A method, system, device and medium for identifying retinal blood vessels

Abstract

The application discloses a retinal blood vessel identification method, system, device and medium, the method comprises the following steps: acquiring a retinal blood vessel image; identifying the retinal blood vessel image according to a retinal blood vessel model to obtain a retinal blood vessel identification result; wherein the retinal blood vessel model is obtained through the following steps: acquiring and preprocessing an original retinal image set to obtain a first retinal image set; performing feature extraction processing on the first retinal image set to obtain a second retinal image set; performing adaptive clustering processing on the second retinal image to obtain a clustering identification result; updating the retinal blood vessel model according to the clustering identification result to obtain an updated retinal blood vessel model. The method can effectively guarantee the integrity and connectivity of the retinal blood vessel structure, effectively improve the anti-noise performance and identification accuracy of the retinal blood vessel identification, and improve the detection sensitivity of the low-contrast small blood vessel structure. The application relates to the technical field of image processing.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, system, device, and medium for identifying retinal vessels. Background Technology

[0002] In recent years, as the structural information of retinal vessels has contributed more and more to the early screening and diagnosis of diseases such as diabetic retinopathy, glaucoma, age-related macular degeneration, and hypertension, automatically capturing changes in vascular morphology in retinal vessel images has become one of the focuses of attention.

[0003] Currently, traditional retinal vessel recognition methods based on C-means clustering often segment retinal vessels based on the clustering attributes of sample points in the feature space. This method is susceptible to noise and outliers, resulting in low accuracy in vessel segmentation. Furthermore, this method to some extent ignores the features of pixels located at cluster centers. When the distance from a pixel at a cluster center to both the vessel and background cluster centers is relatively close, misclassification of that pixel can easily occur, leading to oversmoothing of fine vessel structures and reducing the integrity and connectivity of the retinal vessel structure. Additionally, this method is easily affected by the brightness of the image region, readily misclassifying dark areas or low-contrast fine vessel structures as background, resulting in low accuracy in vessel segmentation.

[0004] Therefore, the problems existing in the current technology still need to be solved and optimized. Summary of the Invention

[0005] The purpose of this invention is to at least partially solve one of the technical problems existing in the related art.

[0006] Therefore, one objective of this invention is to provide a method for identifying retinal vessels, which can effectively improve the noise resistance and accuracy of retinal vessel identification, increase the detection sensitivity of low-contrast small blood vessel structures, and effectively improve the connectivity and integrity of the vascular network.

[0007] Another objective of this application is to provide a system for recognizing retinal vessels.

[0008] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of this application include:

[0009] In a first aspect, embodiments of this application provide a method for identifying retinal vessels, comprising:

[0010] Acquire images of the retinal vessels to be identified;

[0011] The retinal vessel image is identified based on the updated retinal vessel model to obtain retinal vessel identification results;

[0012] The updated retinal vascular model is obtained through the following steps:

[0013] Obtain the raw retinal image set;

[0014] The original retinal image set is preprocessed to obtain the first retinal image set;

[0015] The first retinal image set is subjected to feature extraction processing to obtain a second retinal image set, which includes multiple second retinal images. The second retinal images are used to record the basic vascular network structure and fine vascular structure of the retinal vessels.

[0016] Adaptive clustering is performed on all the second retinal images to obtain clustering recognition results, which are used to record the association information between adjacent pixels in the second retinal images;

[0017] Based on the clustering recognition results, the initialized retinal vascular model is updated to obtain the updated retinal vascular model.

[0018] In addition, the identification method according to the above embodiments of this application may also have the following additional technical features:

[0019] Furthermore, in one embodiment of this application, the preprocessing of the original retinal image set to obtain a first retinal image set includes:

[0020] Get the preset color channels;

[0021] Based on the color channels, the original retinal image set is subjected to channel filtering processing to obtain a channel image set;

[0022] The channel image set is subjected to equalization processing to obtain the first retinal image set.

[0023] Furthermore, in one embodiment of this application, the step of performing feature extraction processing on the first retinal image set to obtain a second retinal image set includes:

[0024] The first retinal image set is subjected to shift filtering to obtain a shift feature set;

[0025] Phase consistency processing is performed on the first retinal image set to obtain a phase feature set;

[0026] Based on the shift feature set, wavelet feature fusion processing is performed on the phase feature set to obtain the second retinal image set.

[0027] Furthermore, in one embodiment of this application, the step of performing shift filtering on the first retinal image set to obtain a shift feature set includes:

[0028] The first retinal image set is subjected to convolutional filtering to obtain a filtered image set and a local extremum set. The filtered image set includes multiple convolutionally filtered first retinal images, and the local extremum set includes multiple local maxima points. Each local maxima point corresponds to one of the convolutionally filtered first retinal images.

[0029] Based on the local extremum set, the filtered image set is subjected to multi-layer shifting processing to obtain the shift feature set.

[0030] Further, in one embodiment of this application, the step of performing wavelet feature fusion processing on the phase feature set according to the shift feature set to obtain the second retinal image set includes:

[0031] The shift feature set is subjected to a first wavelet decomposition to obtain a first wavelet set, which includes a first slowly varying component and a first rapidly varying component.

[0032] The phase feature set is subjected to a second wavelet decomposition to obtain a second wavelet set, which includes a second slowly varying component and a second rapidly varying component.

[0033] Based on the first rapidly changing component, the second rapidly changing component is subjected to absolute value maximization fusion processing to obtain a fused rapidly changing component; and based on the first slowly changing component, the second slowly changing component is subjected to weighted average fusion processing to obtain a fused slowly changing component.

[0034] The second retinal image set is obtained based on the fused fast-changing component and the fused slow-changing component.

[0035] Further, in one embodiment of this application, the first rapid change component includes a first horizontal rapid change component, a first vertical rapid change component, and a first diagonal rapid change component; the second rapid change component includes a second horizontal rapid change component, a second vertical rapid change component, and a second diagonal rapid change component; the step of performing absolute value maxima fusion processing on the second rapid change component based on the first rapid change component to obtain a fused rapid change component includes:

[0036] Based on the first level rapid change component, the second level rapid change component is subjected to level extremum fusion processing to obtain the fused level component;

[0037] Based on the first vertical rapid change component, the second vertical rapid change component is subjected to vertical extremum fusion processing to obtain the fused vertical component;

[0038] Based on the first diagonal rapid change component, the second diagonal rapid change component is subjected to diagonal extremum fusion processing to obtain the fused diagonal component;

[0039] The fused fast-change component is obtained based on the fused horizontal component, the fused vertical component, and the fused diagonal component.

[0040] Furthermore, in one embodiment of this application, the adaptive clustering processing of the second retinal image to obtain the clustering recognition result includes:

[0041] Obtain the preset clustering threshold and number threshold, as well as the current cluster centers, membership matrix, and constraint coefficients;

[0042] Based on the membership matrix, the second retinal image is subjected to local similarity processing to obtain a local similarity set. The local similarity set includes multiple local similarity values, which are used to record the similarity measure between the target pixel and the neighboring pixels of the second retinal image.

[0043] Spatial constraint processing is performed on the local similarity set to obtain the constraint number and the constraint coefficients after constraint.

[0044] Based on the current constraint coefficients, perform absolute difference processing on the constrained constraint coefficients to obtain the constraint difference;

[0045] If the constraint difference is greater than or equal to the clustering threshold and the number of constraints is less than the number threshold, then the current cluster centers and membership matrix are updated according to the local similarity set, and then the process returns to obtain the preset clustering threshold and number threshold, as well as the current cluster centers, membership matrix, and constraint coefficients; or, if the constraint difference is less than the clustering threshold, or the number of constraints is greater than or equal to the number threshold, then the second retinal image is subjected to membership division processing according to the current cluster centers and membership matrix to obtain the clustering recognition result.

[0046] Secondly, embodiments of this application provide a retinal vessel recognition system, comprising:

[0047] The acquisition module is used to acquire images of the retinal vessels to be identified;

[0048] The recognition module is used to recognize the retinal vessel image based on the updated retinal vessel model to obtain the retinal vessel recognition result;

[0049] The updated retinal vascular model is obtained through the following steps:

[0050] Obtain the raw retinal image set;

[0051] The original retinal image set is preprocessed to obtain the first retinal image set;

[0052] The first retinal image set is subjected to feature extraction processing to obtain a second retinal image set, which includes multiple second retinal images. The second retinal images are used to record the basic vascular network structure and fine vascular structure of the retinal vessels.

[0053] Adaptive clustering is performed on all the second retinal images to obtain clustering recognition results, which are used to record the association information between adjacent pixels in the second retinal images;

[0054] Based on the clustering recognition results, the initialized retinal vascular model is updated to obtain the updated retinal vascular model.

[0055] Thirdly, embodiments of this application also provide a computer device, including:

[0056] At least one processor;

[0057] At least one memory for storing at least one program;

[0058] When the at least one program is executed by the at least one processor, the at least one processor implements the retinal vessel recognition method of the first aspect described above.

[0059] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a processor-executable program, which, when executed by the processor, is used to implement the retinal vessel recognition method of the first aspect described above.

[0060] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application:

[0061] This application discloses a method, system, device, and medium for identifying retinal vessels. The method involves acquiring an image of a retinal vessel to be identified; identifying the retinal vessel image based on an updated retinal vessel model to obtain a retinal vessel identification result; wherein the updated retinal vessel model is obtained through the following steps: acquiring an original retinal image set; preprocessing the original retinal image set to obtain a first retinal image set; performing feature extraction processing on the first retinal image set to obtain a second retinal image set, the second retinal image set including multiple second retinal images used to record the basic vascular network structure and fine vascular structure of the retinal vessels; performing adaptive clustering processing on all second retinal images to obtain a clustering recognition result, the clustering recognition result used to record the association information between adjacent pixels of the second retinal images; and updating the initialized retinal vessel model based on the clustering recognition result to obtain the updated retinal vessel model. This recognition method uses adaptive clustering of retinal images to determine the clustering recognition results that record the association information between adjacent pixels in the retinal image. This not only effectively ensures the integrity and connectivity of the retinal vascular structure, but also effectively improves the noise resistance and recognition accuracy of retinal vascular recognition, and enhances the detection sensitivity of low-contrast small vascular structures. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of this application or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions in this application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0063] Figure 1 A schematic flowchart illustrating a method for identifying retinal vessels provided in an embodiment of this application;

[0064] Figure 2 A schematic flowchart illustrating another method for identifying retinal vessels provided in this application embodiment;

[0065] Figure 3 A schematic diagram of the structure of a retinal blood vessel recognition system provided in an embodiment of this application;

[0066] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0067] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0069] Currently, traditional retinal vessel recognition methods based on C-means clustering often segment retinal vessels based on the clustering attributes of sample points in the feature space. This method is susceptible to noise and outliers, resulting in low accuracy in vessel segmentation. Furthermore, this method to some extent ignores the features of pixels located at cluster centers. When the distance from a pixel at a cluster center to both the vessel and background cluster centers is relatively close, misclassification of that pixel can easily occur, leading to oversmoothing of fine vessel structures and reducing the integrity and connectivity of the retinal vessel structure. Additionally, this method is easily affected by the brightness of the image region, readily misclassifying dark areas or low-contrast fine vessel structures as background, resulting in low accuracy in vessel segmentation.

[0070] In view of this, embodiments of the present invention provide a method, system, device and medium for identifying retinal vessels. The identification method can effectively improve the noise resistance and accuracy of retinal vessel identification, improve the detection sensitivity of low-contrast small blood vessel structures, and effectively improve the connectivity and integrity of the vascular network.

[0071] Reference Figure 1 and Figure 2 In this embodiment of the application, a method for identifying retinal vessels includes:

[0072] Step 110: Obtain the image of the retinal vessels to be identified;

[0073] Step 120: Identify the retinal vessel image based on the updated retinal vessel model to obtain the retinal vessel identification result;

[0074] In this embodiment of the application, the retinal vessel image to be identified can be a color fundus image; after obtaining the retinal vessel image to be identified, the retinal vessel image can be input into the updated retinal vessel model for identification, and the retinal vessel identification result output by the updated retinal vessel model can be obtained.

[0075] The updated retinal vascular model is obtained through the following steps:

[0076] Step 130: Obtain the raw retinal image set;

[0077] Step 140: Preprocess the original retinal image set to obtain the first retinal image set;

[0078] Step 140, preprocessing the original retinal image set to obtain a first retinal image set, includes:

[0079] Step 141: Obtain the preset color channels;

[0080] Step 142: Based on the color channels, perform channel filtering processing on the original retinal image set to obtain a channel image set;

[0081] Step 143: Perform equalization processing on the channel image set to obtain the first retinal image set.

[0082] In the embodiments of this application, the original retinal image set can be obtained by acquiring a fundus camera or by obtaining a publicly available fundus image dataset on the Internet. There are various specific implementation methods, which will not be elaborated here.

[0083] It is understood that the preset color channel can be any one of the red channel, blue channel, or green channel. In this embodiment, the green channel is used as an example. First, each original retinal image in the original retinal image set is filtered according to the green channel to select several original retinal images under the green channel. All the original retinal images under the green channel are then used as the channel image set. Next, histogram equalization is performed on each original retinal image in the channel set. Contrast Limited Adaptive Histogram Equalization (CLAHE) is a preferred choice for histogram equalization in this application to improve the contrast of the original retinal images under the green channel and obtain the first retinal image set.

[0084] Step 150: Perform feature extraction processing on the first retinal image set to obtain a second retinal image set. The second retinal image set includes multiple second retinal images, which are used to record the basic vascular network structure and fine vascular structure of the retinal vessels.

[0085] Step 150, performing feature extraction processing on the first retinal image set to obtain a second retinal image set, includes:

[0086] Step 151: Perform shift filtering on the first retinal image set to obtain a shift feature set;

[0087] Step 151 involves performing shift filtering on the first retinal image set to obtain a shift feature set, including:

[0088] Step 1511: Perform convolutional filtering on the first retinal image set to obtain a filtered image set and a local extremum set. The filtered image set includes multiple convolutionally filtered first retinal images, and the local extremum set includes multiple local maxima points. Each local maxima point corresponds to one of the convolutionally filtered first retinal images.

[0089] Step 1512: Based on the local extremum set, perform multi-layer shifting processing on the filtered image set to obtain the shifted feature set.

[0090] In this embodiment, convolutional filtering can be implemented using a B-COSFIRE filter. Specifically, for a first retinal image in the first retinal image set, the first retinal image is first convolved using a B-COSFIRE filter to obtain the Difference of Gaussian (DoG) filter response of the first retinal image on concentric circles, as well as the local maxima corresponding to the DoG filter response. The DoG filter response is then used as one of the filtered images in the filter image set. Next, the DoG filter response is sequentially subjected to blur shift and reverse shift operations to complete multi-layer shift processing in sequence, thereby obtaining the shift features.

[0091] It is understandable that the blur shift can be achieved by calculating the maximum weighted response of the DoG filter in the polar coordinate neighborhood of the local extremum point, thereby obtaining the blurred DoG filter response. Then, the blurred DoG filter response is shifted in the opposite direction to the polar coordinates of the local extremum point, and the shifted DoG filter response is used as the shift feature. The same principle applies to the remaining first retinal images, and can be easily deduced. This application will not elaborate further here. Furthermore, using the local extremum point's maximum coordinates as... For example, reverse shifting can be used to shift the blurred DoG filter response along... The direction of movement can be ρ, and the specific distance of movement can be ρ. The example in this application is for illustration only and does not constitute any limitation on this application.

[0092] Step 152: Perform phase consistency processing on the first retinal image set to obtain a phase feature set;

[0093] In the embodiments of this application, for a certain first retinal image in the first retinal image set, phase consistency processing can be performed by convolving the first retinal image with log-Gabor filter banks of different scales and directions to obtain the two-dimensional phase consistency response (i.e., phase feature) of the first retinal image. The same principle applies to the other first retinal images, and can be easily deduced by analogy. This application will not elaborate further here.

[0094] It is understood that the log-Gabor filter bank can specifically be an even-symmetric log-Gabor filter and an odd-symmetric log-Gabor filter. By convolving the first retinal image with the even-symmetric log-Gabor filter and the odd-symmetric log-Gabor filter respectively, and then calculating the amplitude and phase angle of the convolution output, the phase consistency response of the first retinal image is obtained. Furthermore, the scale and orientation of the log-Gabor filter can be set according to the actual situation. For example, for the orientation, six orientations can be used: 0°, 60°, 120°, 180°, 240°, and 300°. This application's example is for illustration only; the specific scale content is similar to the aforementioned orientation and can be easily deduced, so this application will not elaborate further here.

[0095] Step 153: Based on the shift feature set, perform wavelet feature fusion processing on the phase feature set to obtain the second retinal image set.

[0096] Step 153, performing wavelet feature fusion processing on the phase feature set based on the shift feature set to obtain the second retinal image set, includes:

[0097] Step 1531: Perform a first wavelet decomposition on the shift feature set to obtain a first wavelet set, which includes a first slowly varying component and a first rapidly varying component.

[0098] Step 1532: Perform a second wavelet decomposition on the phase feature set to obtain a second wavelet set, which includes a second slowly varying component and a second rapidly varying component.

[0099] In this embodiment of the application, taking the displacement feature and phase feature corresponding to a certain first retinal image as an example, the first wavelet decomposition can be a wavelet transform of the displacement feature, and the second wavelet decomposition can be a wavelet transform of the phase feature. Specifically, it can be processed by a second-order Daubechies wavelet to obtain the high-frequency and low-frequency components of the displacement feature, as well as the high-frequency and low-frequency components of the phase feature.

[0100] It is understandable that the low-frequency component of the shift feature is the first slowly varying component, and the high-frequency component is the first rapidly varying component; while the low-frequency component of the phase feature is the second slowly varying component, and the high-frequency component is the second rapidly varying component.

[0101] Step 1533: Based on the first fast-changing component, perform absolute value maximization fusion processing on the second fast-changing component to obtain a fused fast-changing component; and based on the first slow-changing component, perform weighted average fusion processing on the second slow-changing component to obtain a fused slow-changing component.

[0102] In some embodiments, the first rapid change component includes a first horizontal rapid change component, a first vertical rapid change component, and a first diagonal rapid change component; the second rapid change component includes a second horizontal rapid change component, a second vertical rapid change component, and a second diagonal rapid change component; step 1533, performing absolute value maximization fusion processing on the second rapid change component based on the first rapid change component to obtain a fused rapid change component, includes:

[0103] Step 15331: Based on the first level rapid change component, perform level extremum fusion processing on the second level rapid change component to obtain the fused level component;

[0104] Step 15332: Based on the first vertical rapid change component, perform vertical extremum fusion processing on the second vertical rapid change component to obtain the fused vertical component;

[0105] Step 15333: Based on the first diagonal rapid change component, perform diagonal extremum fusion processing on the second diagonal rapid change component to obtain the fused diagonal component;

[0106] Step 15334: Obtain the fused fast-change component based on the fused horizontal component, the fused vertical component, and the fused diagonal component.

[0107] In this embodiment, the first rapid transition component of the shift feature can be a first horizontal rapid transition component, a first vertical rapid transition component, and a first diagonal rapid transition component, and the second rapid transition component of the phase feature includes a second horizontal rapid transition component, a second vertical rapid transition component, and a second diagonal rapid transition component. Furthermore, since the shift feature contains a more complete basic vascular network structure, and the basic vascular network structure is mainly present in the first slow transition component, in this embodiment, the low-frequency fusion coefficient can be determined by a weighted average rule, and then the first and second slow transition components can be fused by a weighted average to obtain the fused slow transition component. The equivalent formula for the fused slow transition component can be expressed as:

[0108] L * (x, y) = w1L1(x, y) + w2L2(x, y)

[0109] Where x is the x-coordinate of a pixel in the first retinal image, y is the y-coordinate of a pixel in the first retinal image, and L * () represents the fused gradually varying components, w1 is the low-frequency fusion coefficient corresponding to the first gradually varying component, w2 is the low-frequency fusion coefficient corresponding to the second gradually varying component, L1() is the first gradually varying component, and l2() is the second gradually varying component.

[0110] It is understood that the absolute value maxima fusion in this embodiment can preserve as many detailed features as possible in the retinal image to improve the response to small vascular structures. Specifically, absolute value maxima fusion can use the maximum value of the same fast-changing component of the shift feature and the phase feature as the fused fast-changing component. The equivalent formula for fusion of the fast-changing component can be expressed as:

[0111]

[0112] Specifically, x is the x-coordinate of a pixel in the first retinal image, and y is the y-coordinate of a pixel in the first retinal image. To integrate the horizontal components, max[] is the extremum function. For the first level of fast variable components This is the second level of rapidly changing components.

[0113] It is also understandable that the fusion of vertical and diagonal components is similar to that of the aforementioned fusion of rapidly changing components, and can be easily deduced by analogy; therefore, this application will not elaborate further here. After obtaining the fused horizontal, vertical, and diagonal components, they can be integrated to obtain the fused rapidly changing components.

[0114] Step 1534: Obtain the second retinal image set based on the fused fast-changing component and the fused slow-changing component.

[0115] In this embodiment of the application, for a certain first retinal image, after obtaining the fused fast-changing component and the fused slow-changing component, the fused fast-changing component and the fused slow-changing component can be subjected to inverse wavelet transform, and then the feature image after inverse wavelet transform is output, and the image after inverse wavelet transform is used as the second retinal image. The other first retinal images are similar, and this application will not elaborate further here.

[0116] It is understood that the embodiments of this application, through wavelet feature fusion, significantly improve the response of small blood vessel structures while preserving the basic vascular network structure. This not only avoids the hollow phenomenon in thick blood vessels but also reflects a more complete small blood vessel structure.

[0117] Step 160: Perform adaptive clustering processing on all the second retinal images to obtain clustering recognition results. The clustering recognition results are used to record the association information between adjacent pixels in the second retinal images.

[0118] Step 160, performing adaptive clustering processing on the second retinal image to obtain clustering recognition results, includes:

[0119] Step 161: Obtain the preset clustering threshold and frequency threshold, as well as the current cluster centers, membership matrix, and constraint coefficients;

[0120] Step 162: Based on the membership matrix, perform local similarity processing on the second retinal image to obtain a local similarity set. The local similarity set includes multiple local similarity values, which are used to record the similarity measure between the target pixel and its neighboring pixels in the second retinal image.

[0121] Step 163: Perform spatial constraint processing on the local similarity set to obtain the constraint degree and the constraint coefficients after constraint;

[0122] Step 164: Based on the current constraint coefficients, perform absolute difference processing on the constraint coefficients after constraint to obtain the constraint difference;

[0123] Step 165: When the constraint difference is greater than or equal to the clustering threshold and the constraint count is less than the count threshold, the current cluster centers and membership matrix are updated according to the local similarity set, and then the process returns to obtain the preset clustering threshold and count threshold, as well as the current cluster centers, membership matrix, and constraint coefficients; or, when the constraint difference is less than the clustering threshold, or the constraint count is greater than or equal to the count threshold, the second retinal image is subjected to membership partitioning processing according to the current cluster centers and membership matrix to obtain the clustering recognition result.

[0124] It is understandable that the clustering threshold and the number threshold are used as update conditions for the cluster centers and membership matrix. The specific clustering threshold and number threshold can be set according to the actual situation, and this application will not elaborate further here. The membership matrix is ​​used to record the membership degree of sample pixels to each category (i.e., cluster center). For the cluster centers and membership matrix, the cluster centers can be initialized in the first update process, and the initialized cluster centers can be used as the current cluster centers, while the membership matrix can be calculated according to the original membership degree formula; after the second update process, the updated cluster centers and membership matrix from the previous update process can be used as the current cluster centers and membership matrix.

[0125] In this embodiment of the application, taking a target pixel in the second retinal image as an example, firstly, a neighborhood window, which can be 3×3 pixels in size, can be used to select several neighboring pixels of the target pixel. Then, the local similarity value is determined based on the local spatial distance between the neighboring pixels and the target pixel, as well as the feature value elements of the pixel.

[0126] For example, taking the local similarity value of the i-th target pixel as an example, the equivalent formula for its local similarity value can be expressed as:

[0127]

[0128] Where, α ij (j) represents the local similarity value between the i-th target pixel and the r-th neighboring pixel in the neighborhood window, corresponding to the j-th cluster center; u ij u represents the membership degree of the i-th target pixel to the j-th cluster center; rj D represents the membership degree of the r-th neighboring pixel to the j-th cluster center; ir Let be the spatial distance between the r-th neighboring pixel and the i-th target pixel.

[0129] It is understandable that after obtaining all local similarity values ​​of the second retinal image, a local similarity set can be determined based on all local similarity values. Then, spatial constraints are applied to the local similarity set to obtain the corresponding constraint degree and the constraint coefficients after constraint. The equivalent formula for the constraint coefficients after constraint can be expressed as:

[0130]

[0131] Among them, J p The constraint coefficients are defined as follows: N is the total number of pixels in the second retinal image; c is the total number of cluster centers; u ij F represents the membership degree of the i-th target pixel to the j-th cluster center; m is the fuzzy weighting index of the membership degree; iv is the feature value of the i-th target pixel; j Let F be the eigenvalue of the j-th cluster center; i -v j || 2 N is the Euclidean distance between the i-th target pixel and the j-th cluster center; i α is the total number of neighboring pixels corresponding to the i-th target pixel; Wi is the neighborhood window of the i-th target pixel; α ij (j) represents the local similarity value between the i-th target pixel and the r-th neighboring pixel in the neighborhood window, corresponding to the j-th cluster center.

[0132] It is worth mentioning that in this embodiment, the cluster center c can be 2, which can represent the background cluster center and the blood vessel cluster center respectively; the fuzzy weighting index m can be 2, and this value can be set according to the actual situation. This example is only for illustration; the feature value F of the i-th target pixel. i The eigenvalues ​​v of the j-th cluster center j The results can be obtained from the shift response diagrams corresponding to the aforementioned shift characteristics and the phase response diagrams corresponding to the phase characteristics, which will not be elaborated further in this application.

[0133] It should be noted that after obtaining the constraint coefficients after the current number of constraint iterations, the difference between the constraint coefficients after the current constraint iterations and the current constraint coefficients (i.e., the constraint coefficients after the previous number of constraint iterations) can be calculated. Then, it is determined whether the difference is positive. If the difference is positive, it is used as the constraint difference; if the difference is negative, the absolute value of the difference is used as the constraint difference. After obtaining the constraint difference, proceed to step 165. When the constraint difference is greater than or equal to the clustering threshold and the number of constraint iterations is less than the threshold, the current cluster centers and membership matrix are updated based on the local similarity set, and then the process returns to step 161. When the constraint difference is less than the clustering threshold, or the number of constraint iterations is greater than or equal to the threshold, the second retinal image is subjected to membership division processing based on the current cluster centers and membership matrix to obtain the clustering recognition result.

[0134] It is worth mentioning that the updates to the cluster centers and membership matrix can be calculated using the equivalent formula of minimizing the constraint coefficients after constraints using the Lagrange multiplier method, which will not be elaborated further in this application. After exiting the loop, two cluster centers and the final membership matrix are obtained. These two cluster centers correspond to the background cluster center and the blood vessel cluster center, respectively. Then, the clustering recognition result is determined based on the two cluster centers and the final membership matrix. Specifically, each pixel can be divided into the class to which the cluster center with the highest membership degree belongs, thus obtaining the clustering recognition result. This clustering recognition result fully records the correlation information between adjacent pixels of the corresponding retinal image, and can preserve a more complete fine blood vessel structure while retaining the basic blood vessel network structure, exhibiting high sensitivity for detecting low-contrast fine blood vessel structures.

[0135] For example, taking a small blood vessel pixel as the target pixel, within the neighborhood window, the similarity metric between the neighboring small blood vessel pixels and the target pixel is relatively high, meaning they are strongly correlated with the target pixel; while the similarity metric between the neighboring background pixels and the target pixel is very low, meaning they are weakly correlated with the target pixel. Constrained by the strong correlation with the neighboring small blood vessel pixels, in step 165, the membership degree of the target pixel belonging to the blood vessel center is increased during the cluster center and membership matrix update, so that after clustering convergence, the target pixel is classified as a blood vessel pixel. Furthermore, taking a large blood vessel pixel as the target pixel, the similarity metric between the neighboring blood vessel pixels and the target pixel is relatively high, increasing the membership degree of the target pixel belonging to the blood vessel center. When the target pixel is a noisy point representing a large blood vessel, the similarity metric between the target pixel and neighboring blood vessel pixels is very small. In steps 161 to 165, to minimize the constraint coefficient, the similarity metric between the target pixel and neighboring pixels is updated in an increasing trend. This increases the membership degree of the target pixel to blood vessels and decreases the membership degree to the background, resulting in the target pixel being classified as a blood vessel pixel after clustering convergence. When the target pixel is a pixel in the background region, the similarity metric between the target pixel and neighboring background pixels is large, increasing the membership degree of the target pixel to the background center, resulting in the target pixel still being classified as a background pixel after clustering convergence.

[0136] Step 170: Update the initialized retinal vascular model based on the clustering recognition results to obtain the updated retinal vascular model.

[0137] In this embodiment of the application, before the retinal blood vessel model is put into use, it is necessary to ensure that the retinal blood vessel model has a high recognition accuracy for the input retinal blood vessel images. Therefore, its internal parameters can be adjusted according to the clustering recognition results so that the retinal blood vessel model can achieve a better prediction effect.

[0138] The retinal vessel recognition system according to embodiments of this application is described in detail below with reference to the accompanying drawings.

[0139] Reference Figure 3 The retinal vessel recognition system proposed in this application includes:

[0140] The acquisition module 101 is used to acquire an image of the retinal vessels to be identified;

[0141] The recognition module 102 is used to recognize the retinal vessel image based on the updated retinal vessel model to obtain the retinal vessel recognition result;

[0142] The updated retinal vascular model is obtained through the following steps:

[0143] Obtain the raw retinal image set;

[0144] The original retinal image set is preprocessed to obtain the first retinal image set;

[0145] The first retinal image set is subjected to feature extraction processing to obtain a second retinal image set, which includes multiple second retinal images. The second retinal images are used to record the basic vascular network structure and fine vascular structure of the retinal vessels.

[0146] Adaptive clustering is performed on all the second retinal images to obtain clustering recognition results, which are used to record the association information between adjacent pixels in the second retinal images;

[0147] Based on the clustering recognition results, the initialized retinal vascular model is updated to obtain the updated retinal vascular model.

[0148] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0149] Reference Figure 4 This application also provides a computer device, including:

[0150] At least one processor 201;

[0151] At least one memory 202 is used to store at least one program;

[0152] When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the method embodiment described above.

[0153] Similarly, it can be understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0154] This application also provides a computer-readable storage medium storing a program executable by a processor 201, which, when executed by the processor 201, is used to implement the above-described method embodiments.

[0155] Similarly, the content of the above method embodiments is applicable to the present computer-readable storage medium embodiments. The specific functions implemented by the present computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0156] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0157] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0158] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0160] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0161] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0162] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0163] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0164] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for identifying retinal vessels, characterized in that, include: Acquire images of the retinal vessels to be identified; The retinal vessel image is identified based on the updated retinal vessel model to obtain retinal vessel identification results; The updated retinal vascular model is obtained through the following steps: Obtain the raw retinal image set; The original retinal image set is preprocessed to obtain the first retinal image set; The first retinal image set is subjected to feature extraction processing to obtain a second retinal image set, which includes multiple second retinal images. The second retinal images are used to record the basic vascular network structure and fine vascular structure of the retinal vessels. Adaptive clustering is performed on all the second retinal images to obtain clustering recognition results, which are used to record the association information between adjacent pixels in the second retinal images; Based on the clustering recognition results, the initialized retinal vascular model is updated to obtain the updated retinal vascular model. The step of performing feature extraction processing on the first retinal image set to obtain the second retinal image set includes: The first retinal image set is subjected to shift filtering to obtain a shift feature set; Phase consistency processing is performed on the first retinal image set to obtain a phase feature set; Based on the shift feature set, wavelet feature fusion processing is performed on the phase feature set to obtain the second retinal image set; The shift filtering process performed on the first retinal image set to obtain a shift feature set includes: The first retinal image set is subjected to convolutional filtering to obtain a filtered image set and a local extremum set. The filtered image set includes multiple convolutionally filtered first retinal images, and the local extremum set includes multiple local maxima points. Each local maxima point corresponds to one of the convolutionally filtered first retinal images. Based on the local extremum set, the filtered image set is subjected to multi-layer shifting processing to obtain the shift feature set; The second retinal image set is obtained by performing wavelet feature fusion processing on the phase feature set based on the shift feature set, including: The shift feature set is subjected to a first wavelet decomposition to obtain a first wavelet set, which includes a first slowly varying component and a first rapidly varying component. The phase feature set is subjected to a second wavelet decomposition to obtain a second wavelet set, which includes a second slowly varying component and a second rapidly varying component. Based on the first rapidly changing component, the second rapidly changing component is subjected to absolute value maximization fusion processing to obtain a fused rapidly changing component; and based on the first slowly changing component, the second slowly changing component is subjected to weighted average fusion processing to obtain a fused slowly changing component. The second retinal image set is obtained based on the fused fast-changing component and the fused slow-changing component.

2. The method for identifying retinal vessels according to claim 1, characterized in that, The preprocessing of the original retinal image set to obtain the first retinal image set includes: Get the preset color channels; Based on the color channels, the original retinal image set is subjected to channel filtering processing to obtain a channel image set; The channel image set is subjected to equalization processing to obtain the first retinal image set.

3. The method for identifying retinal vessels according to claim 1, characterized in that, The first rapid change component includes a first horizontal rapid change component, a first vertical rapid change component, and a first diagonal rapid change component; the second rapid change component includes a second horizontal rapid change component, a second vertical rapid change component, and a second diagonal rapid change component; the step of performing absolute value maximization fusion processing on the second rapid change component based on the first rapid change component to obtain a fused rapid change component includes: Based on the first level rapid change component, the second level rapid change component is subjected to level extremum fusion processing to obtain the fused level component; Based on the first vertical rapid change component, the second vertical rapid change component is subjected to vertical extremum fusion processing to obtain the fused vertical component; Based on the first diagonal rapid change component, the second diagonal rapid change component is subjected to diagonal extremum fusion processing to obtain the fused diagonal component; The fused fast-change component is obtained based on the fused horizontal component, the fused vertical component, and the fused diagonal component.

4. The method for identifying retinal vessels according to claim 1, characterized in that, The adaptive clustering process performed on the second retinal image to obtain the clustering recognition result includes: Obtain the preset clustering threshold and number threshold, as well as the current cluster centers, membership matrix, and constraint coefficients; Based on the membership matrix, the second retinal image is subjected to local similarity processing to obtain a local similarity set. The local similarity set includes multiple local similarity values, which are used to record the similarity measure between the target pixel and the neighboring pixels of the second retinal image. Spatial constraint processing is performed on the local similarity set to obtain the constraint number and the constraint coefficients after constraint. Based on the current constraint coefficients, perform absolute difference processing on the constrained constraint coefficients to obtain the constraint difference; If the constraint difference is greater than or equal to the clustering threshold and the number of constraints is less than the number threshold, then the current cluster centers and membership matrix are updated according to the local similarity set, and then the process returns to obtain the preset clustering threshold and number threshold, as well as the current cluster centers, membership matrix, and constraint coefficients; or, if the constraint difference is less than the clustering threshold, or the number of constraints is greater than or equal to the number threshold, then the second retinal image is subjected to membership division processing according to the current cluster centers and membership matrix to obtain the clustering recognition result.

5. A retinal vessel recognition system, characterized in that, include: The acquisition module is used to acquire images of the retinal vessels to be identified; The recognition module is used to recognize the retinal vessel image based on the updated retinal vessel model to obtain the retinal vessel recognition result; The updated retinal vascular model is obtained through the following steps: Obtain the raw retinal image set; The original retinal image set is preprocessed to obtain the first retinal image set; The first retinal image set is subjected to feature extraction processing to obtain a second retinal image set, which includes multiple second retinal images. The second retinal images are used to record the basic vascular network structure and fine vascular structure of the retinal vessels. Adaptive clustering is performed on all the second retinal images to obtain clustering recognition results, which are used to record the association information between adjacent pixels in the second retinal images; Based on the clustering recognition results, the initialized retinal vascular model is updated to obtain the updated retinal vascular model. The step of performing feature extraction processing on the first retinal image set to obtain the second retinal image set includes: The first retinal image set is subjected to shift filtering to obtain a shift feature set; Phase consistency processing is performed on the first retinal image set to obtain a phase feature set; Based on the shift feature set, wavelet feature fusion processing is performed on the phase feature set to obtain the second retinal image set; The shift filtering process performed on the first retinal image set to obtain a shift feature set includes: The first retinal image set is subjected to convolutional filtering to obtain a filtered image set and a local extremum set. The filtered image set includes multiple convolutionally filtered first retinal images, and the local extremum set includes multiple local maxima points. Each local maxima point corresponds to one of the convolutionally filtered first retinal images. Based on the local extremum set, the filtered image set is subjected to multi-layer shifting processing to obtain the shift feature set; The second retinal image set is obtained by performing wavelet feature fusion processing on the phase feature set based on the shift feature set, including: The shift feature set is subjected to a first wavelet decomposition to obtain a first wavelet set, which includes a first slowly varying component and a first rapidly varying component. The phase feature set is subjected to a second wavelet decomposition to obtain a second wavelet set, which includes a second slowly varying component and a second rapidly varying component. Based on the first rapidly changing component, the second rapidly changing component is subjected to absolute value maximization fusion processing to obtain a fused rapidly changing component; and based on the first slowly changing component, the second slowly changing component is subjected to weighted average fusion processing to obtain a fused slowly changing component. The second retinal image set is obtained based on the fused fast-changing component and the fused slow-changing component.

6. A computer device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method for identifying retinal vessels as described in any one of claims 1-4.

7. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to implement the retinal vessel identification method as described in any one of claims 1-4.

Images