CN111583261A

CN111583261A - Fundus super-wide-angle image analysis method and terminal

Info

Publication number: CN111583261A
Application number: CN202010565477.XA
Authority: CN
Inventors: 林晨; 喻碧莺
Original assignee: Ke Junlong
Current assignee: Lin Chen; Wisdom Medical Shenzhen Co ltd
Priority date: 2020-06-19
Filing date: 2020-06-19
Publication date: 2020-08-25
Anticipated expiration: 2040-06-19
Also published as: CN111583261B

Abstract

The invention provides a fundus super-wide-angle image analysis method and a terminal, wherein fundus super-wide-angle images are cut according to different fundus tissues to obtain area images corresponding to each fundus tissue; respectively carrying out lesion analysis on the area image corresponding to each fundus tissue to obtain the area analysis result of each fundus tissue; integrating the regional analysis result of each fundus tissue to obtain and output a final analysis result; according to the method, the fundus super-wide-angle image is divided and analyzed, the analysis amount of each time is reduced, a plurality of divided images can be correspondingly analyzed at the same time, the analysis influence speed is improved, finally, the analysis results of the divided regional images are integrated and analyzed to obtain the final analysis result, the phenomenon that the whole information is lost after the divided images are analyzed is avoided, the accurate analysis result can be obtained, and a better reference is provided for a doctor.

Description

Fundus super-wide-angle image analysis method and terminal

Technical Field

The invention relates to the field of image analysis, in particular to a fundus super-wide-angle image analysis method and a fundus super-wide-angle image analysis terminal.

Background

In the field of fundus image analysis, the prior art generally puts the whole fundus image into an AI model for analysis, outputs the disease probability of each ophthalmic disease corresponding to the fundus image, but the image analysis has poor pertinence, wastes computing resources, and is easy to cause program errors for some pictures with poor quality and excessive interference factors due to non-normative acquisition, which is particularly obvious in the analysis of fundus super-wide-angle images; in addition, the black box effect of the existing method is obvious, only the disease probability of a part of specified diseases is given, the decision process of the disease probability can not be explained by the corresponding inference, the physical sign and the phenomenon are not described, the report can not be displayed according to the clinical reading habit of a doctor, and the method has great limitation in the actual clinical application of auxiliary diagnosis.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: provides a fundus super-wide-angle image analysis method, which improves the analysis speed of the image.

In order to solve the technical problems, the invention adopts a technical scheme that:

a fundus ultra-wide angle image analysis method comprises the following steps:

s1, cutting the fundus super-wide-angle image according to different fundus tissues to obtain a region image corresponding to each fundus tissue;

s2, respectively carrying out lesion analysis on the area image corresponding to each fundus tissue to obtain the area analysis result of each fundus tissue;

and S3, integrating the area analysis result of each fundus tissue, and obtaining and outputting a final analysis result.

In order to solve the technical problem, the invention adopts another technical scheme as follows:

a fundus super-wide-angle image analysis terminal comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor executes the computer program to realize the following steps:

The invention has the beneficial effects that: the fundus super-wide-angle image is cut according to different fundus tissues, more than one cut image is analyzed respectively, the specific fundus tissue condition, namely the regional analysis result can be obtained, and the different fundus tissue conditions can be integrated to obtain the final analysis result; the image is cut and then analyzed, compared with the method for analyzing the whole image, the size of the input image is reduced, a plurality of cut images can be analyzed simultaneously, and the analysis speed of the image is improved; meanwhile, the specific conditions of different tissues of the eyeground can be acquired to be integrated and analyzed as required, the specific conditions of the eyeground tissues and the analysis results of the combination of the conditions of the eyeground tissues can be obtained, and the reference with higher clinical practicability is provided.

Drawings

Fig. 1 is a flowchart illustrating steps of a method for analyzing a super-wide-angle fundus image according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a fundus super-wide-angle image analysis terminal according to an embodiment of the present invention;

description of reference numerals:

1. an eye fundus super wide-angle image analysis terminal; 2. a processor; 3. a memory.

Detailed Description

In order to explain technical contents, achieved objects, and effects of the present invention in detail, the following description is made with reference to the accompanying drawings in combination with the embodiments.

Referring to fig. 1, a method for analyzing a super-wide-angle image of a fundus includes the steps of:

From the above description, the beneficial effects of the present invention are: the fundus super-wide-angle image is cut according to different fundus tissues, more than one cut image is analyzed respectively, the specific fundus tissue condition, namely the regional analysis result can be obtained, and the different fundus tissue conditions can be integrated to obtain the final analysis result; the image is cut and then analyzed, compared with the method for analyzing the whole image, the size of the input image is reduced, a plurality of cut images can be analyzed simultaneously, and the analysis speed of the image is improved; meanwhile, the specific conditions of different tissues of the eyeground can be acquired to be integrated and analyzed as required, the specific conditions of the eyeground tissues and the analysis results of the combination of the conditions of the eyeground tissues can be obtained, and the reference with higher clinical practicability is provided.

Further, before S1, the method further includes:

training an AI model to automatically mark different fundus tissues;

the S1 specifically includes:

s11, inputting the fundus super-wide-angle image into the AI model to obtain fundus super-wide-angle images with different fundus tissue marks;

s12, judging whether the position difference between the marks exceeds a threshold value, if so, returning to S11; otherwise, cutting the fundus super-wide-angle image according to the mark.

According to the description, the AI model is trained, different fundus tissues are automatically marked by the AI model, the marking result can be judged according to the preset threshold value, if the corresponding condition is not met, the marking can be carried out again, the marking accuracy of the AI model is ensured, the efficiency of marking the picture is improved, the cutting is carried out according to the marked image, and the completeness of the fundus tissues in the cut image can be ensured.

Further, the fundus tissue includes the optic disc, macula lutea, and the middle and peripheral portions of the retina;

the obtaining of the area image corresponding to each fundus tissue in S1 includes:

obtaining a first area image including a optic disc, a second area image including a macula lutea, and a third area image including a middle periphery and a peripheral portion of a retina;

classifying the first area image, the second area image and the third area image according to lesions by using an image classification model to obtain a lesion classification result of each area image;

performing semantic segmentation on the first area image, the second area image and the third area image respectively by using an image semantic segmentation model to obtain the shape of the fundus tissue in each area image;

and obtaining the region analysis result of each region image according to the lesion classification result and the shape of the fundus tissue.

According to the description, the integrality of the optic disc, the macula lutea and the retina middle periphery and the periphery in the cut images is guaranteed, the subsequent analysis of the images is facilitated, the images are classified through the image classification model and segmented through the semantic segmentation model, the pathological change probability can be obtained, meanwhile, the shape of the fundus tissue can be obtained, the subsequent formation of a report with high readability is facilitated, and a more detailed reference is provided for the diagnosis of a doctor.

Further, the S3 specifically includes:

and inputting the lesion classification result and the shape of the fundus tissue of each region image into a preset logic model to obtain and output a final analysis result.

From the above description, the lesion classification result and the fundus tissue shape of each region image are input into the preset logical model, the analysis results of the plurality of segmented images can be combined and analyzed, the local analysis results can be integrated to obtain the overall analysis result while the local analysis is considered to be accurate, the detailed analysis process can be seen in the logical model, and the clear analysis process can provide data with more reference for doctors.

Further, the step S12 is specifically;

judging whether the position of the mark of the macula lutea is on the left side or the right side of the level of the mark of the optic disc, if not, returning to the step S11;

determining whether the distance between the position of the macular marker and the position of the optic disc marker exceeds ten times the diameter of the optic disc marker, and if so, returning to S11;

judging whether the mark of the middle circumference part of the retina is not the mark of the equator part, if not, returning to the step S11

And if the position of the macular marking is on the left side or the right side of the level of the marking position of the optic disc, the distance between the position of the macular marking and the position of the marking of the optic disc is less than ten times the diameter of the marking of the optic disc, and the mark of the middle periphery of the retina is outside the mark of the equator part, cutting the fundus super-wide-angle image according to the mark.

According to the description, the specific standard judgment AI model is set to verify the judgment result of the fundus super-wide-angle image, and if the judgment result does not meet the threshold value, the marking is carried out again, so that the marked result with obvious errors can be eliminated, and the accuracy of the marked result is improved.

Referring to fig. 2, a fundus super-wide-angle image analysis terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the following steps when executing the computer program:

Further, before S1, the method further includes:

training an AI model to automatically mark different fundus tissues;

the S1 specifically includes:

Further, the S3 specifically includes:

Further, the step S12 is specifically;

According to the description, the specific standard judgment AI model is set to verify the judgment result of the fundus super-wide-angle image, and if the judgment result does not meet the threshold value, the marking is carried out again, so that the marked result with obvious errors can be eliminated, and the correctness of the marked result is improved.

Referring to fig. 1, a first embodiment of the present invention is:

a fundus ultra-wide angle image analysis method specifically comprises the following steps:

training an AI model to automatically mark different fundus tissues;

in an optional implementation mode, boundary tracing is carried out on areas such as eyelash shadows, eyelids and instrument borders in a large number of fundus super-wide-angle images through marking software (such as LableME, and also independently developed marking software), a generated mask image is input into an AI model (which can be FCN (fiber channel Network, full Convolutional neural Network; U-NET; U-NET + +, and other image semantic segmentation models) for training, the trained model can automatically mark areas such as the eyelash shadows, the eyelids and the instrument borders, and the marked instrument borders can be directly cut;

in an optional implementation mode, the optic disc, the macula lutea and the vortex veins are marked by marking software, images generated by marking are input into an AI model (image segmentation models such as FCN, U-NET and the like) for training, the marking range of other areas is set by taking the position of the optic disc, the macula lutea or the vortex veins as a reference, and different fundus tissues are marked by the AI model according to the training result and the set position of the other areas relative to the optic disc, the macula lutea or the vortex veins;

for example, according to the marked positions of the macula lutea and the optic disc, determining the distance between the macula lutea and the optic disc, taking the macula lutea as the center and taking two times to four times of the distance as the radius to make a circle, wherein the circle is the posterior pole part; determining an equatorial part according to the position of the vortex vein, wherein the periphery of the retina is outside the equatorial part;

specifically, the marking range of the macula lutea is determined by the fovea maculata;

wherein the fundus tissue includes the optic disc, macula lutea, and the mid-peripheral and peripheral portions of the retina;

the S1 specifically includes:

s12, judging whether the position difference between the marks exceeds a threshold value, if so, returning to S11; otherwise, cutting the fundus super-wide-angle image according to the mark;

specifically, the method comprises the following steps of;

If the position of the macular marking is on the left side or the right side of the level of the disc marking position, the distance between the position of the macular marking and the disc marking position is less than ten times the diameter of the disc marking, and the retina middle periphery marking is outside the equatorial marking, cutting the fundus super-wide-angle image according to the marking;

The second embodiment of the invention is as follows:

a method for analyzing a super-wide-angle fundus image, which is different from the first embodiment in that:

classifying the first area image, the second area image and the third area image according to lesions by using an image classification model (such as CNN (Convolutional Neural Networks)) to obtain a lesion classification result of each area image;

in an optional implementation mode, the lesion condition and the lesion classification result of each regional image are obtained by using an image classification model;

specifically, the image classification model is obtained through training, the pathological changes of different fundus tissues in a large number of fundus super-wide-angle images are marked, the pathological changes comprise normal, exudation, bleeding, atrophy and degeneration, pathological changes corresponding to the pathological changes of the different fundus tissues are marked, the marked result is input into a CNN model for training, and the trained CNN model can classify the pathological changes or mark the pathological changes according to the input images;

the image classification model can also be efficient NET, YOLO and the like;

if a glaucoma fundus image is marked as: pathological conditions, optic disc atrophy; lesions, glaucoma;

obtaining a region analysis result of each region image according to the lesion classification result and the shape of the fundus tissue;

specifically, for the first region image, determining the pathological change condition of the optic disc, namely normal, exudation, bleeding, atrophy or degeneration, through an image classification model, and obtaining the range of the optic cup and the optic disc through a semantic segmentation model; determining the pathological change condition of the macula lutea of the second region image through an image classification model, and acquiring the range of the macula lutea through a semantic segmentation model; determining whether the periphery of the retina has degeneration or not by an image classification model for the third area image;

the S3 specifically includes:

inputting the lesion classification result and the fundus tissue shape of each region image into a preset logic model to obtain and output a final analysis result;

specifically, lesion classification results of a plurality of fundus tissues and the shape of the fundus tissues are combined and analyzed, or a whole image is classified, the combined and analyzed results and the whole image lesion classification results are input into a logic model (such as a logistic model, a mixed linear model and the like) to be judged, a decision tree result is formed, and finally a conclusion is generated;

specifically, for the first area image, a cup-to-disc ratio is calculated according to the range of the optic cup and the optic disc, and if the cup-to-disc ratio exceeds a threshold value and the classification result of the image classification model is optic disc atrophy, the probability that the first area image analysis result is glaucoma is high; if the cup-disc ratio does not exceed the threshold value, but the classification result of the image classification model is optic disc atrophy, the first area image analysis result indicates that the possibility of glaucoma exists, and attention is paid to manual verification of the condition of the cup-disc; if the cup-to-disc ratio exceeds the threshold value and the classification result of the image classification model is that the optic disc is normal, the first area image analysis result is that the optic disc is normal;

in an alternative embodiment, the fundus super-wide-angle image is segmented to obtain a fourth area image including a posterior pole part, and the pathological condition and the pathological type of the posterior pole part are obtained through the image classification model, wherein the pathological condition is bleeding and exudation, the pathological type is diabetic retinopathy and the probability is 0.9, the pathological condition of the macula is obtained simultaneously, the bleeding and exudation are obtained, the pathological type is macular pathological and the probability is 0.94; because the two diseases rarely occur simultaneously in general, according to a preset logic model, in diabetic retinopathy, the weight of the lesion in the macular region is 0.3, and the weight of the lesion in the posterior pole part is 0.7, then 0.94 × 0.3+0.9 × 0.7 is calculated to be 0.912, and at this time, the final judgment result is diabetic retinopathy; if the lesion type of the posterior pole part is diabetic retinopathy and the probability is 0.1, calculating that 0.94 × 0.3+0.1 × 0.7 is 0.352, and finally judging that the result is macular degeneration;

specifically, a corresponding threshold may be set to determine a final determination result, the threshold may be set in a range of 0.5 to 0.8 according to the model calculation result, if the threshold is exceeded, the threshold is regarded as the final determination result, the diabetic retinopathy is calculated as above, if the threshold is set to 0.8, 0.912>0.8, and the final determination result in this case is the diabetic retinopathy; 0.352<0.8, and the final judgment result of the condition is macular degeneration;

when the final judgment result is output, the whole judgment process is output;

specifically, when the final analysis result is output, the lesion status, the lesion classification result, and the merged analysis result of each region are output, and if it is determined that glaucoma and diabetic retinopathy are present and retinal periphery degeneration is present, the final analysis result is: optic disc atrophy, cup size is large, variable zones are visible in the periretinal part, and bleeding and exudation of macula and posterior pole are suspected to be glaucoma and diabetic retinopathy.

Referring to fig. 2, a third embodiment of the present invention is:

a fundus super-wide-angle image analysis terminal 1 comprises a processor 2, a memory 3 and a computer program which is stored on the memory 3 and can run on the processor 2, wherein the processor 2 realizes the steps of the first embodiment or the second embodiment when executing the computer program.

In summary, the present invention provides a fundus super-wide angle image analysis method and a terminal, which reduce single input amount and can simultaneously analyze multiple regional images by analyzing multiple regional images obtained by segmenting a fundus super-wide angle image according to fundus tissues, thereby greatly improving the speed of image analysis; automatically marking the fundus tissues by training an AI model so as to divide images according to the marks, setting the threshold value of the marks in a matching manner, judging that the marks are wrong if the marking result exceeds the threshold value, and re-marking the fundus tissues, so that the accuracy of marking the fundus tissues is ensured, and a foundation is laid for subsequent analysis; the segmented image is subjected to lesion classification through the image classification model, the shape of the segmented image is obtained through the image semantic segmentation model, not only can the lesion condition be obtained, but also the lesion range can be obtained, and more reliable reference is provided for determination of subsequent lesions.

The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent changes made by using the contents of the present specification and the drawings, or applied directly or indirectly to the related technical fields, are included in the scope of the present invention.

Claims

1. A fundus ultra-wide angle image analysis method is characterized by comprising the following steps:

2. The method for analyzing the ultra-wide-angle image of the fundus of claim 1, further comprising, before S1:

training an AI model to automatically mark different fundus tissues;

the S1 specifically includes:

3. The method for analyzing the ultra-wide-angle image of the fundus oculi of claim 1, wherein the fundus tissues include the optic disc, the macula lutea, and the middle and peripheral parts of the retina;

4. The method for analyzing the ultra-wide-angle image of the fundus of the eye according to claim 3, wherein the step S3 is specifically as follows:

5. The method for analyzing the ultra-wide-angle image of the fundus according to claims 2-3, wherein the step S12 is specifically;

6. An eye fundus super-wide-angle image analysis terminal, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, characterized in that the processor implements the following steps when executing the computer program:

7. The fundus super-wide-angle image analysis terminal according to claim 6, further comprising, before said S1:

training an AI model to automatically mark different fundus tissues;

the S1 specifically includes:

8. The ultra-wide-angle fundus image analysis terminal according to claim 6, wherein the fundus tissues comprise the optic disc, macula lutea, and the middle and peripheral parts of retina;

9. The fundus super-wide-angle image analysis terminal according to claim 8, wherein the S3 specifically is:

10. The fundus super-wide-angle image analysis terminal according to claims 7-8, wherein said S12 is specifically;

judging whether the mark of the middle circumference part of the retina is not the mark of the equator part, if not, returning to S11;