Retinal thickness prediction method, regression model training method, and related devices

Abstract

This application provides a method for predicting retinal thickness, a regression model training method, and related equipment. The retinal thickness prediction method includes: acquiring a fundus image to be identified; inputting the fundus image into a preset prediction model, performing regression calculations on the fundus image, and obtaining the retinal thickness. Using the above method, by improving the training effect of the regression model and the prediction accuracy of retinal thickness, the retinal thickness of fundus images can be accurately predicted, thereby simplifying the measurement process of retinal thickness, reducing measurement costs, and increasing the application and popularity of retinal thickness monitoring.

Description

Technical Field

[0001] This application belongs to the biomedical field and relates to image processing technology, particularly to a method for predicting retinal thickness, a regression model training method, and related equipment. Prior Technology

[0002] In the field of medical technology, retinal thickness has become an important basis for doctors to diagnose diseases. Accurate measurement of retinal thickness often relies on advanced imaging techniques, such as optical coherence tomography (OCT). However, OCT equipment is expensive, making it difficult to widely adopt methods for measuring retinal thickness using OCT. Summary of the Invention

[0003] In view of the above, it is necessary to provide a method for predicting retinal thickness, a regression model training method, and related equipment, which can solve the technical problems of high cost and difficulty in widespread application of retinal thickness measurement.

[0004] On the one hand, this application provides a method for predicting retinal thickness, the method comprising: acquiring a fundus image to be identified, inputting the fundus image into a preset prediction model, performing regression calculation on the fundus image, and obtaining the retinal thickness.

[0005] In some embodiments of this application, before inputting the fundus image into a preset prediction model, the retinal thickness prediction method further includes: classifying the fundus image using a preset classification model to obtain the disease category of the fundus image, and deleting fundus images from the fundus image that do not match the preset disease category.

[0006] In some embodiments of this application, the prediction model includes a regression model and an inverse standardization model; the step of inputting the fundus image into the preset prediction model and performing regression calculation on the fundus image to obtain the retinal thickness includes: using the regression model to detect the fundus image to obtain an initial retinal thickness, and using the inverse standardization model to perform inverse standardization processing on the initial retinal thickness to obtain the retinal thickness.

[0007] In some embodiments of this application, the regression model includes a feature extraction module and a perception module. The step of using the regression model to detect the fundus image and obtain the initial retinal thickness includes: using the feature extraction module to extract features from the fundus image to obtain a feature map, and using the perception module to perform regression calculations on the feature map to obtain the initial retinal thickness corresponding to at least one region in the fundus image.

[0008] In some embodiments of this application, the sensing module includes a fully connected layer, and the number of output neurons of the fully connected layer determines the initial retinal thickness.

[0009] In some embodiments of this application, the step of using the inverse standardization model to inverse standardize the initial retinal thickness to obtain the retinal thickness includes: calculating the retinal thickness based on the standardization coefficient and the initial retinal thickness.

[0010] In some embodiments of this application, the method for determining the normalization coefficient includes: determining the normalization coefficient from retinal thickness data from real-world sources.

[0011] In some embodiments of this application, determining the normalization coefficient from retinal thickness data in real-world data includes: selecting the largest retinal thickness from the retinal thickness data in real-world data as the normalization coefficient.

[0012] In the retinal thickness prediction scheme provided in this application embodiment, the retinal thickness is predicted by inputting fundus images into the prediction model and performing regression calculations on the fundus images. This simplifies the measurement process of retinal thickness, reduces measurement costs, and increases the application popularity of retinal thickness monitoring.

[0013] On the other hand, this application provides a regression model training method, which includes: acquiring multiple fundus images as training samples, wherein each fundus image corresponds to at least one retinal thickness label; standardizing the retinal thickness label according to a preset standardization model to obtain the standard retinal thickness corresponding to the training sample; and training a preset neural network according to the training sample and the corresponding standard retinal thickness to obtain a regression model.

[0014] In some embodiments of this application, the step of standardizing the retinal thickness labels according to a preset standardization model to obtain the standard retinal thickness corresponding to the training samples includes: calculating the standard retinal thickness corresponding to each retinal thickness label based on the standardization coefficient and each retinal thickness label.

[0015] In some embodiments of this application, the preset neural network includes a feature extraction module and a perception module. The step of training the preset neural network to obtain a regression model based on the training samples and the corresponding standard retinal thickness includes: using the feature extraction module to extract features from the training samples to obtain sample feature maps; using the perception module to perform regression operations on the sample feature maps to obtain the sample retinal thickness corresponding to the training samples; calculating a loss value based on the sample retinal thickness and the corresponding standard retinal thickness; and adjusting the preset neural network based on the loss value to obtain the regression model.

[0016] In some embodiments of this application, the sensing module includes a fully connected layer, the number of output neurons of the fully connected layer corresponding to the number of retinal thickness tags corresponding to each fundus image.

[0017] In the regression model training method provided in this application embodiment, the retinal thickness label is standardized. This standardization process controls the numerical or scale range of the standardized retinal thickness, ensuring that the standard retinal thickness and the corresponding sample retinal thickness are within the same numerical or scale range. This reduces the numerical discrepancy between the standard retinal thickness and the corresponding sample retinal thickness. By reducing the numerical discrepancy between the standard retinal thickness and the model's output data, excessively large loss values ​​can be avoided, preventing gradient explosion and the regression model from converging. This ensures the training effect of the regression model and the accuracy of retinal thickness prediction.

[0018] On the other hand, this application provides an electronic device, the electronic device comprising: a memory for storing at least one instruction; and a processor for executing the at least one instruction to implement the retinal thickness prediction method or the regression model training method.

[0019] On the other hand, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the retinal thickness prediction method or the regression model training method. Simple Explanation of the Diagram

[0020] Figure 1 is a flowchart of a regression model training method provided in an embodiment of this application.

[0021] Figure 2 is a schematic diagram of a fundus tomographic image of normal retinal thickness provided in an embodiment of this application.

[0022] Figure 3 is a schematic diagram of a fundus tomographic image of abnormal retinal thickness provided in an embodiment of this application.

[0023] Figure 4 is a schematic diagram of region division in a fundus image provided in an embodiment of this application.

[0024] Figure 5 is a schematic diagram of regression model training provided in an embodiment of this application.

[0025] Figure 6 is a flowchart of a retinal thickness prediction method provided in an embodiment of this application.

[0026] Figure 7 is a schematic diagram of retinal thickness prediction provided in an embodiment of this application.

[0027] Figure 8 is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Implementation

[0028] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, scope of the invention application, and drawings of this application are used to distinguish similar objects, not to describe a specific order or sequence.

[0029] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner. Unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0030] This application provides a regression model training method and a retinal thickness prediction method. By improving the training effect of the regression model and the prediction accuracy of retinal thickness, the retinal thickness of fundus images can be accurately predicted, thereby simplifying the measurement process of retinal thickness, reducing measurement costs, and increasing the application popularity of retinal thickness monitoring.

[0031] Figure 1 shows a flowchart of a regression model training method provided in an embodiment of this application. Depending on different needs, the order of the steps in this flowchart can be adjusted according to actual requirements, and some steps can be omitted. The regression model training method is applied to electronic devices, such as the electronic device 10 shown in Figure 8.

[0032] Step S11: Obtain multiple fundus images as training samples, wherein each fundus image corresponds to at least one retinal thickness label.

[0033] In some embodiments of this application, fundus images can be obtained by taking pictures using devices such as fundus cameras. For example, considering that diseases such as age-related macular degeneration (AMD) and diabetic retinopathy (DR) may cause macular edema, which may lead to abnormal retinal thickness, in order for the trained regression model to accurately predict retinal thickness, the fundus images in the training samples can be fundus images corresponding to diseases such as macular degeneration and diabetic retinopathy.

[0034] Abnormal retinal thickness can refer to a situation where the central retinal thickness (CRT) exceeds a preset thickness threshold. Figure 2 shows a schematic diagram of a fundus tomographic image of normal retinal thickness according to an embodiment of this application. Figure 3 shows a schematic diagram of a fundus tomographic image of abnormal retinal thickness according to an embodiment of this application. Compared to the central retinal center in the fundus image of Figure 2, the central retinal center in the fundus image of Figure 3 bulges upwards, indicating abnormal thickness.

[0035] Retinal thickness labels can be the actual retinal thickness obtained through equipment such as optical coherence tomography (OCT), or the retinal thickness in real-world data (RWD) such as multi-source heterogeneous data from medical systems, insurance records, and patient reports.

[0036] Each fundus image in the training samples can have one or more corresponding retinal thickness labels. For example, each fundus image can correspond to the same retinal thickness label, or the area around the optic disc in the fundus image can be divided into multiple regions, each with a corresponding retinal thickness label. The division of the eyeball in the fundus image can be flexibly set. For example, Figure 4 shows a schematic diagram of the region division in a fundus image according to an embodiment of this application. In Figure 4, the area around the optic disc in the fundus image is divided into 9 regions. Referring to the region represented by each digit in Figure 4, the digit in each region indicates the corresponding retinal thickness, such as a central retinal thickness of 419.

[0037] Considering that the retinal thickness may vary in different regions of the eye, in this embodiment, by using retinal thickness labels from multiple regions in the fundus image for training, the trained regression model can accurately predict the retinal thickness of the corresponding region.

[0038] Step S12: Standardize the retinal thickness label according to the preset standardization model to obtain the standard retinal thickness corresponding to the training sample.

[0039] In some embodiments of this application, the electronic device performs standardization processing on the retinal thickness labels according to a preset standardization model to obtain the standard retinal thickness corresponding to the training samples, including: calculating the standard retinal thickness corresponding to each retinal thickness label based on the standardization coefficient and each retinal thickness label.

[0040] For example, the ratio between each retinal thickness label and the standardization coefficient can be used as the standard retinal thickness corresponding to each retinal thickness label.

[0041] The standardization factor can be determined from all retinal thickness labels of the training samples. For example, the largest retinal thickness can be selected as the standardization factor from all collected true retinal thicknesses or retinal thicknesses in real-world data (RWD).

[0042] Considering that if the difference between the output data and the input data of the regression model is too large, the gradient explosion phenomenon may occur due to the excessive loss value, which may cause the model to fail to converge and make it impossible for the model to accurately detect retinal thickness, this embodiment can control the scale or numerical range corresponding to the standard retinal thickness by selecting a standardized coefficient from the retinal thickness labels and using the ratio between each retinal thickness label and the standardized coefficient as the corresponding standard retinal thickness.

[0043] Step S13: Train the preset neural network based on the training samples and the corresponding standard retinal thickness to obtain the regression model.

[0044] In some embodiments of this application, the preset neural network includes a feature extraction module and a perception module. The electronic device trains the preset neural network based on training samples and corresponding standard retinal thicknesses to obtain a regression model, including: using the feature extraction module to extract features from the training samples to obtain sample feature maps; using the perception module to perform regression operations on the sample feature maps to obtain the sample retinal thicknesses corresponding to the training samples; calculating loss values ​​based on the sample retinal thicknesses and corresponding standard retinal thicknesses; and adjusting the preset neural network based on the loss values ​​to obtain the regression model.

[0045] The feature extraction module can be a convolutional neural network, such as the feature extraction modules in networks like RestNet, GoogleNet, VGG, EfficientNet, and Long Short-Term Memory (LSTM). The regression operation can be a linear regression operation.

[0046] Before using the sensing module to perform regression calculations on the sample feature map, the sample feature map can be flattened / vectorized first, and then the vector obtained by vectorization can be input into the sensing module for regression calculations to obtain the sample retinal thickness.

[0047] The perception module may include a fully connected layer, the number of output neurons in which corresponds to the number of retinal thickness tags for each fundus image. For example, if there is one retinal thickness tag for each fundus image, the fully connected layer has one output neuron; if there are nine retinal thickness tags for each fundus image, the fully connected layer has nine output neurons. The number of output neurons in the fully connected layer can be flexibly adjusted according to the number of retinal thickness tags for each fundus image.

[0048] The number of sample retinal thicknesses corresponding to each fundus image in the training samples corresponds to the number of output neurons in the fully connected layer, thus obtaining sample retinal thicknesses that correspond one-to-one with the retinal thickness labels, facilitating the calculation of the loss value. For example, if each fundus image corresponds to 1 retinal thickness label, then the number of output neurons in the fully connected layer is 1; if each fundus image corresponds to 9 retinal thickness labels, then the number of output neurons in the fully connected layer is 9.

[0049] In some embodiments, the data output by the fully connected layer of the sensing module can be directly used as the sample retinal thickness.

[0050] The electronic device can calculate the loss value based on the retinal thickness of each sample and the corresponding standard retinal thickness using a loss function. This application does not restrict the type of loss function. For example, the loss function can be cross-entropy loss or binary cross-entropy loss.

[0051] In this embodiment, the retinal thickness label is standardized using a standardization coefficient. This controls the numerical or scale range of the standardized retinal thickness, ensuring that the standard retinal thickness and the corresponding sample retinal thickness are within the same numerical or scale range. This reduces the numerical discrepancy between the standard and sample retinal thicknesses. By minimizing this discrepancy, excessive loss values ​​can be avoided, preventing gradient explosion and the regression model from converging. This ensures the training effectiveness of the regression model and the accuracy of its retinal thickness prediction.

[0052] Figure 5 illustrates a schematic diagram of regression model training according to an embodiment of this application. In Figure 5, retinal thickness labels are input into a standardized model for standardization to obtain standard retinal thickness. Fundus images with macular degeneration and diabetic retinopathy are used as training samples and input into a feature extraction module to obtain sample feature maps. These sample feature maps are then input into a fully connected layer in a perception module to obtain sample retinal thickness. A loss value is calculated based on the sample retinal thickness and the standard retinal thickness, and the feature extraction module and perception module are adjusted according to the loss value.

[0053] In the regression model training method provided in this application embodiment, the retinal thickness label is standardized. This standardization process controls the numerical or scale range of the standardized retinal thickness, ensuring that the standard retinal thickness and the corresponding sample retinal thickness are within the same numerical or scale range. This reduces the numerical discrepancy between the standard retinal thickness and the corresponding sample retinal thickness. By reducing the numerical discrepancy between the standard retinal thickness and the model's output data, excessively large loss values ​​can be avoided, preventing gradient explosion and the regression model from converging. This ensures the training effect of the regression model and the accuracy of retinal thickness prediction.

[0054] Following the regression model training method shown in Figure 1, after obtaining the regression model, it can be used to predict the fundus image to be identified. Figure 6 shows a flowchart of a retinal thickness prediction method provided in an embodiment of this application. Depending on different needs, the order of each step in the flowchart can be adjusted according to actual requirements, and some steps can be omitted. The retinal thickness prediction method is applied to electronic devices, such as the electronic device 10 shown in Figure 8.

[0055] Step S21: Obtain the fundus image to be identified.

[0056] In some embodiments of this application, before inputting the fundus image into a preset prediction model, the electronic device can use a preset classification model to classify the fundus image, obtain the disease category of the fundus image, and delete fundus images corresponding to disease categories that do not match the preset disease categories.

[0057] The classification model can be obtained by training convolutional neural networks, including but not limited to: RestNet, GoogleNet, VGG, and EfficientNet. Preset disease categories can include macular degeneration and diabetic retinopathy. Disease categories that do not match the preset disease categories can be different from the preset disease categories. For example, only fundus images corresponding to macular degeneration and diabetic retinopathy can be retained.

[0058] Step S22: Input the fundus image into the preset prediction model, perform regression calculation on the fundus image, and obtain the retinal thickness.

[0059] In some embodiments of this application, the prediction model includes a regression model and an inverse standardization model, wherein the regression model includes a feature extraction module and a perception module. The feature extraction module can be a convolutional neural network, such as the feature extraction modules in networks like RestNet, GoogleNet, VGG, EfficientNet, and Long Short-Term Memory (LSTM). The perception module can include fully connected layers. The training method for the regression model can be referred to the description of the training method shown in Figure 1 or Figure 5, and will not be repeated in this embodiment.

[0060] In some embodiments of this application, the electronic device inputs a fundus image into a preset prediction model, performs regression calculations on the fundus image, and obtains the retinal thickness by: using the regression model to detect the fundus image and obtain an initial retinal thickness, and using an inverse standardization model to inverse standardize the initial retinal thickness to obtain the retinal thickness.

[0061] For example, an electronic device uses a regression model to detect fundus images and obtain initial retinal thickness by: extracting features from the fundus image using a feature extraction module to obtain a feature map, and performing regression calculations on the feature map using a perception module to obtain the initial retinal thickness corresponding to at least one region in the fundus image.

[0062] The initial retinal thickness is determined by the number of output neurons in the fully connected layer. For explanations of the feature extraction module and the perception module, please refer to step S13. The calculation method for the initial retinal thickness can be found in the explanation of the sample retinal thickness calculation method in step S13.

[0063] In this embodiment, since the regression model is obtained through training, it is possible to accurately detect fundus images, thereby ensuring the accuracy of the initial retinal thickness prediction.

[0064] For example, an electronic device uses an inverse normalization model to inverse normalize the initial retinal thickness to obtain the retinal thickness, including: calculating the retinal thickness based on the normalization coefficient and the initial retinal thickness.

[0065] Denormalization is the inverse operation of normalization. For example, the product or weighted product of the normalization coefficients and the initial retinal thickness can be used as the retinal thickness. The weights used for weighting can be customized, and this application does not impose any restrictions on them. The method for determining the normalization coefficients can be found in the description of step S12.

[0066] Considering that the regression model is trained using a standardized retinal thickness, the initial retinal thickness output by the regression model is a value under a standard scale, which cannot reflect the true retinal thickness, in this embodiment, the initial retinal thickness is de-standardized by a standard coefficient, which can restore the initial retinal thickness to the original scale, so that the retinal thickness can accurately reflect the true retinal thickness.

[0067] Figure 7 is a schematic diagram of retinal thickness prediction provided in an embodiment of this application. In Figure 7, the fundus image to be identified, which has macular degeneration and diabetic retinopathy, is input into the feature extraction module of the regression model of the prediction model to obtain a feature map. The feature map is then input into the fully connected layer of the perception module in the regression model to obtain the initial retinal thickness. Finally, the initial retinal thickness is input into the inverse normalization model of the prediction model to obtain the final retinal thickness.

[0068] In the retinal thickness prediction scheme provided in this application embodiment, since the regression model is obtained through training, it can accurately detect fundus images, thereby ensuring the accuracy of the initial retinal thickness. Considering that the regression model is trained using a standardized retinal thickness, the initial retinal thickness output by the regression model is a value at a standard scale and cannot reflect the true retinal thickness. Therefore, destandardization is performed on the initial retinal thickness to obtain the predicted retinal thickness, restoring the initial retinal thickness to the original scale, thus ensuring the accuracy of the predicted retinal thickness. This simplifies the retinal thickness measurement process, reduces measurement costs, and increases the application and popularity of retinal thickness monitoring.

[0069] Figure 8 shows a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 10 can be a computer, tablet computer, laptop computer, mobile phone, server, or other electronic devices. This application embodiment does not impose any restrictions on the specific type of electronic device.

[0070] In Figure 8, the electronic device 10 may include a communication module 101, a memory 102, a processor 103, an input / output (I / O) interface 104, and a bus 105. The processor 103 is coupled to the communication module 101, the memory 102, and the input / output interface 104 through the bus 105.

[0071] Communication module 101 may include a wired communication module and / or a wireless communication module. The wired communication module may provide one or more wired communication solutions such as Universal Serial Bus (USB) and Controller Area Network (CAN). The wireless communication module may provide one or more wireless communication solutions such as Wireless Fidelity (Wi-Fi), Bluetooth (BT), mobile communication networks, frequency modulation (FM), near field communication (NFC), and infrared (IR).

[0072] Memory 102 may include one or more random access memory (RAM) and one or more non-volatile memory (NVM). The RAM can be directly read and written by the processor 103, and can be used to store executable programs (e.g., machine instructions) of other running programs, as well as user and application data. The RAM may include static random-access memory (SRAM), dynamic random-access memory (DRAM), synchronous dynamic random-access memory (SDRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.

[0073] Non-volatile memory can also store executable programs and user and application data, and can be pre-loaded into random access memory for direct reading and writing by the processor 103. Non-volatile memory can include disk memory and flash memory. For example, flash memory can be NAND flash.

[0074] Memory 102 is used to store one or more computer programs. The one or more computer programs are configured to be executed by processor 103. The one or more computer programs include multiple instructions that, when executed by processor 103, can implement a regression model training method or a retinal thickness prediction method that is executed on electronic device 10.

[0075] In other embodiments, the electronic device 10 shown in FIG8 also includes an external memory interface for connecting external memory to expand the storage capacity of the electronic device 10.

[0076] Processor 103 may include one or more processing units, such as: application processor (AP), modem processor, graphics processing unit (GPU), image signal processor (ISP), controller, video transcoder, digital signal processor (DSP), and / or neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0077] The processor 103 provides computing and control capabilities. For example, the processor 103 is used to execute computer programs stored in memory 102 to implement the regression model training method or retinal thickness prediction method described above.

[0078] The input / output interface 104 is used to provide a channel for user input or output. For example, the input / output interface 104 can be used to connect various input / output devices, such as mice, keyboards, touch devices, and display screens, so that users can enter information or visualize information.

[0079] Bus 105 is used to provide at least a channel for communication between communication modules 101, memory 102, processor 103, and input / output interface 104 in electronic device 10.

[0080] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device 10. In other embodiments of this application, the electronic device 10 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of hardware and software.

[0081] This application also provides a computer-readable storage medium storing a computer program, which includes program instructions. When the program instructions are executed, the method implemented can refer to the methods in the above embodiments of this application.

[0082] The computer-readable storage medium can be the internal memory of the electronic device described in the above embodiments, such as the hard drive or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card.

[0083] In some embodiments, a computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, applications required for at least one function, etc.; and the data storage area may store data created based on the use of the electronic device, etc.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0085] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0087] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be incorporated herein by reference. No appended diagram markings in the claims should be construed as limiting the scope of the claims involved.

[0088] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices described in this application may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

[0090] 10: Electronic devices 101: Communication Module 102: Memory 103: Processor 104: Input / Output Interface 105: Busbar S11~S13: Steps S21~S22: Steps

Claims

1. A method for predicting retinal thickness, executed in an electronic device, wherein, The retinal thickness prediction method includes: acquiring a fundus image to be identified, wherein the fundus image is obtained by taking a picture through a fundus camera; inputting the fundus image into a preset prediction model, performing regression calculation on the fundus image, and obtaining the retinal thickness.

2. The retinal thickness prediction method as described in claim 1, wherein, Before inputting the fundus image into a preset prediction model, the retinal thickness prediction method further includes: classifying the fundus image using a preset classification model to obtain the disease category of the fundus image; and deleting fundus images from the fundus image that do not match the preset disease category.

3. The retinal thickness prediction method as described in claim 1, wherein, The prediction model includes a regression model and an inverse standardization model; the step of inputting the fundus image into the preset prediction model and performing regression calculation on the fundus image to obtain the retinal thickness includes: using the regression model to detect the fundus image and obtain an initial retinal thickness; using the inverse standardization model to inverse standardize the initial retinal thickness and obtain the retinal thickness.

4. The retinal thickness prediction method as described in claim 3, wherein, The regression model includes a feature extraction module and a perception module. The step of using the regression model to detect the fundus image and obtain the initial retinal thickness includes: using the feature extraction module to extract features from the fundus image to obtain a feature map; and using the perception module to perform regression calculations on the feature map to obtain the initial retinal thickness corresponding to at least one region in the fundus image.

5. The retinal thickness prediction method as described in claim 4, wherein, The sensing module includes a fully connected layer, and the initial retinal thickness is determined by the number of output neurons of the fully connected layer.

6. The retinal thickness prediction method as described in claim 3, wherein, The step of using the inverse standardization model to inverse standardize the initial retinal thickness to obtain the retinal thickness includes: calculating the retinal thickness based on the standardization coefficient and the initial retinal thickness.

7. The retinal thickness prediction method as described in claim 6, wherein, The method for determining the standardized coefficients includes: determining the standardized coefficients from retinal thickness data from real-world sources.

8. The retinal thickness prediction method as described in claim 7, wherein, Determining the standardized coefficient from real-world retinal thickness data includes: selecting the largest retinal thickness from the real-world retinal thickness data as the standardized coefficient.

9. A regression model training method, executed on an electronic device, wherein, The regression model training method includes: acquiring multiple fundus images as training samples, wherein each fundus image corresponds to at least one retinal thickness label, and the multiple fundus images are obtained by taking pictures through a fundus camera; standardizing the retinal thickness label according to a preset standardization model to obtain the standard retinal thickness corresponding to the training sample; and training a preset neural network according to the training sample and the corresponding standard retinal thickness to obtain a regression model.

10. The regression model training method as described in claim 9, wherein, The process of standardizing the retinal thickness labels according to a preset standardization model to obtain the standard retinal thickness corresponding to the training samples includes: calculating the standard retinal thickness corresponding to each retinal thickness label based on the standardization coefficient and each retinal thickness label.

11. The regression model training method as described in claim 9, wherein, The preset neural network includes a feature extraction module and a perception module. Training the preset neural network to obtain a regression model based on the training samples and the corresponding standard retinal thickness includes: extracting features from the training samples using the feature extraction module to obtain a sample feature map; performing regression calculations on the sample feature map using the perception module to obtain the sample retinal thickness corresponding to the training samples; calculating a loss value based on the sample retinal thickness and the corresponding standard retinal thickness; and adjusting the preset neural network based on the loss value to obtain the regression model.

12. The regression model training method as described in request item 11, wherein, The perception module includes a fully connected layer, and the number of output neurons in the fully connected layer corresponds to the number of retinal thickness tags for each fundus image.

13. An electronic device, wherein, include: Memory stores at least one instruction; and a processor, executing the at least one instruction to implement the retinal thickness prediction method as described in any one of claims 1 to 8 or the regression model training method as described in any one of claims 9 to 12.

14. A computer-readable storage medium, wherein, The computer-readable storage medium stores at least one instruction, which is executed by a processor in an electronic device to implement the retinal thickness prediction method as described in any one of claims 1 to 8 or the regression model training method as described in any one of claims 9 to 12.

Images