Method for processing ultra-wide-angle eye fundus image and related product

Abstract

The invention discloses a method for processing an ultra-wide-angle eye fundus image and a related product. The method comprises the following steps: generating a plurality of to-be-processed images of which the resolution is smaller than that of an ultra-wide-angle eye fundus image based on the ultra-wide-angle eye fundus image; processing the plurality of images based on a trained image segmentation model to output a segmentation result about the plurality of images, the image segmentation model being obtained by training a common fundus image; and determining a segmentation result of the ultra-wide-angle eye fundus image according to the segmentation results of the plurality of images. Through the scheme disclosed by the invention, effective segmentation of the ultra-wide-angle fundus image can be realized by means of the image segmentation model trained by the common fundus image.

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 for processing ultra-wide-angle fundus images, an electronic device for executing the aforementioned method, and a computer-readable storage medium. Background Art

[0002] Compared with ordinary fundus images, ultra-wide-angle fundus images have a larger field of view. Therefore, the resolution of each ultra-wide-angle fundus image is much higher than that of ordinary fundus images. This also means that when ultra-wide-angle fundus images need to be labeled with blood vessel labels, lesion labels, etc. (for example, when ultra-wide-angle fundus images are used as training samples for training segmentation models), the annotation cost of ultra-wide-angle images is much greater than the annotation cost of general fundus images. For example, a skilled annotator can finely annotate the vascular distribution of more than 20 ordinary fundus images a day, but cannot annotate the vascular distribution of an ultra-wide-angle image in a day. Therefore, due to the difficulty in sample collection and annotation, training the segmentation model using ultra-wide-angle fundus images as training samples is not only difficult but also has unsatisfactory training effects, resulting in poor segmentation effects for ultra-wide-angle fundus images.

[0003] In view of this, there is an urgent need to provide a solution for processing ultra-wide-angle fundus images so as to achieve effective segmentation of the ultra-wide-angle fundus images. Summary of the invention

[0004] In order to at least solve one or more of the technical problems mentioned above, the present disclosure proposes a solution for processing ultra-wide-angle fundus images in multiple aspects.

[0005] In a first aspect, the present disclosure provides a method for processing an ultra-wide-angle fundus image, the method comprising: generating a plurality of images to be processed having a resolution smaller than that of the ultra-wide-angle fundus image based on the ultra-wide-angle fundus image; processing the plurality of images based on a trained image segmentation model to output segmentation results regarding the plurality of images, wherein the image segmentation model is trained based on ordinary fundus images; and determining a segmentation result of the ultra-wide-angle fundus image based on the segmentation results of the plurality of images.

[0006] In some embodiments, generating a plurality of images to be processed based on the ultra-wide-angle fundus image includes: performing rasterization processing on the ultra-wide-angle fundus image to obtain the plurality of images.

[0007] In some embodiments, rasterizing the ultra-wide-angle fundus image includes: determining rasterization parameters according to the resolution of the ultra-wide-angle fundus image and the resolution of the image to be processed; and rasterizing the ultra-wide-angle fundus image using the rasterization parameters.

[0008] In some embodiments, performing rasterization processing on the ultra-wide-angle fundus image includes: performing rasterization processing on the ultra-wide-angle fundus image according to preset rasterization parameters.

[0009] In some embodiments, the image segmentation model is trained by the following operations: rasterizing ordinary fundus images and their corresponding segmentation labels as training samples to obtain raster fundus images and image labels; and optimizing the image classification model using the raster fundus images and the image labels.

[0010] In some embodiments, optimizing the training of the image segmentation model using the grid fundus image and the image label includes: processing the grid fundus image using the image segmentation model to obtain a prediction result; calculating a loss function between the prediction result and the image label; and optimizing the parameters of the image segmentation model based on the loss function to achieve optimized training of the image segmentation model.

[0011] In some embodiments, wherein the loss function comprises a cross entropy loss function, optimizing the parameters of the image segmentation model based on the loss function comprises: optimizing the parameters of the image segmentation model based on minimization processing of the cross entropy loss function.

[0012] In some embodiments, determining the segmentation result of the ultra-wide-angle fundus image according to the segmentation results of the multiple images includes: merging the segmentation results of the multiple images to obtain the segmentation result of the ultra-wide-angle fundus image.

[0013] In a second aspect, the present disclosure provides an electronic device comprising: a processor; and a memory storing computer instructions for processing ultra-wide-angle fundus images, wherein when the computer instructions are executed by the processor, the electronic device executes the method described in the first aspect of the present disclosure.

[0014] In a third aspect, the present disclosure provides a computer-readable storage medium comprising program instructions for processing ultra-wide-angle fundus images. When the program instructions are executed by a processor, the method described in the first aspect of the present disclosure is implemented.

[0015] Through the scheme for processing ultra-wide-angle fundus images provided above, the disclosed embodiment generates multiple images with a resolution smaller than that of the ultra-wide-angle fundus image from the ultra-wide-angle fundus image, and processes these images with the help of an image segmentation model trained with ordinary fundus images, so as to determine the segmentation result of the ultra-wide-angle fundus image based on the segmentation results of these images, thereby achieving the segmentation of the ultra-wide-angle fundus image. It can be seen that the scheme disclosed in the present invention divides the ultra-wide-angle fundus image into multiple images with a resolution close to that of the ordinary fundus image, effectively reducing the data difference between the ultra-wide-angle fundus image and the ordinary fundus image, so that a segmentation model that can be used in the ultra-wide-angle fundus image can be trained using the annotated ordinary fundus image. As a result, not only the segmentation processing process of the ultra-wide-angle fundus image is simplified, but also the efficient segmentation of the ultra-wide-angle fundus image is achieved. In addition, the entire process does not rely on the label of the ultra-wide-angle fundus image, which can greatly save the annotation cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. 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, wherein:

[0017] Figure 1 A comparison diagram of the rasterization of a normal fundus image and an ultra-wide-angle fundus image according to an embodiment of the present disclosure is shown;

[0018] Figure 2 An exemplary flow chart of a method for processing an ultra-wide-angle fundus image according to an embodiment of the present disclosure is shown;

[0019] Figure 3 An exemplary flow chart showing a method for processing an ultra-wide-angle fundus image according to another embodiment of the present disclosure;

[0020] Figure 4 An exemplary flow chart showing a method for processing an ultra-wide-angle fundus image according to yet another embodiment of the present disclosure;

[0021] Figure 5 An exemplary flow chart of a method for training an image segmentation model according to an embodiment of the present disclosure is shown;

[0022] Figure 6 A schematic diagram showing the segmentation result of the ultra-wide-angle fundus image of an embodiment of the present disclosure; and

[0023] Figure 7 An exemplary structural block diagram of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0025] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0026] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0027] As used in this specification and claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0028] In the prior art, segmentation of blood vessels, lesions, etc. is mostly performed on ordinary fundus images. However, the segmentation of ultra-wide-angle fundus images is not only difficult but also has unsatisfactory segmentation effects. Specifically, in the related art, ultra-wide-angle fundus images are often used as training samples to train segmentation models to cope with the segmentation task of ultra-wide-angle fundus images. However, ultra-wide-angle fundus images are not only difficult to collect, but also have a much higher labeling cost than ordinary fundus images, making the segmentation of ultra-wide-angle fundus images a technical bottleneck. In this regard, the inventors have found through research and a large amount of data analysis that when segmenting ultra-wide-angle fundus images and ordinary fundus images, the difference in data is mainly caused by resolution. Figure 1As shown, after the resolution of the ordinary fundus image 11 is adjusted, a low-resolution first image 12 is obtained, and after the resolution of the ultra-wide-angle fundus image is adjusted, a low-resolution second image 14 is obtained, wherein the resolutions of the first image 12 and the second image 14 are close. It can be seen that the data difference between the first image 12 and the second image is small. Therefore, it is considered that the ultra-wide-angle fundus image can be processed to make the resolution close to that of the ordinary fundus image, so that the data difference between the images can be reduced, so that the transfer learning strategy can be used to process the ultra-wide-angle fundus image using the image classification model trained by the labeled ordinary fundus image.

[0029] In view of this, the presently disclosed embodiment provides a solution for processing ultra-wide-angle fundus images, which processes low-resolution pictures generated by ultra-wide-angle fundus images with the help of an image segmentation model trained with ordinary fundus images, so as to determine the segmentation results of the ultra-wide-angle fundus images based on the segmentation results of these pictures, thereby achieving segmentation of the ultra-wide-angle fundus images.

[0030] The specific implementation of the present disclosure is described in detail below with reference to the accompanying drawings.

[0031] Figure 2 An exemplary flow chart of a method for processing an ultra-wide-angle fundus image according to an embodiment of the present disclosure is shown.

[0032] like Figure 2 As shown, at step S201, multiple images to be processed with a resolution smaller than that of the ultra-wide-angle fundus image can be generated based on the ultra-wide-angle fundus image. The resolution of the ultra-wide-angle fundus image is much greater than that of the ordinary fundus image, so that the data distribution of the ultra-wide-angle fundus image is greatly different from that of the ordinary fundus image. In practical applications, the ultra-wide-angle fundus image can be processed to reduce the resolution of the ultra-wide-angle fundus image, so as to reduce the data difference between the ultra-wide-angle fundus image and the ordinary fundus image.

[0033] Next, at step S202, the aforementioned multiple images can be processed based on the trained image segmentation model to output segmentation results for the multiple images. The aforementioned image segmentation model is trained based on ordinary fundus images. When the data difference between the obtained multiple images and the ordinary fundus images is small, a data migration strategy can be adopted to process these images using the image segmentation model trained by the annotated ordinary fundus images. This process does not require the image segmentation model to be trained by annotated ultra-wide-angle fundus images to implement the segmentation task of the ultra-wide-angle fundus images.

[0034] Finally, at step S203, the segmentation result of the ultra-wide-angle fundus image can be determined according to the segmentation results of the multiple images. After the segmentation results of the multiple images are obtained, the segmentation results of these images can be further used to form the segmentation result of the ultra-wide-angle fundus image.

[0035] The image segmentation model (such as the blood vessel extraction model) trained directly using ordinary fundus images can only be applied to ordinary fundus images, and cannot have good performance on ultra-wide-angle images. Specifically, there is a great difference in the data distribution between ultra-wide-angle images and ordinary fundus images, which is also the reason why the model trained on ordinary fundus images cannot be directly applied to ultra-wide-angle images. Therefore, by dividing the ultra-wide-angle fundus image into multiple images with a resolution close to that of the ordinary fundus image, the data difference between the ultra-wide-angle fundus image and the ordinary fundus image is effectively reduced, so that the segmentation model that can be used in the ultra-wide-angle fundus image can be trained using the annotated ordinary fundus image. Therefore, not only is the segmentation process of the ultra-wide-angle fundus image simplified, but also the efficient segmentation of the ultra-wide-angle fundus image is achieved. In addition, the entire process does not rely on the label of the ultra-wide-angle fundus image, which can greatly save the annotation cost.

[0036] In some embodiments, the resolution of the ultra-wide-angle fundus image can be reduced by rasterizing the ultra-wide-angle fundus image, so that the image obtained by rasterization is close to the data distribution of the ordinary fundus image. The technical personnel in this field have a certain inertia of thinking about the application of rasterization. Rasterization can be used in the model training of segmentation tasks or other tasks. However, the application scenario of rasterization is limited to increasing the diversity of samples in the training stage of the model. For example, in the training stage, the number of samples is increased by rasterizing the training samples, thereby realizing the optimization training of the model. In the inference stage of the model, the sample is directly input into the model for processing, and the output result of the model is the processing result of the sample. In the scheme disclosed in the present invention, the inventor breaks through this inertia of thinking about rasterization, uses rasterization to reduce the resolution of the picture to reduce the data difference between the pictures, and can apply rasterization to the inference stage of the model, so that the rasterization is innovatively applied in the inference stage of the classification model.

[0037] The following combination Figure 3 and Figure 4 The application of rasterization processing in the model reasoning stage is explained.

[0038] Figure 3 FIG. 3 is an exemplary flow chart of a method 300 for processing an ultra-wide-angle fundus image according to another embodiment of the present disclosure. It should be noted that: Figure 3 yes Figure 2 A specific exemplary embodiment of the method 200 is a further limitation and expansion of the method 200. Figure 2 The relevant detailed description also applies to the following.

[0039] like Figure 3As shown, at step S301, the rasterization parameters can be determined according to the resolution of the ultra-wide-angle fundus image and the resolution of the image to be processed, and then the ultra-wide-angle fundus image is rasterized using the aforementioned rasterization parameters to obtain multiple images. Ordinary fundus images generally have a lower resolution, so they can be rasterized into fewer tiles, and ultra-wide-angle fundus images generally have a larger resolution, so they can be rasterized into more tiles to keep the tile sizes of the two similar. In some embodiments, the rasterization parameters may include the number of parts into which the image is divided in different dimensions (such as width and height). Specifically, the calculation can be performed using the formula n1=W / / w, n2=H / / h. Wherein, / / represents the operation of the quotient of division. n1 and n2 represent how many parts the ultra-wide-angle fundus image is to be divided into in width and height, and the ultra-wide-angle fundus image will be rasterized into a total of n1×n2 images. W and H represent the width and height resolutions of the ultra-wide-angle fundus image, while w and h represent the width and height resolutions of the image to be processed. The value range of w and h is [128-512], usually an integer multiple of 16. Preferably, w and h are 256 respectively, because the resolutions of ultra-wide-angle images and ordinary fundus images are not fixed, and different cameras may produce images of different resolutions. When w and h are 256 respectively, not only is the compatibility with images of different resolutions better, but the stability of the model is also better. Of course, the description of the value range here is only an example, and can be adjusted according to factors such as the architecture of the classification model and application requirements.

[0040] Next, at step S302, the aforementioned multiple images can be processed based on the image segmentation model trained according to the common fundus image. The multiple images obtained by rasterization processing have a resolution close to that of the common fundus image, that is, the data difference between the multiple images obtained by rasterization processing and the common fundus image is small. At this time, these images can be processed with the aid of the image segmentation model trained according to the common fundus image to obtain the segmentation results of these images.

[0041] Finally, at step S303, the segmentation results of the multiple images may be merged to obtain the segmentation result of the ultra-wide-angle fundus image. For example, the segmentation results of the images may be de-rasterized to obtain the segmentation result of the ultra-wide-angle fundus image.

[0042] In some embodiments, assuming that the segmentation task of the image segmentation model is to segment the location of blood vessels or lesions, the segmentation results corresponding to each small block can be merged (ie, de-rasterized) into a segmentation result of the entire ultra-wide-angle image scale. Figure 6As shown, taking blood vessel segmentation as an example, the ultra-wide-angle fundus image is rasterized into multiple images and output to the image segmentation model trained for ordinary fundus images. Finally, the ultra-wide-angle blood vessel position is obtained by merging the prediction results of each image output by the model.

[0043] Figure 4 FIG. 4 is an exemplary flow chart of a method 400 for processing an ultra-wide-angle fundus image according to another embodiment of the present disclosure. It should be noted that: Figure 4 yes Figure 2 Another specific exemplary embodiment of the method 200 is a further limitation and expansion of the method 200. Figure 2 The relevant detailed description also applies to the following.

[0044] like Figure 4 As shown, at step S401, the ultra-wide-angle fundus image can be rasterized according to the preset rasterization parameters. In practical applications, the rasterization parameters can be preset according to the specific architecture of the classification model and the resolution of the ultra-wide-angle fundus image and the ordinary fundus image. In the subsequent processing process, the image can be rasterized directly according to the unified rasterization parameters. For example, the ultra-wide-angle fundus image can be uniformly divided into 64 images of 8×8 size. In addition, when the model is trained using ordinary fundus images, the ordinary fundus images can be uniformly divided into 16 images of 4×4 size. In this way, the entire image processing process can be further simplified. It should be noted that the description of the preset rasterization parameter values ​​here is only an exemplary description.

[0045] Next, at step S402, the aforementioned multiple images can be processed based on the image segmentation model trained according to the ordinary fundus image. The multiple images obtained by rasterization processing have a resolution close to that of the ordinary fundus image (although there is a difference in resolution between the image rasterized by the preset rasterization image and the ordinary fundus image, the difference is relatively small), that is, the data difference between the multiple images obtained by rasterization processing and the ordinary fundus image is small. At this time, these images can be processed with the help of the image segmentation model trained according to the ordinary fundus image to obtain the segmentation results of these images.

[0046] Finally, at step S403, the segmentation results of the multiple images may be merged to obtain the segmentation result of the ultra-wide-angle fundus image. For example, the segmentation results of the images may be de-rasterized to obtain the segmentation result of the ultra-wide-angle fundus image.

[0047] In some embodiments, the image segmentation model may be trained in a variety of ways. Figure 5 A specific implementation method 500 of using common fundus images to train an image segmentation model is shown. Figure 5 The specific training process of the image segmentation model is explained by taking the blood vessel segmentation model as an example.

[0048] like Figure 5 As shown, at step S501, a normal fundus image (eg, a 45-degree fundus image) and its segmentation label as a training sample may be obtained, wherein the segmentation label here is also a label in image form (eg, a blood vessel segmentation label).

[0049] Next, at step S502, rasterization parameters may be calculated. The rasterization parameters may be determined based on factors such as the architecture of the model, the original resolution of the common fundus image, and the resolution of the rasterized fundus image. Of course, preset unified rasterization parameters may also be used.

[0050] Next, at step S503, the common fundus image and the segmented label image are rasterized using rasterization parameters.

[0051] Next, at step S504, the rasterized fundus image and the label of the rasterized image obtained by the rasterization process are obtained.

[0052] Next, at step S505, the rasterized fundus image obtained by rasterization and the label of the raster image can be used for model optimization training. Specifically, in some embodiments, the rasterized fundus image can be processed using an image segmentation model (e.g., a blood vessel segmentation model) to obtain a prediction result. Then, a loss function between the prediction result and the image label is calculated, and the parameters of the image segmentation model are optimized based on the loss function to achieve optimization training of the image segmentation model.

[0053] Furthermore, in some embodiments, the cross entropy loss function between the prediction result of the grid fundus image and the image label can be calculated, and the parameters of the image segmentation model can be optimized based on the minimization of the cross entropy loss function. For example, the formula Calculate the cross entropy loss function between the prediction result of the raster fundus image and the image label. is the predicted position given by the model, and y is the actual position marked manually. It should be noted that the type of loss function is not limited here, and different types of loss functions can be calculated according to application requirements.

[0054] Finally, in step S506, an optimized image segmentation model is obtained. Thus, it is achieved that a segmentation model that can be used for ultra-wide-angle fundus images can be trained using annotated ordinary fundus images.

[0055] It should be noted that the solution disclosed herein is not limited to the processing of ultra-wide-angle fundus images, but is also compatible with the segmentation processing of ordinary fundus images.

[0056] Compared with the prior art, the disclosed solution can completely save the huge labeling cost of annotating the blood vessel distribution or lesion location of ultra-wide-angle fundus images, and the model can achieve high accuracy without the need for ultra-wide-angle image sample labeling, and the model can be used on both ordinary fundus images and ultra-wide-angle fundus images. In addition, the transfer learning method can be applied to various segmentation tasks of ultra-wide-angle fundus images (such as blood vessel extraction, lesion identification, etc.).

[0057] The disclosed embodiment also provides an electronic device. Figure 7 FIG. 8 shows an exemplary structural block diagram of an electronic device 700 according to an embodiment of the present disclosure. Figure 7 As shown, the electronic device 700 may include a processor 701 and a memory 702. The memory 702 stores computer instructions for processing ultra-wide-angle fundus images. When the computer instructions are executed by the processor 701, the electronic device 700 executes the above-mentioned Figures 2 to 5 The method described. For example, in some embodiments, the electronic device 700 can generate multiple images to be processed based on the ultra-wide-angle fundus image (for example, rasterization processing of the ultra-wide-angle fundus image), calculate rasterization parameters, use ordinary fundus images to train image segmentation models, and perform inverse rasterization processing on the segmentation results of multiple images to obtain a segmentation model of the ultra-wide-angle fundus image. Based on this, the electronic device 700 divides the ultra-wide-angle fundus image into multiple images with a resolution close to that of the ordinary fundus image, effectively reducing the data difference between the ultra-wide-angle fundus image and the ordinary fundus image, so that the segmentation model that can be used in the ultra-wide-angle fundus image can be trained using the annotated ordinary fundus image. As a result, not only the segmentation processing process of the ultra-wide-angle fundus image is simplified, but also the efficient segmentation of the ultra-wide-angle fundus image is achieved. In addition, the entire process does not rely on the label of the ultra-wide-angle fundus image, which can greatly save the annotation cost.

[0058] Although multiple embodiments of the present disclosure have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art may think of many changes, modifications, and alternatives without departing from the thought and spirit of the present disclosure. It should be understood that in the process of practicing the present disclosure, various alternatives to the embodiments of the present disclosure described herein may be adopted. The attached claims are intended to define the scope of protection of the present disclosure, and therefore cover equivalents or alternatives within the scope of these claims.

Claims

1. A method for processing ultra-wide-angle fundus images, characterized in that: The method comprises: generating, based on the ultra-wide-angle fundus image, a plurality of images to be processed whose resolution is smaller than that of the ultra-wide-angle fundus image; Processing the multiple images based on a trained image segmentation model to output segmentation results for the multiple images, wherein the image segmentation model is trained based on common fundus images; and The segmentation result of the ultra-wide-angle fundus image is determined according to the segmentation results of the multiple images.

2. The method according to claim 1, characterized in that Generating a plurality of images to be processed based on the ultra-wide-angle fundus image comprises: The ultra-wide-angle fundus image is rasterized to obtain the multiple images.

3. The method according to claim 2, characterized in that The rasterization processing of the ultra-wide-angle fundus image includes: determining a rasterization parameter according to the resolution of the ultra-wide-angle fundus image and the resolution of the image to be processed; and The ultra-wide-angle fundus image is rasterized using the rasterization parameters.

4. The method according to claim 2, characterized in that: The rasterization processing of the ultra-wide-angle fundus image includes: The ultra-wide-angle fundus image is rasterized according to preset rasterization parameters.

5. The method according to claim 1, characterized in that The image segmentation model is trained by the following operations: rasterizing the common fundus image and its corresponding segmentation label as training samples to obtain a raster fundus image and image label; and The image classification model is optimized and trained using the grid fundus image and the image label.

6. The method according to claim 5, characterized in that Optimizing and training the image segmentation model using the grid fundus image and the image label includes: Processing the grid fundus image using the image segmentation model to obtain a prediction result; Calculating a loss function between the prediction result and the image label; and The parameters of the image segmentation model are optimized based on the loss function to achieve optimized training of the image segmentation model.

7. The method according to claim 5, characterized in that The loss function includes a cross entropy loss function, and optimizing the parameters of the image segmentation model based on the loss function includes: Based on the minimization of the cross entropy loss function, the parameters of the image segmentation model are optimized.

8. The method according to any one of claims 1 to 7, characterized in that Determining the segmentation result of the ultra-wide-angle fundus image according to the segmentation results of the multiple images includes: The segmentation results of the multiple images are merged to obtain the segmentation result of the ultra-wide-angle fundus image.

9. An electronic device, characterized in that: include: processor; as well as A memory storing computer instructions for processing ultra-wide-angle fundus images, wherein when the computer instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The system comprises program instructions for processing ultra-wide-angle fundus images, and when the program instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images