifying eye signs of thyroid-related eye diseases, based on any one of the above methods for identifying eye signs of thyroid-related eye diseases, comprises:

[0037] An acquisition module is used to acquire an eye image of a patient, perform partition processing on the eye image, obtain a plurality of sub-eye images, and divide the edge area of ​​the sub-eye images;

[0038] The recognition module is used to recognize abnormal signs within the range of each sub-eye image through an image recognition model and output first abnormal information; at the same time, the image recognition model is used to recognize texture information of the edge area divided by each sub-eye image;

[0039] The first verification module is used to perform fusion processing on the sub-eye images based on texture information to obtain a verification image, and compare the verification image with the eye image before the partition processing to evaluate the validity of the recognition result;

[0040] A judgment module, used to control the recognition module and the verification module to re-run when the validity of the recognition result is less than a threshold value; and to control the second verification module to run when the validity of the recognition result is greater than or equal to the threshold value;

[0041] A second verification module is used to use the verification image corresponding to the recognition result validity being greater than or equal to the threshold as the physical sign auxiliary recognition image, identify the abnormal physical signs within the range of the physical sign auxiliary recognition image through the image recognition model, and output second abnormality information;

[0042] A result output module is used to output an eye sign recognition result of thyroid-related eye disease in combination with the first abnormal information and the second abnormal information.

[0043] The above scheme has the following beneficial effects:

[0044] 1. The present invention, firstly, partitions the eye image to obtain several sub-eye images, and divides the edge area of ​​the sub-eye image. Compared with the prior art, the recognition complexity can be simplified, so that the recognition model can more accurately locate and identify abnormal eye signs, thereby improving the accuracy and reliability of recognition and reducing the impact of individual differences. Secondly, after performing sign recognition in the sub-eye image, the sub-eye image is fused and spliced, and a preliminary judgment is made on the sign recognition effect based on its fusion and splicing effect. When it is judged that the preset conditions are not met, the sign recognition and edge recognition in the sub-eye image are repeated until the fusion and splicing effect meets the preset conditions, and then the secondary sign recognition is continued to ensure the accuracy of eye sign recognition of thyroid-related eye diseases.

[0045] 2. The present invention can effectively retain and enhance the characteristics of abnormal signs by identifying the texture information of the edge area of ​​the sub-eye image and fusing the sub-eye image based on the texture information in the S3 stage, thereby further improving the accuracy of recognition.

[0046] 3. The present invention uses the verification image corresponding to the recognition result whose validity is greater than or equal to the threshold as the auxiliary physical sign recognition image, and further identifies abnormal physical signs in the S5 stage, thereby realizing multi-level and multi-angle eye sign recognition of thyroid-related eye diseases, and can provide more comprehensive recognition results.

[0047] 4. The present invention uses a multi-information fusion mechanism to identify eye signs of thyroid-related eye diseases by combining the first abnormal information and the second abnormal information, which can make full use of multiple information sources, thereby improving the accuracy and reliability of identification. The edge matching mechanism compares the overlap between the first abnormal sign edge and the second abnormal sign edge of the same abnormal sign, dynamically adjusts the identification strategy, and then outputs the identification result, further improving the accuracy of identification.

[0048] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of an embodiment of a method for identifying eye signs of thyroid-related eye diseases of the present invention;

[0050] Figure 2 A simplified flow chart of an embodiment of a method for identifying ocular signs of thyroid-related eye diseases according to the present invention;

[0051] Figure 3 A schematic diagram of eye image acquisition direction of an embodiment of a method for identifying eye signs of thyroid-related eye diseases of the present invention;

[0052] Figure 4 This is a structural block diagram of an embodiment of an eye sign recognition device for thyroid-related eye diseases of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The following is further described in detail through specific implementation methods:

[0055] Embodiment 1:

[0056] As attached Figure 1-3 As shown: A method for identifying eye signs of thyroid-related eye diseases, comprising the following steps:

[0057] S1: Collect the patient's eye image, perform partition processing on the eye image, obtain a number of sub-eye images, and divide the edge area of ​​the sub-eye images.

[0058] The eye images collected from the patient are images of gazes in different directions, including: front gaze image, upper right gaze image, right gaze image, lower right gaze image, upper left gaze image, left gaze image and lower left gaze image. For details, please refer to Figure 3 , (a1) is the upper right gaze image, (a2) is the upper left gaze image, (b1) is the right gaze image, (b2) is the left gaze image, (c1) is the lower right gaze image, and (c2) is the lower left gaze image. By collecting eye images of patients looking in different directions, more comprehensive eye information can be collected through image acquisition.

[0059] In addition, the images of different gaze directions need to be preprocessed before the collected eye images are partitioned, including: cropping, resizing, graying and denoising, to ensure the uniform size of the eye images and the quality of the eye images, so as to facilitate subsequent accurate vital sign recognition.

[0060] Partitioning the eye image specifically includes the following steps:

[0061] S101: Preset the sub-eye image size and the starting point of the partition processing; since the size of the collected eye image is cropped to a uniform size, the preset sub-eye image size determines the number of sub-eye images for subsequent partition processing. The sub-eye image size needs to be flexibly set based on the size of the collected eye image. Too large or too small will reduce the difficulty of subsequent feature recognition of the sub-eye image. In this embodiment, the size of the collected eye image is cropped to 512*256px, and the preset sub-eye image size is 64*32px. The starting point of the partition processing indicates the starting point for starting the sub-eye image division. In this embodiment, the starting point of the sub-eye image division is set as the origin.

[0062] S102: performing partition processing on the eye image according to the sub-eye image size and the partition processing starting point.

[0063] S103: performing denoising, contrast enhancement and edge detection processing on the sub-eye image in sequence; by performing denoising, contrast enhancement and edge detection processing on the sub-eye image in sequence, its overall quality can be improved, which facilitates subsequent accurate recognition of vital signs and texture information.

[0064] S104: based on the edge detection result, dividing the edge region into each sub-eye image; in this embodiment, the 1 / 4 part of the edge of the sub-eye image is defined as the edge, and based on the edge detection result, the 1 / 4 part of the entire sub-eye image from the edge is divided into the edge region.

[0065] S2: Identify abnormal signs within the range of each sub-eye image through an image recognition model and output first abnormal information; at the same time, identify texture information of edge areas divided by each sub-eye image through the image recognition model.

[0066] Among them, the abnormal signs within the range of each sub-eye image are identified by an image recognition model, and the first abnormal information is output, including: identifying the edge of the abnormal sign within the range of each sub-eye image by an image recognition model to obtain the first abnormal sign edge; and identifying the center of the abnormal sign within the range of each sub-eye image based on the first abnormal sign edge to obtain the first abnormal sign center; combining the first abnormal sign edge and the first abnormal sign center to integrate and output the abnormal signs within the range of each sub-eye image, and integrating them into the first abnormal information.

[0067] The edge contains the most basic features of abnormal signs and useful information for identification. Therefore, based on the edge recognition of abnormal signs, the area corresponding to the abnormal signs can be extracted more accurately, which provides a more favorable basis for judging the type of abnormal signs. By identifying the center of the abnormal sign, the abnormal sign target within the sub-eye image can be more accurately located to improve the accuracy of feature recognition. Abnormal signs usually include: eyelid edema, eyelid congestion, conjunctival congestion, conjunctival edema, lacrimal caruncle congestion and edema, and corneal ulcers.

[0068] Secondly, texture information of the preset edge area of ​​each sub-eye image is identified through the image recognition model, including: calculating the gradient direction and gradient magnitude of each pixel in the edge area divided by each sub-eye image, forming a directional gradient histogram of the gradient direction and gradient magnitude of each pixel in each sub-eye image; and converting the directional gradient histogram into texture information of the sub-eye image to which it belongs.

[0069] S3: Perform fusion processing on the sub-eye images based on texture information to obtain a verification image, and compare the verification image with the eye image before partition processing to evaluate the effectiveness of the recognition result.

[0070] Specifically, in S3, the sub-eye images are fused based on the texture information to obtain a verification image, including:

[0071] S301: Obtaining overall brightness information and contrast information of each sub-eye image according to texture information of each sub-eye image;

[0072] S302: splicing the sub-eye images according to the overall brightness information and contrast information of the images to generate a verification image.

[0073] In S3, the verification image is compared with the eye image before the partitioning process to evaluate the effectiveness of the recognition result, including: comparing the verification image with the eye image before the partitioning process, evaluating the similarity between the verification image and the eye image before the partitioning process through a convolutional neural network, and using the similarity as the effectiveness of the recognition result. In addition to using a convolutional neural network to evaluate the similarity, a histogram comparison can also be used: the similarity between the two images is evaluated by comparing their grayscale histograms. Various histogram similarity measurement methods can be used, such as a chi-square test, a cross entropy, etc.

[0074] S4: When the validity of the recognition result is less than the threshold, S2 and S3 are repeated; when the validity of the recognition result is greater than or equal to the threshold, S5 is performed. Through this judgment step, the eye sign recognition result of the early thyroid-related eye disease is judged, and if the preset requirements are met, the subsequent sign recognition steps are continued.

[0075] S5: Using the verification image corresponding to the recognition result whose validity is greater than or equal to the threshold as the physical sign auxiliary recognition image, identifying the abnormal physical signs within the range of the physical sign auxiliary recognition image through the image recognition model, and outputting the second abnormal information.

[0076] Specifically, similar to the S2 stage, the edge of the abnormal physical sign within the range of the physical sign assisted identification image is identified through the image recognition model to obtain the second abnormal physical sign edge; and based on the second abnormal physical sign edge, the center of the abnormal physical sign within the range of the physical sign assisted identification image is identified to obtain the second abnormal physical sign center; the second abnormal physical sign edge and the second abnormal sign center are combined to integrate the abnormal physical sign within the range of the physical sign assisted identification image and output it as the second abnormal information.

[0077] S6: Output the eye sign recognition result of thyroid-related eye disease in combination with the first abnormal information and the second abnormal information. Specifically, locate each abnormal sign area based on the first abnormal information and the second abnormal information, compare the first abnormal sign edge and the second abnormal sign edge of the same abnormal sign, if the overlap degree of the first abnormal sign edge and the second abnormal sign edge is greater than or equal to a threshold, and if the abnormal signs of the corresponding area recorded in the first abnormal information and the second abnormal information are the same, directly output the eye sign recognition result of thyroid-related eye disease in the area.

[0078] If the overlap degree between the first abnormal sign edge and the second abnormal sign edge is less than a threshold value, the number of first abnormal sign centers within the second abnormal sign edge is obtained; if the number of first abnormal sign centers within the second abnormal sign edge is 1, the abnormal signs of the corresponding area recorded in the first abnormal information are output as the eye sign recognition result of thyroid-related eye diseases in the area; if the number of first abnormal sign centers within the second abnormal sign edge is greater than 1, the second abnormal sign center is redefined based on the first abnormal information, and the abnormal sign recognition of the corresponding abnormal sign area is performed in combination with the redefined second abnormal sign center, and the eye sign recognition result of thyroid-related eye diseases in the area is output.

[0079] In this embodiment, by partitioning the eye image, obtaining several sub-eye images, and dividing the edge areas of the sub-eye images, the recognition complexity can be simplified, so that the recognition model can more accurately locate and identify abnormal eye signs, thereby improving the accuracy and reliability of recognition and reducing the impact of individual differences. A multi-information fusion mechanism is adopted to identify eye signs of thyroid-related eye diseases by combining the first abnormal information and the second abnormal information, which can make full use of multiple information sources, thereby improving the accuracy and reliability of recognition. And the edge matching mechanism compares the overlap between the first abnormal sign edge and the second abnormal sign edge of the same abnormal sign, dynamically adjusts the recognition strategy and outputs the recognition result, further improving the accuracy of recognition.

[0080] Embodiment 2:

[0081] As attached Figure 4 As shown: A device for identifying eye signs of thyroid-related eye diseases, which is based on the eye signs identification method for thyroid-related eye diseases in Example 1, includes: an acquisition module, an identification module, a first verification module, a judgment module, a second verification module and a result output module, and the acquisition module, the identification module, the first verification module, the judgment module, the second verification module and the result output module interact and communicate with each other in sequence.

[0082] The acquisition module is used to acquire the patient's eye image, perform partition processing on the eye image, obtain a number of sub-eye images, and divide the edge area of ​​the sub-eye images;

[0083] The recognition module is used to recognize abnormal signs within the range of each sub-eye image through an image recognition model and output first abnormal information; at the same time, the image recognition model is used to recognize texture information of the edge area divided by each sub-eye image;

[0084] The first verification module is used to perform fusion processing on the sub-eye images based on texture information to obtain a verification image, and compare the verification image with the eye image before the partition processing to evaluate the effectiveness of the recognition result;

[0085] The judgment module is used to control the recognition module and the verification module to re-run when the validity of the recognition result is less than the threshold; and to control the second verification module to run when the validity of the recognition result is greater than or equal to the threshold;

[0086] The second verification module is used to use the verification image corresponding to the recognition result validity being greater than or equal to the threshold as the physical sign auxiliary recognition image, identify the abnormal physical signs within the range of the physical sign auxiliary recognition image through the image recognition model, and output the second abnormal information;

[0087] The result output module is used to output the eye sign recognition result of thyroid-related eye disease in combination with the first abnormal information and the second abnormal information.

[0088] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A method for identifying ocular signs of thyroid-related eye disease, characterized in that: The steps include: S1: collecting the patient's eye image, performing partition processing on the eye image, obtaining a number of sub-eye images, and dividing the edge area of ​​the sub-eye images; S2: Identify abnormal signs within the range of each sub-eye image through the image recognition model, and output first abnormal information; at the same time, identify texture information of the edge area divided by each sub-eye image through the image recognition model; S3: Perform fusion processing on the sub-eye images based on texture information to obtain a verification image, and compare the verification image with the eye image before the partition processing to evaluate the effectiveness of the recognition result; S4: When the effectiveness of the recognition result is less than the threshold, S2 and S3 are repeated; When the validity of the recognition result is greater than or equal to the threshold, proceed to S5; S5: taking the verification image corresponding to the recognition result being greater than or equal to the threshold as the physical sign auxiliary recognition image, identifying the abnormal physical signs within the range of the physical sign auxiliary recognition image through the image recognition model, and outputting the second abnormal information; S6: Outputting the eye sign recognition result of thyroid-related eye disease by combining the first abnormal information and the second abnormal information.

2. The method for identifying eye signs of thyroid-related eye disease according to claim 1, characterized in that: In S1, the eye images of the patient are collected as images of gaze in different directions, including: a front gaze image, an upper right gaze image, a right gaze image, a lower right gaze image, an upper left gaze image, a left gaze image, and a lower left gaze image; In S1, before the eye images are partitioned, the images of different gaze directions are preprocessed, including: cropping, resizing, graying and denoising.

3. The method for identifying eye signs of thyroid-related eye disease according to claim 2, characterized in that: In S1, the eye image is partitioned, including: S101: Preset the sub-eye image size and the starting point of the partition processing; S102: performing partition processing on the eye image according to the sub-eye image size and the partition processing starting point; S103: performing denoising, contrast enhancement and edge detection processing on the sub-eye image in sequence; S104: Divide the edge regions into sub-eye images based on the edge detection processing result.

4. The method for identifying eye signs of thyroid-related eye disease according to claim 1, characterized in that: In S2, the abnormal signs within the range of each sub-eye image are identified by the image recognition model, and the first abnormal information is output, including: The edge of abnormal signs within the range of each sub-eye image is identified through an image recognition model to obtain a first abnormal sign edge; and the center of abnormal signs within the range of each sub-eye image is identified based on the first abnormal sign edge to obtain the first abnormal sign center; the first abnormal sign edge and the first abnormal sign center are combined to integrate and output the abnormal signs within the range of each sub-eye image, and integrate them into the first abnormal information.

5. The method for identifying eye signs of thyroid-related eye disease according to claim 4, characterized in that: In S2, texture information of the preset edge area of ​​each sub-eye image is recognized by an image recognition model, including: The gradient direction and gradient magnitude of each pixel in the edge area divided by each sub-eye image are calculated, and the gradient direction and gradient magnitude of each pixel in each sub-eye image are formed into a directional gradient histogram; and the directional gradient histogram is converted into texture information of the corresponding sub-eye image.

6. The method for identifying eye signs of thyroid-related eye disease according to claim 5, characterized in that: In S3, the sub-eye images are fused based on the texture information to obtain a verification image, including: S301: Obtaining overall brightness information and contrast information of each sub-eye image according to texture information of each sub-eye image; S302: splicing the sub-eye images according to the overall brightness information and contrast information of the images to generate a verification image.

7. The method for identifying eye signs of thyroid-related eye disease according to claim 6, characterized in that: In S3, the verification image is compared with the eye image before the partitioning process to evaluate the validity of the recognition result, including: comparing the verification image with the eye image before the partitioning process, evaluating the similarity between the verification image and the eye image before the partitioning process through a convolutional neural network, and taking the similarity as the validity of the recognition result.

8. The method for identifying eye signs of thyroid-related eye disease according to claim 4, characterized in that: In S5, the abnormal physical signs within the physical sign auxiliary recognition image are identified by the image recognition model, and the second abnormality information is output, including: The edge of abnormal physical signs within the range of the physical sign assisted identification image is identified through the image recognition model to obtain the second abnormal physical sign edge; and the center of abnormal physical signs within the range of the physical sign assisted identification image is identified based on the second abnormal physical sign edge to obtain the second abnormal physical sign center; the abnormal physical signs within the range of the physical sign assisted identification image are integrated and output in combination with the second abnormal physical sign edge and the second abnormal sign center to integrate them into the second abnormal information.

9. The method for identifying eye signs of thyroid-related eye disease according to claim 8, characterized in that: S6: Combining the first abnormal information and the second abnormal information, outputting an eye sign recognition result of thyroid-related eye disease, including: Locate each abnormal sign area based on the first abnormal information and the second abnormal information, compare the first abnormal sign edge and the second abnormal sign edge of the same abnormal sign, if the overlap degree of the first abnormal sign edge and the second abnormal sign edge is greater than or equal to a threshold, and if the abnormal signs of the corresponding area recorded in the first abnormal information and the second abnormal information are the same, directly output the eye sign recognition result of thyroid-related eye disease in the area; If the overlap between the first abnormal physical sign edge and the second abnormal physical sign edge is less than a threshold, the number of the first abnormal physical sign centers within the second abnormal physical sign edge is obtained; If the number of first abnormal sign centers within the edge of the second abnormal sign is 1, the abnormal signs of the corresponding area recorded in the first abnormal information are output as the eye sign recognition result of thyroid-related eye diseases in the area; if the number of first abnormal sign centers within the edge of the second abnormal sign is greater than 1, the second abnormal sign centers are redefined based on the first abnormal information, and the abnormal signs of the corresponding abnormal sign area are recognized in combination with the redefined second abnormal sign centers, and the eye sign recognition result of thyroid-related eye diseases in the area is output.

10. An eye sign recognition device for thyroid-related eye diseases, operating based on the eye sign recognition method for thyroid-related eye diseases according to any one of claims 1 to 9, characterized in that: include: An acquisition module is used to acquire an eye image of a patient, perform partition processing on the eye image, obtain a plurality of sub-eye images, and divide the edge area of ​​the sub-eye images; The recognition module is used to recognize abnormal signs within the range of each sub-eye image through an image recognition model and output first abnormal information; at the same time, the image recognition model is used to recognize texture information of the edge area divided by each sub-eye image; The first verification module is used to perform fusion processing on the sub-eye images based on texture information to obtain a verification image, and compare the verification image with the eye image before the partition processing to evaluate the validity of the recognition result; A judgment module, used to control the identification module and the verification module to re-run when the validity of the identification result is less than a threshold; When the validity of the recognition result is greater than or equal to a threshold, controlling the second verification module to run; A second verification module is used to use the verification image corresponding to the recognition result validity being greater than or equal to the threshold as the physical sign auxiliary recognition image, identify the abnormal physical signs within the range of the physical sign auxiliary recognition image through the image recognition model, and output the second abnormal information; A result output module is used to output an eye sign recognition result of thyroid-related eye disease in combination with the first abnormal information and the second abnormal information.