Method for training a model for identifying hyperthyroidism and related products

Abstract

The present disclosure relates to a method and related products for training a model for identifying hyperthyroidism. The method is implemented by a computing device and includes: acquiring healthy fundus images and fundus images of patients with hyperthyroidism; performing quality screening on the healthy and hyperthyroid fundus images to obtain a first training set of fundus images without auxiliary information and a second training set of fundus images containing the auxiliary information; and training a model for identifying hyperthyroidism based on fundus images based on the first and second training sets. Using the disclosed solution, hyperthyroidism can be efficiently identified based on fundus images.

Description

Technical Field

[0001] The present disclosure generally relates to the field of image processing technology. More specifically, the present disclosure relates to a method and apparatus for training a model for identifying hyperthyroidism, a method and apparatus for identifying hyperthyroidism, and a computer-readable storage medium. Background Art

[0002] Hyperthyroidism, commonly known as "hyperthyroidism," is caused by the thyroid gland producing and releasing excessive thyroid hormones, leading to hypermetabolism and sympathetic nervous system activation, resulting in palpitations, sweating, increased eating and bowel movements, and weight loss. Hyperthyroidism is also associated with symptoms such as exophthalmos, eyelid edema, and decreased vision. In iodine-sufficient regions, the prevalence of clinical hyperthyroidism ranges from 0.2% to 1.3%. According to the Whickham Study in the UK, the incidence of hyperthyroidism is estimated to be 100-200 per 100,000 person-years, with a prevalence of 2.7% in women and 0.23% in men. Furthermore, the National Nutrition and Health Examination Survey (NHANES III) in the United States showed that 0.5% of the general population had clinical hyperthyroidism and 0.7% had subclinical hyperthyroidism, for an overall prevalence of 1.3%. Similar incidence and prevalence rates have been reported in several studies from Sweden, Denmark, Norway, and Japan. A meta-analysis of European studies revealed an estimated average prevalence of 0.75% and an incidence of 51 per 100,000 person-years.

[0003] Hyperthyroidism is often closely linked to hormone levels, family medical history, and medication use, and requires a series of clinical examinations. Currently, screening for hyperthyroidism is typically performed through blood analysis. However, this method can lead to misdiagnosis and missed diagnosis, making large-scale population screening extremely time-consuming and labor-intensive. Therefore, developing more advanced hyperthyroidism identification systems is crucial. In recent years, deep learning technology has been widely used in computer-assisted automated fundus screening. Recent studies have demonstrated that deep learning can be used to model large-scale color fundus image data and identify patient biometrics such as age and gender, and even non-fundus-related diseases such as anemia and Alzheimer's disease. However, no such research has been conducted for hyperthyroidism. Therefore, how to quickly and efficiently identify hyperthyroidism has become a technical challenge that needs to be addressed. Summary of the Invention

[0004] To at least partially address the technical issues mentioned in the background technology, the present disclosure provides a method for training a model for identifying hyperthyroidism based on fundus images and for identifying hyperthyroidism based on fundus images. Using the present disclosure, the trained model can be used to identify hyperthyroidism based on fundus images, enabling accurate hyperthyroidism identification results to be obtained quickly and efficiently. To this end, the present disclosure provides solutions in the following aspects.

[0005] In a first aspect, the present disclosure provides a method for training a model for identifying hyperthyroidism, the method being implemented by a computing device and comprising: acquiring healthy fundus images and fundus images of patients with hyperthyroidism; performing quality screening on the healthy fundus images and the fundus images of patients with hyperthyroidism to obtain a first training set formed by fundus images not containing auxiliary information and a second training set formed by fundus images containing the auxiliary information; and training a model for identifying hyperthyroidism based on fundus images based on the first training set and the second training set.

[0006] In one embodiment, in obtaining a first training set formed by fundus images that do not contain auxiliary information, the method includes: counting a first distribution of the auxiliary information contained in the fundus images of patients with hyperthyroidism; extracting samples from the healthy fundus images based on the first distribution; and matching a second distribution of the extracted healthy fundus images that contain the auxiliary information with the first distribution to obtain a first training set formed by fundus images that do not contain auxiliary information.

[0007] In another embodiment, the auxiliary information includes at least age information and gender information.

[0008] In another embodiment, the method further includes: performing an image transformation operation on the fundus images in the first training set and / or the second training set to augment the training set of the first training set and / or the second training set, wherein the image transformation operation includes an image cropping operation, an image flipping operation and / or an image rotation operation.

[0009] In another embodiment, the model includes a verification model and a recognition model, wherein in training the model for identifying hyperthyroidism based on fundus images based on the first training set and the second training set, the method includes: using the first training set to train the verification model, wherein the verification model is used to verify whether the fundus image is correlated with hyperthyroidism; and in response to the verification model verifying that the fundus image is correlated with hyperthyroidism, using the second training set to train the recognition model, wherein the recognition model is used to output a recognition result for identifying hyperthyroidism based on fundus images.

[0010] In another embodiment, during the training of the verification model and the recognition model, the method includes: respectively calculating the corresponding binary cross entropy losses during the training of the verification model and the training of the recognition model; and optimizing the corresponding binary cross entropy losses using an optimizer until the output results of the verification model and the recognition model respectively meet the corresponding preset conditions and stop training.

[0011] In yet another embodiment, the verification model and the recognition model both include residual network models.

[0012] In a second aspect, the present disclosure also provides a device for training a model for identifying hyperthyroidism, comprising: a processor; and a memory storing program instructions for training a model for identifying hyperthyroidism, wherein when the program instructions are executed by the processor, the device implements multiple embodiments of the aforementioned first aspect.

[0013] In a third aspect, the present disclosure further provides a computer-readable storage medium having stored thereon computer-readable instructions for training a model for identifying hyperthyroidism. When the computer-readable instructions are executed by one or more processors, multiple embodiments of the aforementioned first aspect are implemented.

[0014] In a fourth aspect, the present disclosure also provides a method for identifying hyperthyroidism, comprising: obtaining a fundus image to be used for hyperthyroidism identification; inputting the fundus image into a recognition model trained by multiple embodiments in the aforementioned first aspect to output a recognition result for hyperthyroidism identification based on the fundus image.

[0015] In a fifth aspect, the present disclosure also provides a device for identifying hyperthyroidism, comprising: a processor; and a memory storing program instructions for identifying hyperthyroidism, wherein when the program instructions are executed by the processor, the device implements the embodiment of the aforementioned fourth aspect.

[0016] In a sixth aspect, the present disclosure further provides a computer-readable storage medium having stored thereon computer-readable instructions for identifying hyperthyroidism. When the computer-readable instructions are executed by one or more processors, the embodiments of the aforementioned fourth aspect are implemented.

[0017] Through the solution disclosed herein, the model is trained by acquiring healthy fundus images and fundus images of patients with hyperthyroidism, and then training the model based on two training sets formed after quality screening of the healthy fundus images and the fundus images of patients with hyperthyroidism to ensure the reliability of the trained model. Furthermore, the embodiment of the disclosure eliminates auxiliary information through quality screening, thereby improving the accuracy of hyperthyroidism identification. In addition, by inputting fundus images into the trained model, the embodiment of the disclosure can quickly and efficiently obtain hyperthyroidism identification results, so as to provide medical staff and patients with auxiliary treatment recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0019] Figure 1 is an exemplary flowchart illustrating a method for training a model for identifying hyperthyroidism based on fundus images according to an embodiment of the present disclosure;

[0020] Figure 2 is an exemplary schematic diagram illustrating a residual network model;

[0021] Figure 3 is an exemplary schematic diagram illustrating performance of a verification model and a recognition model trained based on a first training set and a second training set according to an embodiment of the present disclosure;

[0022] Figure 4 is an exemplary flowchart illustrating a method for identifying hyperthyroidism based on fundus images according to an embodiment of the present disclosure;

[0023] Figure 5 is an exemplary schematic diagram showing a recognition result of hyperthyroidism recognition based on a fundus image according to an embodiment of the present disclosure; and

[0024] Figure 6 2 is a block diagram illustrating a device for training a model for identifying hyperthyroidism based on fundus images and identifying hyperthyroidism based on fundus images according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. It should be understood that the embodiments described in this specification are only some of the embodiments provided by the present disclosure to facilitate a clear understanding of the solutions and to comply with legal requirements, and are not all embodiments that can implement the present disclosure. Based on the embodiments disclosed in this specification, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this disclosure.

[0026] Figure 1 is an exemplary flow chart illustrating a method 100 for training a model for identifying hyperthyroidism based on fundus images according to an embodiment of the present disclosure. Based on the description below, those skilled in the art will appreciate that method 100 herein can be performed by a computing device. Depending on the implementation, the computing device can be a general-purpose computing device including a processor and memory, or an artificial intelligence device including a dedicated processor (e.g., an artificial intelligence processor) and memory.

[0027] like Figure 1As shown in , at step S102, healthy fundus images and fundus images of a patient with hyperthyroidism are acquired. In one embodiment, the healthy fundus images and fundus images of a patient with hyperthyroidism can be acquired using, for example, a digital fundus camera. In particular, the healthy fundus images and fundus images of a patient with hyperthyroidism can be color fundus images acquired without dilating the pupil of the subject (or patient). It is understood that color fundus images are a very important means of fundus examination, and many diseases can be reflected in the fundus. Because the retinal blood vessels in the fundus are visible blood vessels in the human body, medical personnel use them as a window to understand the vascular conditions of other organs. Changes in fundus blood vessels reflect, to a certain extent, the extent of changes in certain organs. For example, retinal arteriosclerosis can be seen in the fundus of patients with hypertension, and capillary hemangiomas, small hemorrhages, or exudates can be seen in the fundus of patients with diabetes, which, to a certain extent, reflect changes in blood vessels throughout the body. In the disclosed embodiment, hyperthyroidism is identified using healthy color fundus images and color fundus images of a patient with hyperthyroidism.

[0028] After obtaining the above-mentioned healthy fundus images and fundus images of patients with hyperthyroidism, at step S104, the healthy fundus images and the fundus images of patients with hyperthyroidism are quality screened to obtain a first training set formed by fundus images that do not contain auxiliary information and a second training set formed by fundus images that contain auxiliary information. That is, by quality screening the healthy fundus images and the fundus images of patients with hyperthyroidism, a set of training sets that exclude auxiliary information and a set of training sets that contain auxiliary information are obtained. In some embodiments, the auxiliary information may include, for example, age information and gender information. This is because fundus images can distinguish the age and gender of the subject (or patient), and the probability of hyperthyroidism in women is higher than that in men. Therefore, when the model is trained without excluding age information and gender information, the model output result will be to distinguish, for example, gender, rather than to identify hyperthyroidism. In one implementation scenario, the first training set formed by fundus images that do not contain auxiliary information can be obtained by the following operations. That is, first, a first distribution of fundus images containing auxiliary information in hyperthyroidism is calculated, and then samples are extracted from the healthy fundus images based on the first distribution. Furthermore, the second distribution of the extracted healthy fundus images containing auxiliary information is matched with the first distribution to obtain a first training set of fundus images without auxiliary information. In other words, by registering the healthy fundus images with the fundus images of the hyperthyroidism patients based on age and gender consistency, auxiliary information such as age and gender is excluded.

[0029] In another implementation scenario, because age and gender information can assist in identifying hyperthyroidism, a second training set of fundus images containing auxiliary information such as age and gender can be obtained without registering these auxiliary information. In this scenario, since the prevalence of hyperthyroidism in the population is also somewhat related to age and gender, random samples can be drawn from healthy fundus images and fundus images of patients with hyperthyroidism based on the prevalence. In other words, samples are randomly drawn from the original distribution of healthy fundus images and fundus images of patients with hyperthyroidism to obtain a second training set of fundus images containing auxiliary information.

[0030] It should be understood that hyperthyroidism is not an eye fundus disease, and both healthy eye fundus images and eye fundus images with hyperthyroidism may have eye fundus diseases (or eye fundus lesions). Similar to the above-mentioned auxiliary information, when the eye fundus lesions are not excluded, the trained model will output the result of judging whether there are eye fundus lesions instead of identifying hyperthyroidism. Therefore, when quality screening healthy eye fundus images and eye fundus images with hyperthyroidism, it is also necessary to eliminate healthy eye fundus images with eye fundus diseases and eye fundus images with hyperthyroidism to improve the accuracy of hyperthyroidism recognition results.

[0031] Based on the first training set and the second training set obtained above, at step S106, a model for identifying hyperthyroidism based on fundus images is trained based on the first training set and the second training set. In one embodiment, the model may include a verification model and a recognition model. Specifically, the verification model is first trained using the first training set. The verification model is used to verify whether the fundus image is correlated with hyperthyroidism. Then, in response to the verification model verifying that the fundus image is correlated with hyperthyroidism, the recognition model is trained using the second training set. The aforementioned recognition model is used to output a recognition result for identifying hyperthyroidism based on the fundus image. That is, the embodiment of the present disclosure first determines that the fundus image is correlated with hyperthyroidism through the verification model, and then uses the recognition model to identify hyperthyroidism based on the fundus image.

[0032] In some embodiments, in order to avoid over-fitting in the training network, the disclosed embodiments may also perform image transformation operations on the fundus images in the first training set and / or the second training set to perform training set augmentation on the first training set and / or the second training set. The aforementioned image transformation operations include image cropping operations, image flipping operations, and / or image rotation operations. For example, the fundus images in the first training set and / or the second training set are randomly cropped into image areas of size 512*512 by image cropping operations. The aforementioned image flipping operations include but are not limited to horizontal flipping and vertical flipping, and the aforementioned image rotation operations include randomly selecting an offset angle within a preset angle range (0°-15°) to rotate the fundus image.

[0033] In one implementation scenario, the above-mentioned verification model and recognition model include but are not limited to residual network models. In the implementation scenario, during the aforementioned training of the verification model and the recognition model, the corresponding binary cross entropy loss during the verification model training period and the recognition model training period can be calculated respectively, and then the corresponding binary cross entropy loss is optimized using an optimizer (such as Adamoptimizer) until the output results of the verification model and the recognition model meet the corresponding preset conditions. The training is stopped. Based on the trained recognition model, the recognition result of hyperthyroidism recognition can be output based on the fundus image.

[0034] In combination with the above description, it can be seen that the embodiment of the present disclosure performs quality screening on healthy fundus images and fundus images of patients with hyperthyroidism to divide healthy fundus images and fundus images of patients with hyperthyroidism into a first training set (i.e., registered data) formed by fundus images that do not contain auxiliary information and a second training set (i.e., unregistered data) formed by fundus images that contain auxiliary information. Furthermore, a verification model is trained based on the aforementioned first training set to verify whether the fundus images are correlated with hyperthyroidism, and after verifying that the fundus images are correlated with hyperthyroidism, the second training set is used to train the recognition model. By using the trained recognition model to perform hyperthyroidism recognition based on fundus images, accurate hyperthyroidism recognition results can be obtained quickly and efficiently.

[0035] Figure 2 is an exemplary schematic diagram showing a residual network model. Figure 2Part of the residual network is shown as an example, and it is assumed that the first layer of the residual network is shown. Among them, x represents the input and F(x) represents the residual. It can be understood that F(x) is the output after x undergoes linear change in the first layer and is activated by the activation function relu. Furthermore, the figure also shows that before the second layer undergoes linear change and activation, F(x) adds the input value x of the layer, and then outputs after activation, and the path formed by adding x from the second layer output to the activation is a shortcut connection. The residual network is relatively easy to optimize, and the accuracy can be improved by increasing the depth considerably. In addition, the residual blocks inside the residual network use jump connections to alleviate the gradient vanishing problem caused by increasing the depth in the deep neural network.

[0036] In the embodiment of the present disclosure, a residual network of, for example, 50 layers can be used to train based on the first training set and the second training set. For the verification model and the recognition model, their inputs x are the first training set and the second training set respectively, and their outputs are probabilities. For example, the verification model outputs the probability of the association between the fundus image and hyperthyroidism, and the recognition model outputs the probability of identifying hyperthyroidism. As mentioned above, during the training of the verification model and the recognition model, the corresponding binary cross entropy loss during the training of the verification model and the recognition model is calculated respectively, and the corresponding binary cross entropy loss is optimized using an optimizer (e.g., Adam optimizer). The batch size can be 16. That is, the number of fundus images input in each iteration. The aforementioned optimizer is continuously optimized until the output results of the verification model and the recognition model meet the corresponding preset conditions, and the training is stopped. The preset conditions can be based on the output results of the verification set and / or the test set. In the implementation scenario, the training of the model can be supervised by setting a verification set and a test set, wherein the verification set can be used to determine whether the training is stopped, and the test set can be used to reflect the performance of the model. For example, when the output results of the verification model and the recognition model reach the output results of their respective verification sets, their training is stopped. Based on this, the training of the verification model and the recognition model can be completed.

[0037] In some embodiments, the performance of the model can be indicated by multiple indicators, wherein the aforementioned multiple indicators may include but are not limited to AUC, sensitivity, specificity, accuracy, positive predictive value, negative predictive value and F1-score, for example Figure 3 shown.

[0038] Figure 3 1 is an exemplary diagram showing the performance of the verification model and the recognition model trained based on the first training set and the second training set according to an embodiment of the present disclosure. Figure 3The results of the validation model and the recognition model trained on the first training set (i.e., registration data) formed by the fundus images without auxiliary information and the second training set (i.e., unregistered data) formed by the fundus images containing auxiliary information under multiple indicators are shown in the figure. Among them, the multiple indicators are AUC, accuracy, sensitivity, specificity, positive predictive value, negative predictive value and F1-score. For the registered data, the results corresponding to each of the above indicators are 0.850, 0.780, 0.718, 0.824, 0.659, 0.861 and 0.813 respectively, while for the unregistered data, the results corresponding to each indicator are 0.607, 0.809, 0.761, 0.885, 0.761, 0.866 and 0.837 respectively.

[0039] Depend on Figure 3 It can be seen that when using the registered data for training, it can be regarded as completely or partially ignoring other factors that are relevant to the identification of hyperthyroidism (such as age and gender). In this scenario, the AUC of the verification model can reach 0.85, which proves that there is a certain correlation between fundus images and hyperthyroidism. In the case of unregistered data for training, that is, after adding relevant factors such as age and gender as auxiliary information, the performance of the recognition model has been further improved. For example, its AUC reached 0.607, while the sensitivity and specificity reached 0.761 and 0.885 respectively. It can be seen that the hyperthyroidism recognition results obtained by using fundus images to input the trained recognition model for hyperthyroidism recognition have a certain reliability. Based on this, the hyperthyroidism recognition results can be obtained by inputting the fundus images into the recognition model.

[0040] Figure 4 FIG. 4 is an exemplary flow chart showing a method 400 for identifying hyperthyroidism based on fundus images according to an embodiment of the present disclosure. Figure 4 As shown in FIG, at step S402, a fundus image to be used for hyperthyroidism identification is obtained. In one embodiment, the fundus image can be collected using, for example, a digital fundus camera, and the fundus image can be a color fundus image. At step S404, the fundus image is input into the trained recognition model to output a recognition result for hyperthyroidism identification based on the fundus image. Regarding the training of the recognition model, reference can be made to the above-mentioned Figure 1-Figure 2The content described is not repeated in this disclosure. As previously mentioned, the recognition model outputs a binary classification, and its probability range is between (0, 1). In an exemplary scenario, when 0 represents health and 1 represents hyperthyroidism, the output result is 1, indicating that it is recognized as hyperthyroidism.

[0041] Figure 5 FIG. 1 is an exemplary schematic diagram showing the recognition result of hyperthyroidism recognition based on fundus images according to an embodiment of the present disclosure. Figure 5 As shown in the figure, the left half of A, B, and C in the figure are healthy fundus images of healthy people, and the right half of A, B, and C in the figure are the recognition results obtained by inputting healthy fundus images into the recognition model. Furthermore, the left half of D, E, and F in the figure are fundus images of patients with hyperthyroidism, and the right half of D, E, and F in the figure are the recognition results obtained by inputting fundus images of patients with hyperthyroidism into the recognition model. Figure 5 As can be seen, compared to fundus images of healthy individuals, fundus images of patients with hyperthyroidism highlight densely populated areas of large fundus vessels and the optic nerve head. Therefore, hyperthyroidism identification based on the recognition model is reliable, and visualizing the recognition results also helps to improve the interpretability of the recognition model.

[0042] Figure 6 is a block diagram illustrating a device 600 for training a model for identifying hyperthyroidism based on fundus images and for identifying hyperthyroidism based on fundus images according to an embodiment of the present disclosure. It is understood that the device implementing the solution of the present disclosure may be a single device (e.g., a computing device) or a multifunctional device including various peripheral devices.

[0043] like Figure 6 As shown in , the device of the present disclosure may include a central processing unit or central processing unit ("CPU") 611, which may be a general-purpose CPU, a dedicated CPU, or other information processing and program execution unit. Further, the device 600 may also include a large-capacity memory 612 and a read-only memory ("ROM") 613, wherein the large-capacity memory 612 can be configured to store various types of data, including various fundus images related to healthy eye fundus, fundus images of eye fundus with hyperthyroidism, or eye fundus images to be identified as hyperthyroidism, algorithm data, intermediate results, and various programs required to run the device 600. The ROM 613 can be configured to store data and instructions required for the power-on self-test of the device 600, the initialization of various functional modules in the system, the basic input / output drivers of the system, and the booting of the operating system.

[0044] Optionally, the device 600 may also include other hardware platforms or components, such as the tensor processing unit ("TPU") 614, graphics processing unit ("GPU") 615, field programmable gate array ("FPGA") 616, and machine learning unit ("MLU") 617 shown. It will be understood that although a variety of hardware platforms or components are shown in the device 600, this is merely exemplary and not restrictive, and those skilled in the art may add or remove corresponding hardware according to actual needs. For example, the device 600 may include only a CPU, related storage devices, and interface devices to implement the method disclosed herein for training a model for identifying hyperthyroidism based on fundus images or the method for identifying hyperthyroidism based on fundus images.

[0045] In some embodiments, in order to facilitate the transmission and interaction of data with an external network, the device 600 of the present disclosure further includes a communication interface 618, so that it can be connected to a local area network / wireless local area network ("LAN / WLAN") 605 through the communication interface 618, and then connected to a local server 606 or to the Internet ("Internet") 607 through the LAN / WLAN. Alternatively or additionally, the device 600 of the present disclosure can also be directly connected to the Internet or a cellular network based on wireless communication technology through the communication interface 618, such as wireless communication technology based on the third generation ("3G"), the fourth generation ("4G") or the fifth generation ("5G") generation. In some application scenarios, the device 600 of the present disclosure can also access a server 608 and a database 609 of an external network as needed to obtain various known image models, data and modules, and can remotely store various data, such as various data or instructions for presenting recognition results, model training, etc.

[0046] The peripheral devices of device 600 may include a display device 602, an input device 603 and a data transmission interface 604. In one embodiment, the display device 602 may, for example, include one or more speakers and / or one or more visual displays, which are configured to provide voice prompts and / or image video displays for the process of training the model for identifying hyperthyroidism based on fundus images based on the first training set and the second training set or the recognition result of the final hyperthyroidism recognition. The input device 603 may include other input buttons or controls such as a keyboard, a mouse, a microphone, a gesture capture camera, etc., which are configured to receive input and / or user instructions of fundus images. The data transmission interface 604 may include, for example, a serial interface, a parallel interface or a universal serial bus interface ("USB"), a small computer system interface ("SCSI"), a serial ATA, a FireWire ("FireWire"), a PCI Express and a high-definition multimedia interface ("HDMI"), etc., which are configured for data transmission and interaction with other devices or systems. According to the solution of the present disclosure, the data transmission interface 604 can receive healthy fundus images, fundus images of patients with hyperthyroidism, or fundus images to be identified for hyperthyroidism collected from the fundus camera, and transmit fundus images or various other types of data or results to the device 600.

[0047] The CPU 611, mass storage 612, ROM 613, TPU 614, GPU 615, FPGA 616, MLU 617, and communication interface 618 of the device 600 of the present disclosure can be interconnected via a bus 619 and can interact with peripheral devices via the bus. In one embodiment, the CPU 611 can control other hardware components in the device 600 and its peripheral devices via the bus 619.

[0048] Combination of the above Figure 6 The present invention describes a device for training a model for identifying hyperthyroidism based on fundus images and for identifying hyperthyroidism based on fundus images, which can be used to implement the present invention. It should be understood that the device structure or architecture described herein is merely exemplary, and the implementation and implementation of the present invention are not limited thereto and may be modified without departing from the spirit of the present invention.

[0049] According to the above description in conjunction with the accompanying drawings, those skilled in the art can also understand that the embodiments of the present disclosure can also be implemented by software programs. Therefore, the present disclosure also provides a computer program product. The computer program product can be used to implement the embodiments of the present disclosure in conjunction with the accompanying drawings. Figure 1 The method described for training a model for hyperthyroidism recognition based on fundus images and the method described in combination with the accompanying Figure 4 The described method for hyperthyroidism identification based on fundus images.

[0050] It should be noted that although the operations of the disclosed method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in that particular order, or that all of the operations shown must be performed to achieve the desired results. Rather, the steps depicted in the flowcharts may be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into a single step, and / or a single step may be broken down into multiple steps.

[0051] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of the present disclosure, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or combinations thereof.

[0052] It should also be understood that the terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the present disclosure. As used in this disclosure and the claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should further be understood that the term "and / or" as used in this disclosure and the claims refers to any and all possible combinations of one or more of the associated listed items, including and including these combinations.

[0053] Although the embodiments of the present disclosure are as described above, the contents are merely examples used to facilitate understanding of the present disclosure and are not intended to limit the scope and application scenarios of the present disclosure. Any person skilled in the art of the present disclosure may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope of the present disclosure, but the scope of patent protection of the present disclosure shall still be based on the scope defined by the attached claims.

Claims

1. A method for training a model for identifying hyperthyroidism, characterized in that: The method is implemented by a computing device and includes: Obtaining healthy fundus images and fundus images of patients with hyperthyroidism; Performing quality screening on the healthy fundus images and the fundus images of the patient with hyperthyroidism to obtain a first training set formed by fundus images not containing auxiliary information and a second training set formed by fundus images containing the auxiliary information, wherein the auxiliary information includes age information and gender information; Using the first training set to train a verification model, wherein the verification model is used to verify whether fundus images are correlated with hyperthyroidism; and In response to the verification model verifying that the fundus image is correlated with hyperthyroidism, the recognition model is trained using the second training set, wherein the recognition model is used to output a recognition result for hyperthyroidism recognition based on the fundus image.

2. The method according to claim 1, characterized in that In obtaining a first training set formed by fundus images not containing auxiliary information, the method includes: Counting the first distribution of the auxiliary information contained in the fundus image of the patient suffering from hyperthyroidism; Extracting samples from the healthy fundus image according to the first distribution; and The second distribution of the extracted healthy fundus images containing the auxiliary information is matched with the first distribution to obtain a first training set formed by fundus images not containing the auxiliary information.

3. The method according to claim 2, characterized in that The auxiliary information includes at least age information and gender information.

4. The method according to claim 1, wherein Also includes: An image transformation operation is performed on the fundus images in the first training set and / or the second training set to augment the training set of the first training set and / or the second training set, wherein the image transformation operation includes an image cropping operation, an image flipping operation and / or an image rotation operation.

5. The method according to claim 1, wherein During training of the verification model and the recognition model, the method includes: Calculating the binary cross entropy loss corresponding to the validation model training period and the recognition model training period respectively; and The corresponding binary cross entropy loss is optimized using an optimizer, and the training is stopped until the output results of the verification model and the recognition model respectively meet the corresponding preset conditions.

6. The method according to claim 5, characterized in that The verification model and the recognition model both include residual network models.

7. A device for training a model for identifying hyperthyroidism, characterized in that: include: processor; as well as A memory storing program instructions for training a model for identifying hyperthyroidism, wherein when the program instructions are executed by the processor, the device implements the method according to any one of claims 1-6.

8. A computer-readable storage medium storing computer-readable instructions for training a model for identifying hyperthyroidism, wherein the computer-readable instructions, when executed by one or more processors, implement the method according to any one of claims 1 to 6.

9. A method for identifying hyperthyroidism, characterized in that: include: Acquiring a fundus image for hyperthyroidism identification; The fundus image is input into a verification model and a recognition model trained according to any one of the methods of claims 1-6 to output a recognition result of hyperthyroidism recognition based on the fundus image.

10. A device for identifying hyperthyroidism, characterized in that: include: processor; as well as A memory storing program instructions for identifying hyperthyroidism, wherein when the program instructions are executed by the processor, the device implements the method according to claim 9.

11. A computer-readable storage medium having stored thereon computer-readable instructions for identifying hyperthyroidism, wherein when the computer-readable instructions are executed by one or more processors, the method according to claim 9 is implemented.

Images