Retina image scoring method and device, equipment and medium
By identifying key lesions and global feature scores in retinal images and combining them with Dempster-Shafer synthesis rules, the problem of inaccurate lesion grade classification in existing technologies has been solved, achieving refined scoring and improved accuracy of retinal lesion severity.
Patent Information
- Application Number
- CN202510963747.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-14
- Publication Date
- 2025-11-14
AI Technical Summary
Existing AI diagnostic models can only roughly classify lesions in fundus images, failing to accurately reflect the different degrees of lesions within the same grade, resulting in low accuracy in lesion scoring.
By identifying key lesions in retinal images, statistically analyzing their index data, evaluating lesion scores, and combining global feature extraction and Dempster-Shafer synthesis rules, a total retinal score is calculated to refine the characterization of lesion severity.
It enables refined scoring of the severity of retinal lesions, improving the accuracy of lesion scoring and the ability to perform fine-grained analysis.
Smart Images
Figure CN120953178A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, specifically to a method, apparatus, device, and medium for scoring retinal images. Background Technology
[0002] Collecting fundus images of the human eye and applying artificial intelligence diagnostic models to these images to diagnose diabetic retinopathy is a method. However, existing artificial intelligence diagnostic models can only roughly classify the lesion level in fundus images, and the same level of lesion includes different degrees of lesion. Therefore, classifying the lesion level in images alone cannot accurately reflect the true degree of lesion.
[0003] In summary, the accuracy of lesion scoring based on lesion grade in existing technologies is relatively low.
[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a method, apparatus, device, and medium for scoring retinal images, which solves the problem of low accuracy in lesion scoring based on lesion grade in the prior art.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] In a first aspect, the present invention provides a method for scoring retinal images, comprising:
[0008] The process involves acquiring an image of the retina, identifying key lesions in the image, statistically analyzing the index data of the key lesions based on the image, evaluating the key lesions based on the index data, and obtaining a lesion score for the image. The index data is used to characterize the severity of the lesions in the key lesions.
[0009] Extract global features from the image, evaluate the global features, and obtain a feature score for the image;
[0010] The total score of the retina is obtained based on the lesion score and the feature score of the image.
[0011] In one implementation, identifying key lesions in the image and statistically analyzing indicator data of the key lesions based on the image includes:
[0012] Identify microaneurysms, hemorrhage points, hard exudates, soft exudates, and vascular proliferation on the image, and identify the microaneurysms, hemorrhage points, hard exudates, soft exudates, and vascular proliferation as key lesions;
[0013] Based on the images, the number of key lesions, the total area of the key lesions, the ratio of the total area to the retinal area, and the maximum area of the key lesions are statistically analyzed, and the number of key lesions, the total area of the key lesions, the ratio of the total area to the retinal area, and the maximum area of the key lesions are used as indicator data for the key lesions.
[0014] In one implementation, the key lesions are evaluated based on the indicator data to obtain a lesion score, including:
[0015] The lesion score is obtained by weighting the index data of the microaneurysm, the index data of the bleeding point, the index data of the hard exudate, the index data of the soft exudate, and the index data of the vascular proliferation.
[0016] In one implementation, global features of the image are extracted, and the global features are evaluated to obtain a feature score, including:
[0017] Extract global features related to the degree of retinal disease from the image;
[0018] Based on the feature values of the global features corresponding to each of the retinal lesion levels, the global features corresponding to all of the retinal lesion levels are evaluated to obtain a feature score.
[0019] In one implementation, a total score for the retina is obtained based on the lesion score and the feature score of the image, including:
[0020] Each view in the image is determined, and each view is a view taken from a different angle of the retina;
[0021] The total score of the retina is obtained based on the lesion score and feature score of each of the views.
[0022] In one implementation, a total score for the retina is obtained based on the lesion score and feature score of each of the views, including:
[0023] The lesion scores and feature scores of each of the views are combined into a score vector;
[0024] Applying the Dempster-Shafer synthesis rule to the score vectors of each of the views yields joint evidence quality in vector form;
[0025] The total score of the retina is obtained based on the values of each element in the joint evidence quality in vector form.
[0026] In one implementation, the lesion grade of the retina is determined based on the total score.
[0027] Secondly, embodiments of the present invention also provide a scoring device for retinal images, wherein the device comprises the following components:
[0028] The lesion scoring module is used to acquire images of the retina, identify key lesions in the images, statistically analyze the index data of the key lesions based on the images, and evaluate the key lesions based on the index data to obtain a lesion score for the images. The index data is used to characterize the degree of lesion of the key lesions.
[0029] The feature scoring module is used to extract global features of the image, evaluate the global features, and obtain a feature score for the image.
[0030] The total score module is used to obtain the total score of the retina based on the lesion score and the feature score of the image.
[0031] Thirdly, embodiments of the present invention also provide a terminal device, wherein the terminal device includes a memory, a processor, and a retinal image scoring program stored in the memory and executable on the processor, wherein when the processor executes the retinal image scoring program, it implements the steps of the retinal image scoring method described above.
[0032] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a retinal image scoring program, wherein when the retinal image scoring program is executed by a processor, the steps of the retinal image scoring method described above are implemented.
[0033] Beneficial effects: This invention extracts key lesions and their indicator data from retinal images, evaluates the key lesions based on the indicator data to obtain lesion scores, and simultaneously evaluates the global features of the image to obtain feature scores. Finally, a total retinal score is obtained based on the lesion scores and feature scores. Because this invention calculates the total retinal score using quantitative data such as indicator data and global features, the total retinal score can accurately characterize the degree of retinal lesions. Attached Figure Description
[0034] Figure 1 This is an overall flowchart of the present invention;
[0035] Figure 2 This is a schematic diagram of fine-grained scoring in an embodiment of the present invention;
[0036] Figure 3 A structural diagram of the retinal image scoring device provided by the present invention;
[0037] Figure 4This is a block diagram illustrating the internal structure of a terminal device provided in an embodiment of the present invention. Detailed Implementation
[0038] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0039] Research has found that collecting fundus images of the human eye and applying artificial intelligence diagnostic models to these images to diagnose diabetic retinopathy is feasible. However, existing artificial intelligence diagnostic models can only roughly classify the lesion level in fundus images, and the same level of lesion includes different degrees of lesion. Therefore, classifying the lesion level in images alone cannot accurately reflect the true degree of lesion.
[0040] To address the aforementioned technical problems, this invention provides a method, apparatus, device, and medium for scoring retinal images, solving the problem of low accuracy in existing lesion scoring based on lesion severity. Specifically, the invention first identifies key lesions in the retinal image, statistically analyzes the key lesion indicators based on the image, and evaluates the key lesions based on the indicator data to obtain a lesion score for the image. Simultaneously, global features of the image are extracted and evaluated to obtain a feature score for the image. Finally, based on the lesion score and the feature score of the image, a total retinal score is obtained.
[0041] For example, the retinal image includes four views taken from four different angles: a view covering the macula, a view of the optic disc center, a view of the upper rim region, and a view of the lower rim region. From the view covering the macula, five key lesions are identified: microaneurysms, hemorrhages, hard exudates, soft exudates, and vascular proliferation. The index data for each key lesion are statistically analyzed, and a score for each key lesion is calculated based on the index data. The lesion score for each view is obtained by summing the scores corresponding to all lesions in each view.
[0042] Global feature extraction is performed on each of the four views, and the global features of each view are evaluated to obtain a feature score for each view. Finally, based on the lesion scores of all views and the feature scores of all views, a total retinal score is obtained. This total score is used to refine the characterization of the degree of retinal disease.
[0043] The retinal image scoring method of this embodiment can be applied to a terminal device, which can be a terminal product with image processing capabilities, such as a computer. In this embodiment, as... Figure 1 As shown, the scoring method for the aforementioned retinal images specifically includes the following steps:
[0044] S100: Acquire an image of the retina, identify key lesions in the image, statistically analyze the index data of the key lesions based on the image, and evaluate the key lesions based on the index data to obtain a lesion score for the image. The index data is used to characterize the degree of lesion of the key lesions.
[0045] S200, extract the global features of the image, evaluate the global features, and obtain the feature score of the image.
[0046] S300, based on the lesion score and feature score of the image, the total score of the retina is obtained.
[0047] In this embodiment, the retinal images include a view covering the macula, a view of the optic disc center, a view of the upper rim region, and a view of the lower rim region. The macula is located at the center of the retina, and the view covering the macula is the image covering the center of the retina. The optic disc center is the concave area where the optic nerve passes through the retina, and the view of the optic disc center is the image of this concave area. The upper rim region is the upper half of the retina, and the view of the upper rim region is the acquired image of the upper half of the retina. The lower rim region is the lower half of the retina, and the view of the lower rim region is the acquired image of the lower half of the retina. Steps S100 and S200 are performed for each view to obtain lesion scores and feature scores for all views. Then, step S300 is performed on the lesion scores and feature scores for all views to obtain the total retinal score.
[0048] In this embodiment, evidence quality assessment theory is introduced to obtain a fine-grained score for the retina. This involves segmenting key lesion regions at the pixel level from the retinal image, extracting key lesion evidence (i.e., the index data of key lesions), and simultaneously using a feature extraction network to extract global feature evidence from the retinal image. The quality of evidence is then evaluated using a lesion scoring standard established by the physician to obtain fine-grained scores for both the lesions and the global features. Finally, the Dempster-Shafer synthesis rule is employed, combined with evidence information from multiple views, to achieve a more objective decision-making process, resulting in accurate fine-grained scores and interpretable quantitative analysis results for the lesions.
[0049] like Figure 2 As shown, in step S100, each retinal image is processed by an encoder and a decoder (both encoders and decoders are existing technologies) to extract key lesions (extracting key lesions is... Figure 2 The process involves extracting lesion evidence (from the evidence extraction process), then scoring the key lesions on the quality of evidence (i.e., evaluating the score of the key lesions based on their indicator data), and finally obtaining the lesion score for the view based on this score.
[0050]
[0051] In the formula, This represents the indicator vector corresponding to the k-th key lesion in the v-th view. The indicator vector consists of several indicator data. represent In the model, K represents the total number of key lesion types. p represents the weight of the k-th critical lesion in the v-th view. v α represents the lesion score in the v-th view, and α represents the bias.
[0052] When training the lesion extraction model, p v Doctor rating y v By comparing and adjusting the model parameters, p v Approaching y v This is to complete the training of the model.
[0053] like Figure 2 As shown, a feature extraction network is applied to any retinal image in step S100 simultaneously to extract global features, and the evidence quality assessment of the global features is performed to obtain the feature score of the view.
[0054] Step S100: Key lesions include microaneurysms, hemorrhages, hard effusions, soft effusions, and vascular proliferation, as shown in Table 1. Key data indicators include the number of key lesions, the total area of key lesions, the ratio of the total area to the retinal area, and the maximum area of the key lesion, as shown in Table 1. For any key lesion, such as a microaneurysm, the number of key lesions refers to the number of microaneurysms on the retinal image; the total area refers to the area occupied by all microaneurysms; and the maximum area of the key lesion is the largest of the areas of all microaneurysms.
[0055] Table 1
[0056]
[0057]
[0058] In this embodiment, step S200 includes the following specific steps: extracting global features related to the retinal lesion level from the image; evaluating the global features corresponding to all retinal lesion levels based on the feature values of the global features corresponding to each retinal lesion level to obtain a feature score.
[0059] like Figure 2 As shown, a feature extraction network (which is a prior art) is applied to each view contained in the image to obtain the global features corresponding to each view. The global features are then evaluated to obtain a feature score.
[0060]
[0061] In the formula, p′v represents the feature score of the v-th view. The feature value represents the global feature related to the i-th retinal lesion level on the v-th view, and C is the total number of retinal lesion levels, such as retinal lesion levels including stages I, II, III, IV, V, and VI, which are progressively more severe. β represents the weight of the i-th retinal lesion level in the v-th view. β represents the bias.
[0062] In this embodiment, step S300 includes the following specific steps S301, S302, S303, and S304:
[0063] S301, determine the individual views in the image, each of the views being a view taken from a different angle of the retina.
[0064] Each view includes at least the view covering the macula, the view of the center of the optic disc, the view of the upper edge area, and the view of the lower edge area.
[0065] S302, assign lesion scores p to each of the aforementioned views. v The feature score p′v is combined to form the score vector U. v :
[0066]
[0067] S303, apply the Dempster-Shafer synthesis rule to the score vectors of each of the views to obtain the joint evidence quality U in vector form.
[0068]
[0069] Among them, U 1 U represents the rating vector for the first view. 1 ={p 1 ,p′ 1}, U 2 U represents the rating vector for the second view. 2 ={p 2 ,p′ 2}, where V represents the total number of views. This represents the Dempster-Shafer composition rule, which is a data fusion algorithm. That is, calculate first Assume the calculation result is U 12 Then calculate This process is repeated to obtain the final quality U of the joint evidence.
[0070] S304, based on the values of each element in the joint evidence quality in vector form, the total score S of the retina is obtained.
[0071] Use u1, u2,...,u n Let each element in the joint evidence quality U in vector form represent the value of the total score S:
[0072]
[0073] Below, U 1 ={p 1 ,p′ 1}、U 2 ={p 2 ,p′ 2 For example, let's explain U. 1 and U 2 The joint evidence quality U of the two, where the element value of U is u:
[0074]
[0075] In the formula, D is the metric U 1 and U 2 The amount of conflict between them.
[0076] This embodiment yields a total score S, which is used to classify the degree of retinal lesions according to Table 2.
[0077] Table 2
[0078]
[0079] This embodiment also provides a retinal image scoring device, such as... Figure 3 As shown, the device comprises the following components:
[0080] The lesion scoring module 01 is used to acquire images of the retina, identify key lesions in the images, statistically analyze the index data of the key lesions based on the images, and evaluate the key lesions based on the index data to obtain a lesion score for the images. The index data is used to characterize the degree of lesion of the key lesions.
[0081] Feature scoring module 02 is used to extract global features of the image, evaluate the global features, and obtain a feature score of the image;
[0082] The total score module 03 is used to obtain the total score of the retina based on the lesion score and the feature score of the image.
[0083] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 4 As shown, the terminal device includes a processor, memory, network interface, and display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a retinal image scoring method. The display screen can be a liquid crystal display (LCD) or an e-ink display.
[0084] Those skilled in the art will understand that Figure 4 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.
[0085] In one embodiment, a terminal device is provided, the terminal device including a memory, a processor, and a retinal image scoring program stored in the memory and executable on the processor. When the processor executes the retinal image scoring program, it implements the following operation instructions:
[0086] The process involves acquiring an image of the retina, identifying key lesions in the image, statistically analyzing the index data of the key lesions based on the image, evaluating the key lesions based on the index data, and obtaining a lesion score for the image. The index data is used to characterize the severity of the lesions in the key lesions.
[0087] Extract global features from the image, evaluate the global features, and obtain a feature score for the image;
[0088] The total score of the retina is obtained based on the lesion score and the feature score of the image.
[0089] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.
[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for scoring retinal images, characterized in that, include: The process involves acquiring an image of the retina, identifying key lesions in the image, statistically analyzing the index data of the key lesions based on the image, evaluating the key lesions based on the index data, and obtaining a lesion score for the image. The index data is used to characterize the severity of the lesions in the key lesions. Extract global features from the image, evaluate the global features, and obtain a feature score for the image; The total score of the retina is obtained based on the lesion score and the feature score of the image.
2. The retinal image scoring method as described in claim 1, characterized in that, Identifying key lesions in the image and statistically analyzing indicator data of the key lesions based on the image, including: Identify microaneurysms, hemorrhage points, hard exudates, soft exudates, and vascular proliferation on the image, and identify the microaneurysms, hemorrhage points, hard exudates, soft exudates, and vascular proliferation as key lesions; Based on the images, the number of key lesions, the total area of the key lesions, the ratio of the total area to the retinal area, and the maximum area of the key lesions are statistically analyzed, and the number of key lesions, the total area of the key lesions, the ratio of the total area to the retinal area, and the maximum area of the key lesions are used as indicator data for the key lesions.
3. The retinal image scoring method as described in claim 2, characterized in that, The key lesions are evaluated based on the aforementioned indicator data to obtain lesion scores, including: The lesion score is obtained by weighting the index data of the microaneurysm, the index data of the bleeding point, the index data of the hard exudate, the index data of the soft exudate, and the index data of the vascular proliferation.
4. The retinal image scoring method as described in claim 1, characterized in that, Extract global features from the image, evaluate the global features, and obtain a feature score, including: Extract global features related to the degree of retinal disease from the image; Based on the feature values of the global features corresponding to each of the retinal lesion levels, the global features corresponding to all of the retinal lesion levels are evaluated to obtain a feature score.
5. The retinal image scoring method as described in claim 1, characterized in that, Based on the lesion score and feature score of the image, a total score for the retina is obtained, including: Each view in the image is determined, and each view is a view taken from a different angle of the retina; The total score of the retina is obtained based on the lesion score and feature score of each of the views.
6. The retinal image scoring method as described in claim 5, characterized in that, Based on the lesion score and feature score of each of the views, the total score of the retina is obtained, including: The lesion scores and feature scores of each of the views are combined into a score vector; Applying the Dempster-Shafer synthesis rule to the score vectors of each of the views yields joint evidence quality in vector form; The total score of the retina is obtained based on the values of each element in the joint evidence quality in vector form.
7. The method for scoring retinal images as described in any one of claims 1-6, characterized in that, Also includes: The degree of retinal lesion is determined based on the total score.
8. A scoring device for retinal images, characterized in that, The device comprises the following components: The lesion scoring module is used to acquire images of the retina, identify key lesions in the images, statistically analyze the index data of the key lesions based on the images, and evaluate the key lesions based on the index data to obtain a lesion score for the images. The index data is used to characterize the degree of lesion of the key lesions. The feature scoring module is used to extract global features of the image, evaluate the global features, and obtain a feature score for the image. The total score module is used to obtain the total score of the retina based on the lesion score and the feature score of the image.
9. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a retinal image scoring program stored in the memory and executable on the processor. When the processor executes the retinal image scoring program, it implements the steps of the retinal image scoring method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a retinal image scoring program, which, when executed by a processor, implements the steps of the retinal image scoring method as described in any one of claims 1-7.