Eye state evaluation method and electronic device

Abstract

The present application provides an eye state evaluation method and an electronic device. The method comprises: obtaining an optic disc image region from a first fundus image, and generating a plurality of optic cup disc ratio evaluation results based on the optic disc image region by a plurality of first models; obtaining a first evaluation result of the eye based on the optic cup disc ratio evaluation results; performing a plurality of data augmentation operations on the first fundus image to generate a plurality of second fundus images; generating a plurality of optic nerve fiber layer defect evaluation results based on the second fundus images by a plurality of second models; obtaining a second evaluation result of the eye based on the optic nerve fiber layer defect evaluation results; and obtaining an optic nerve evaluation result of the eye based on the first evaluation result and the second evaluation result. Thus, the present application can provide a reference for doctors when evaluating the eye state of patients, thereby helping doctors to give appropriate evaluation results for the eyes of patients.

Description

Technical Field

[0001] This invention relates to a method and electronic device for assessing human physical condition, and more particularly to a method and electronic device for assessing eye condition. Background Technology

[0002] According to statistics, glaucoma is the second leading cause of blindness in the United States. Generally, glaucoma is diagnosed by using optical coherence tomography (OCT) to calculate the retinal nerve fiber layer (RNFL). However, most people do not intentionally have OCT images taken.

[0003] Compared to the high cost and scarcity of OCT, color fundus photography images are relatively easy to obtain. Many signs of eye diseases can be observed from fundus images. The cup-to-disc ratio (CDR) of the optic disc and optic cup, as well as retinal nerve fiber layer defects (RNFL defects), can both be obtained from fundus images, thus helping to determine the degree of glaucoma.

[0004] However, since different doctors may determine the degree of glaucoma based on fundus images, it is an important issue for those skilled in the art to design a mechanism that can assist doctors in making such a determination. Summary of the Invention

[0005] In view of this, the present invention provides an eye condition assessment method and electronic device, which can be used to solve the above-mentioned technical problems.

[0006] This invention provides a method for assessing eye condition, suitable for an electronic device, comprising: acquiring a first fundus image of an eye, wherein the first fundus image includes an optic disc image region; acquiring the optic disc image region from the first fundus image, and generating multiple optic disc cup-disc ratio assessment results based on the optic disc image region using multiple first models; obtaining a first assessment result of the eye based on the multiple optic disc cup-disc ratio assessment results; performing multiple data amplification operations on the first fundus image to generate multiple second fundus images corresponding to the multiple data amplification operations respectively; generating multiple optic nerve fiber layer defect assessment results based on the multiple second fundus images using multiple second models; obtaining a second assessment result of the eye based on the multiple optic nerve fiber layer defect assessment results; and obtaining an optic nerve assessment result of the eye based on the first assessment result and the second assessment result.

[0007] This invention provides an electronic device including a storage circuit and a processor. The storage circuit stores program code. The processor is coupled to the storage circuit and accesses the program code to execute: acquiring a first fundus image of an eye, wherein the first fundus image includes an optic disc image region; acquiring the optic disc image region from the first fundus image and generating multiple optic disc cup-disc ratio assessment results based on the optic disc image region using multiple first models; obtaining a first assessment result of the eye based on the multiple optic disc cup-disc ratio assessment results; performing multiple data amplification operations on the first fundus image to generate multiple second fundus images corresponding to the multiple data amplification operations; generating multiple optic nerve fiber layer defect assessment results based on the multiple second fundus images using multiple second models; obtaining a second assessment result of the eye based on the multiple optic nerve fiber layer defect assessment results; and obtaining an optic nerve assessment result of the eye based on the first assessment result and the second assessment result. Attached Figure Description

[0008] The accompanying drawings are included to further illustrate the invention, and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0009] Figure 1 This is a schematic diagram of an electronic device according to an embodiment of the present invention;

[0010] Figure 2 This is a flowchart illustrating an eye condition assessment method according to an embodiment of the present invention;

[0011] Figure 3 This is a schematic diagram illustrating the acquisition of a first assessment result of the eye according to an embodiment of the present invention;

[0012] Figure 4 It is based on Figure 3 The diagram shows the second assessment results of the eye. Detailed Implementation

[0013] Reference will now be made in detail to exemplary embodiments of the invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same component reference numerals are used in the drawings and description to denote the same or similar parts.

[0014] Please refer to Figure 1 This is a schematic diagram of an electronic device according to an embodiment of the present invention. In different embodiments, the electronic device 100 may be, for example, various computer devices and / or smart devices, but is not limited thereto.

[0015] like Figure 1As shown, the electronic device 100 may include a storage circuit 102 and a processor 104. The storage circuit 102 may be, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk, or other similar devices or combinations thereof, and may be used to record multiple program codes or modules.

[0016] Processor 104 is coupled to storage circuit 102 and may be a general purpose processor, special purpose processor, conventional processor, digital signal processor, multiple microprocessors, one or more microprocessors incorporating a digital signal processor core, controller, microcontroller, application specific integrated circuit (ASIC), field programmable gate array (FPGA), any other type of integrated circuit, state machine, processor based on advanced RISC machine (ARM), and the like.

[0017] In an embodiment of the present invention, the processor 104 can access the modules and program code recorded in the storage circuit 102 to implement the eye state assessment method proposed in the present invention, the details of which are described below.

[0018] Please refer to Figure 2 This is a flowchart illustrating an eye condition assessment method according to an embodiment of the present invention. The method of this embodiment can be... Figure 1 The electronic device 100 performs the following: Figure 1 Component description shown Figure 2 Details of each step.

[0019] First, in step S210, the processor 104 may obtain a first fundus image of eye A. In one embodiment, eye A is, for example, one of the eyes of a patient, and the first fundus image is, for example, a color fundus image of eye A obtained by a medical professional using an instrument for taking fundus images, but is not limited to this.

[0020] In one embodiment, since a typical fundus image may include black borders, the processor 104 can crop the black borders after acquiring the fundus image to produce a fundus image without black borders as the aforementioned first fundus image. However, this is not a limitation. To make the concept of the present invention easier to understand, the following will be supplemented with... Figure 3 Further explanation is needed.

[0021] Please refer to Figure 3 This is a schematic diagram illustrating the acquisition of a first evaluation result of the eye according to an embodiment of the present invention. Figure 3 In this context, it is assumed that processor 104 obtains the following in step S210: Figure 3 The first fundus image 310 shown may include an optic disc image region 311 and an optic cup image region 312, which correspond to the optic disc and optic cup in the aforementioned eye A, respectively.

[0022] After obtaining the first fundus image 310, in step S220, the processor 104 can obtain the optic disc image region 311 from the first fundus image 310, and generate multiple CDR evaluation results based on the optic disc image region 311 by multiple first models.

[0023] In one embodiment, the processor 104 may input a first fundus image 310 into a pre-trained reference neural network 320, wherein the reference neural network 320 may identify and output an optic disc image region 311 in response to the first fundus image 310.

[0024] In one embodiment, before using the reference neural network 320 to identify the optic disc image region 311 in the first fundus image 310, the reference neural network 320 may be trained on training images labeled with specific regions of interest (ROIs) to learn the features of these ROIs through the training images. Therefore, in this embodiment, in order to enable the reference neural network 320 to identify the optic disc image region in any fundus image, the reference neural network 320 may first learn from multiple training images (i.e., fundus images) labeled with optic disc image regions to learn the features of the optic disc image regions in the fundus images, but it is not limited to this.

[0025] Subsequently, the processor 104 can provide the optic disc image region 311 output by the reference neural network 320 to the plurality of first models to generate corresponding CDR evaluation results. In embodiments of the present invention, the CDR evaluation results of each first model can be considered as normal or abnormal based on the ratio between the axial length of the optic cup and the axial length of the optic disc in the aforementioned eye A (i.e., CDR), but are not limited to this.

[0026] exist Figure 3 In this context, it is assumed that the plurality of first models under consideration (the number of which is, for example, N, and N is an odd number) include the first neural network 331, the second neural network 332, and the third neural network 333 shown, but may not be limited thereto.

[0027] In one embodiment, the first neural network 331 may output a CDR evaluation result R11 (e.g., normal or abnormal) in response to the optic disc image region 311.

[0028] In one embodiment, to enable the first neural network 331 to identify whether the corresponding optic disc retraction curve (CDR) is normal for any given optic disc image region, the first neural network 331 may first learn based on multiple optic disc image regions with normal CDRs to learn the characteristics of optic disc image regions with normal CDRs. Alternatively, the first neural network 331 may also learn based on multiple optic disc image regions with abnormal CDRs to learn the characteristics of optic disc image regions with abnormal CDRs, but this is not a limitation.

[0029] In one embodiment, the processor 104 can input the optic disc image region 311 into the second neural network 332, and the second neural network 332 can output the first CDR of the eye A in response to the optic disc image region 311.

[0030] In one embodiment, in order for the second neural network 332 to have the ability to identify the corresponding CDR for any optic disc image region, the second neural network 332 may first learn based on multiple optic disc image regions labeled with corresponding CDRs to learn the features of the optic disc image regions corresponding to various CDRs, but it is not limited to this.

[0031] After obtaining the first CDR, the processor 104 can determine whether the first CDR is higher than a preset threshold (e.g., 0.7). In one embodiment, in response to determining that the first CDR is higher than this preset threshold, the processor 104 can determine that the CDR evaluation result R12 of the second neural network 332 is abnormal. On the other hand, in response to determining that the first CDR is not higher than this preset threshold, the processor 104 can determine that the CDR evaluation result R12 of the second neural network 332 is normal, but this is not limited to this.

[0032] In one embodiment, the processor 104 can input the optic disc image region 311 into the third neural network 333, and the third neural network 333 can output the optic cup axis length and optic disc axis length of the eye A in response to the optic disc image region 311.

[0033] In one embodiment, in order to enable the third neural network 333 to identify the corresponding optic cup axis length and optic disc axis length for any optic disc image region, the third neural network 333 may first learn based on multiple optic disc image regions labeled with corresponding optic cup axis length and optic disc axis length, so as to learn the features of optic disc image regions corresponding to various optic cup axis lengths and optic disc axis lengths, but it is not limited to this.

[0034] After obtaining the optic cup axis length and optic disc axis length corresponding to the optic disc image region 311, the processor 104 can obtain the second CDR of the aforementioned eye A. For example, the processor 104 can divide the aforementioned optic cup axis length by the optic disc axis length to obtain the second CDR, but it is not limited to this.

[0035] Next, the processor 104 can determine whether the second CDR is higher than the aforementioned preset threshold value (e.g., 0.7). In one embodiment, in response to determining that the second CDR is higher than this preset threshold value, the processor 104 can determine that the CDR evaluation result R13 of the third neural network 333 is abnormal. On the other hand, in response to determining that the second CDR is not higher than this preset threshold value, the processor 104 can determine that the CDR evaluation result R13 of the third neural network 333 is normal, but this is not limited to this.

[0036] After obtaining the CDR evaluation results R11 to R13, in step S230, the processor 104 can obtain the first evaluation result RR1 of eye A based on the plurality of CDR evaluation results R11 to R13.

[0037] In one embodiment, the processor 104 outputs multiple first results indicating normality in CDR evaluation results R11 to R13, and identifies multiple second results indicating abnormality in CDR evaluation results R11 to R13. Then, in response to determining that the number of first results exceeds the number of second results, the processor 104 can determine that the first evaluation result RR1 indicates normality (i.e., the CDR of eye A is normal). On the other hand, in response to determining that the number of first results is less than the number of second results, the processor 104 can determine that the first evaluation result RR1 indicates abnormality (i.e., the CDR of eye A is abnormal).

[0038] In short, the processor 104 can make a majority decision based on the CDR evaluation results R11 to R13. If there are more first results indicating normality among the CDR evaluation results R11 to R13, the processor 104 can determine that the first evaluation result RR1 indicates normality; if there are more second results indicating abnormality among the CDR evaluation results R11 to R13, the processor 104 can determine that the first evaluation result RR1 indicates abnormality, but it is not limited to this.

[0039] Furthermore, in step S240, the processor 104 may perform various data amplification operations on the first fundus image 310 to generate multiple second fundus images corresponding to the multiple data amplification operations, respectively. To make the concept of the present invention easier to understand, the following will provide further details. Figure 4 Further explanation is needed.

[0040] Please refer to Figure 4 It is based on Figure 3 This diagram illustrates the acquisition of the second assessment results for the eye. Figure 4In the process, the processor 104 can perform M types of data amplification operations (M is a positive integer) on the first retinal image 310 to generate M second retinal images 411 to 41M. The M types of data amplification operations may include rotating, translating, flipping, scaling, stretching, etc. on the first retinal image 310, but are not limited to these.

[0041] Subsequently, in step S250, the processor 104 can input the second fundus images 411-41M into the second models 421-42M respectively, so that the corresponding prediction confidence values ​​output by each second model 421-42M are used as RNFL defect assessment results R21-R2M. In one embodiment, the prediction confidence values ​​corresponding to each second model 421-42M (i.e., RNFL defect assessment results R21-R2M) can indicate the degree of RNFL abnormality in the aforementioned eye A.

[0042] In one embodiment, in order to enable each second model 421-42M to identify the degree of abnormality of its corresponding RNFL in response to the data-amplified fundus image, each second model 421-42M may first learn based on multiple (data-amplified) fundus images labeled with the corresponding degree of abnormality of RNFL, so as to learn the features of fundus images corresponding to the degree of abnormality of various RNFLs, but is not limited to this.

[0043] After obtaining the prediction confidence values ​​of each of the second models 421 to 42M (i.e., RNFL defect assessment results R21 to R2M), in step S260, the processor 104 can obtain the second assessment result RR2 of eye A based on the plurality of RNFL defect assessment results R21 to R2M.

[0044] In one embodiment, the processor 104 can obtain a squared difference between the predicted confidence value and a reference baseline value (e.g., 0.5) for each of the second models 421-42M, and use this difference to find a candidate model among the second models 421-42M. In one embodiment, the squared difference corresponding to this candidate model can be the largest among the second models 421-42M.

[0045] Then, the processor 104 can determine whether the prediction confidence value corresponding to this candidate model is higher than the aforementioned reference benchmark value. If so, the processor 104 can determine that the second evaluation result RR2 of the aforementioned eye A is abnormal; otherwise, it can determine that the second evaluation result RR2 of the aforementioned eye A is normal.

[0046] After obtaining the first assessment result RR1 (which may indicate normal or abnormal) and the second assessment result RR2 (which may indicate normal or abnormal) of the aforementioned eye A, in step S270, the processor 104 may obtain the optic nerve assessment result of eye A based on the first assessment result RR1 and the second assessment result RR2.

[0047] In one embodiment, in response to the determination that the first assessment result RR1 indicates normal and the second assessment result RR2 indicates normal, the optic nerve assessment result of eye A is determined to belong to a first risk level; in response to the determination that the first assessment result RR1 indicates abnormal and the second assessment result RR2 indicates normal, the optic nerve assessment result of eye A is determined to belong to a second risk level, wherein the second risk level is greater than the first risk level; in response to the determination that the first assessment result RR1 indicates normal and the second assessment result RR2 indicates abnormal, the optic nerve assessment result of eye A is determined to belong to a third risk level, wherein the third risk level is greater than the second risk level; in response to the determination that the first assessment result RR1 indicates abnormal and the second assessment result RR2 indicates abnormal, the optic nerve assessment result of eye A is determined to belong to a fourth risk level, wherein the fourth risk level is greater than the third risk level.

[0048] In one embodiment, the first, second, third, and fourth risk levels can be understood as the probability that eye A will develop glaucoma. For example, when the optic nerve assessment result of eye A is at the first risk level, it means that the probability of eye A developing glaucoma is low. On the other hand, when the optic nerve assessment result of eye A is at the fourth risk level, it means that the probability of eye A developing glaucoma is high, but this is not limited to these categories. Therefore, the present invention can present the obtained optic nerve assessment results to physicians or other relevant personnel as auxiliary information for the diagnosis of eye A, but this is not limited to these categories.

[0049] In one embodiment, the first, second, third and fourth risk levels can be roughly understood as normal, low risk, medium risk and high risk, and can be summarized as shown in Table 1 below, but are not limited thereto.

[0050] First assessment result RR1 Second assessment result RR2 Optic nerve assessment results normal normal normal abnormal normal Low risk normal abnormal Medium risk abnormal abnormal High risk

[0051] Table 1

[0052] In summary, after obtaining the optic disc image region, this invention allows multiple first models with different recognition mechanisms to individually generate corresponding CDR assessment results based on the optic disc image region. A majority vote is then used to determine whether the first assessment result indicates normality or abnormality. Furthermore, after performing various data amplifications on a first fundus image to obtain multiple second fundus images, each second model generates a corresponding RNFL defect result based on one of these second fundus images, thereby obtaining a second assessment result indicating whether the eye's RNFL is normal or abnormal. Subsequently, this invention can comprehensively consider the first and second assessment results to obtain the optic nerve assessment result of the eye. Therefore, this invention can provide physicians with a reference when assessing the patient's eye condition, thereby helping physicians to give appropriate assessment results for the patient's eyes. For example, physicians can make appropriate assessments of whether a patient has glaucoma based on the optic nerve assessment results provided by the embodiments of this invention, thereby improving the quality of diagnosis.

[0053] 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An eye state evaluation method, adapted for an electronic device, characterized by, comprising: obtaining a first fundus image of an eye, wherein the first fundus image comprises an optic disc image region; obtaining the optic disc image region from the first fundus image and generating a plurality of optic cup disc ratio evaluation results based on the optic disc image region by a plurality of first models; obtaining a first evaluation result of the eye based on the plurality of optic cup disc ratio evaluation results; performing a plurality of data augmentation operations on the first fundus image to generate a plurality of second fundus images respectively corresponding to the plurality of data augmentation operations; generating a plurality of optic nerve fiber layer defect evaluation results based on the plurality of second fundus images by a plurality of second models, wherein the step of generating the plurality of optic nerve fiber layer defect evaluation results based on the plurality of second fundus images by the plurality of second models comprises: inputting one of the plurality of second fundus images into one of the plurality of second models, wherein the one of the plurality of second models outputs a prediction confidence value as one of the plurality of optic nerve fiber layer defect evaluation results in response to the one of the plurality of second fundus images, wherein the prediction confidence value indicates a degree of abnormality of the eye; obtaining a second evaluation result of the eye based on the plurality of optic nerve fiber layer defect evaluation results, wherein the step of obtaining the second evaluation result of the eye based on the plurality of optic nerve fiber layer defect evaluation results comprises: obtaining a squared difference between the prediction confidence value corresponding to each of the second models and a reference benchmark value, and finding a candidate model in the plurality of second models based on the squared difference, wherein the squared difference corresponding to the candidate model is the largest among the plurality of second models; determining that the second evaluation result of the eye indicates abnormality in response to determining that the prediction confidence value corresponding to the candidate model is higher than the reference benchmark value; and determining that the second evaluation result of the eye indicates normality in response to determining that the prediction confidence value corresponding to the candidate model is not higher than the reference benchmark value; and obtaining an optic nerve evaluation result of the eye based on the first evaluation result and the second evaluation result.

2. The method of claim 1, wherein the first fundus image comprises a color fundus image of the eye, and the color fundus image does not have a black border.

3. The method of claim 1, wherein the step of obtaining the optic disc image region from the first fundus image comprises: inputting the first fundus image into a reference neural network, wherein the reference neural network identifies the optic disc image region in the first fundus image in response to the first fundus image and outputs the optic disc image region.

4. The method of claim 1, wherein the plurality of first models comprises N neural networks, and each of the optic cup disc ratio evaluation results indicates whether a cup disc ratio of the eye is normal or abnormal, wherein N is an odd number.

5. The method of claim 4, wherein the step of obtaining the first evaluation result of the eye based on the plurality of optic cup disc ratio evaluation results comprises: finding a plurality of first results from the plurality of optic nerve cup-disc ratio evaluation results indicating normal, and finding a plurality of second results from the plurality of optic nerve cup-disc ratio evaluation results indicating abnormal; determining that the first evaluation result indicates normal in response to determining that the plurality of first results is more than the plurality of second results; and determining that the first evaluation result indicates abnormal in response to determining that the plurality of first results is less than the plurality of second results.

6. The method of claim 1, wherein the plurality of first models comprises a first neural network, and the method comprises: inputting the optic disc image region into the first neural network, wherein the first neural network outputs one of the plurality of optic nerve cup-disc ratio evaluation results in response to the optic disc image region.

7. The method of claim 1, wherein the plurality of first models comprises a second neural network, and the method comprises: inputting the optic disc image region into the second neural network, wherein the second neural network outputs a first optic nerve cup-disc ratio of the eye in response to the optic disc image region; determining that one of the plurality of optic nerve cup-disc ratio evaluation results corresponding to the second neural network indicates abnormal in response to determining that the first optic nerve cup-disc ratio is higher than a pre-set threshold value; and determining that one of the plurality of optic nerve cup-disc ratio evaluation results corresponding to the second neural network indicates normal in response to determining that the first optic nerve cup-disc ratio is not higher than the pre-set threshold value.

8. The method of claim 1, wherein the optic disc image region comprises an optic cup image region, the plurality of first models comprises a third neural network, and the method comprises: inputting the optic disc image region into the third neural network, wherein the third neural network outputs an optic cup axial length of the eye in response to the optic cup image region, and outputs an optic disc axial length of the eye in response to the optic disc image region; obtaining a second optic nerve cup-disc ratio of the eye based on the optic cup axial length and the optic disc axial length; determining that one of the plurality of optic nerve cup-disc ratio evaluation results corresponding to the third neural network indicates abnormal in response to determining that the second optic nerve cup-disc ratio is higher than a pre-set threshold value; and determining that one of the plurality of optic nerve cup-disc ratio evaluation results corresponding to the third neural network indicates normal in response to determining that the second optic nerve cup-disc ratio is not higher than the pre-set threshold value.

9. The method of claim 1, wherein the first evaluation result indicates normal or abnormal, the second evaluation result indicates normal or abnormal, and the step of obtaining an optic nerve evaluation result of the eye based on the first evaluation result and the second evaluation result comprises: determining that the optic nerve evaluation result of the eye belongs to a first risk level in response to determining that the first evaluation result indicates normal and the second evaluation result indicates normal; in response to determining that the first assessment result indicates abnormal and the second assessment result indicates normal, determining that the optic nerve assessment result of the eye belongs to a second risk level, wherein the second risk level is greater than the first risk level; in response to determining that the first assessment result indicates normal and the second assessment result indicates abnormal, determining that the optic nerve assessment result of the eye belongs to a third risk level, wherein the third risk level is greater than the second risk level; in response to determining that the first assessment result indicates abnormal and the second assessment result indicates abnormal, determining that the optic nerve assessment result of the eye belongs to a fourth risk level, wherein the fourth risk level is greater than the third risk level.

10. An electronic device, comprising: comprising: a storage circuit storing program codes; a processor coupled to the storage circuit and accessing the program codes to perform: obtaining a first fundus image of an eye, wherein the first fundus image comprises an optic disc image region; obtaining the optic disc image region from the first fundus image and generating a plurality of optic nerve cup-to-disc ratio assessment results based on the optic disc image region by a plurality of first models; obtaining a first assessment result of the eye based on the plurality of optic nerve cup-to-disc ratio assessment results; performing a plurality of data augmentation operations on the first fundus image to generate a plurality of second fundus images respectively corresponding to the plurality of data augmentation operations; generating a plurality of optic nerve fiber layer defect assessment results based on the plurality of second fundus images by a plurality of second models, comprising: inputting one of the plurality of second fundus images into one of the plurality of second models, wherein the one of the plurality of second models outputs a prediction confidence value as one of the plurality of optic nerve fiber layer defect assessment results in response to the one of the plurality of second fundus images, wherein the prediction confidence value indicates an abnormality level of the eye; obtaining a second assessment result of the eye based on the plurality of optic nerve fiber layer defect assessment results, comprising: obtaining a squared difference between the prediction confidence value corresponding to each of the second models and a reference benchmark value, and finding a candidate model in the plurality of second models based on the squared difference, wherein the squared difference corresponding to the candidate model is the largest among the plurality of second models; in response to determining that the prediction confidence value corresponding to the candidate model is higher than the reference benchmark value, determining that the second assessment result of the eye indicates abnormal; and in response to determining that the prediction confidence value corresponding to the candidate model is not higher than the reference benchmark value, determining that the second assessment result of the eye indicates normal; and obtaining an optic nerve assessment result of the eye based on the first assessment result and the second assessment result.

Images