Image processing-based auxiliary diagnosis method and device for glaucoma, and storage medium

Abstract

The application discloses an image processing-based auxiliary glaucoma diagnosis method and device and a storage medium, and belongs to the technical field of glaucoma diagnosis; a glaucoma diagnosis image dataset is constructed, the optic disc region is subjected to phase consistency enhancement, multi-scale structure complexity modeling and anisotropic distance expression, and stable structure features are obtained; a joint feature tensor is generated in combination with a morphological context density map, a multi-scale gradient structure histogram and a morphological flow field divergence significance map, multi-source feature fusion is realized, and a discrimination score is constructed based on a curvature consistency index and a gradient symmetry measure, and a threshold system is used to output health, early and late glaucoma diagnosis results.

Description

Technical Field

[0001] This invention belongs to the field of glaucoma diagnosis technology, specifically relating to glaucoma auxiliary diagnosis methods, devices, and storage media based on image processing. Background Technology

[0002] With the continuous development of medical image analysis technology, the screening and accurate diagnosis of glaucoma has become a key link in ophthalmic diagnosis and preventive treatment. Its diagnosis relies heavily on the quantitative analysis of the morphology of the optic disc region and the structural features of the neuroretina in fundus images. Traditional computer-aided diagnostic methods are mainly based on threshold segmentation, morphological manipulation and manually designed geometric features combined with classifiers for discrimination. These methods have problems such as limited feature expression ability, sensitivity to changes in image contrast and illumination, and poor adaptability to structural lesions at different scales. Especially when the image quality is poor or there is pathological interference, it is easy to lead to misjudgment and missed diagnosis, making it difficult to achieve high-precision and robust automatic diagnosis of glaucoma.

[0003] In recent years, with the in-depth application of computer vision and deep learning technologies, glaucoma diagnosis methods based on convolutional neural networks have emerged. Some studies have improved the perception of lesion areas to a certain extent by introducing multi-scale feature extraction and attention mechanism techniques. Other scholars have combined U-Net segmentation and classification networks to achieve the localization of the optic disc region and the assessment of the degree of lesions. These methods have made initial progress in alleviating the reliance of traditional methods on manual features and improving the end-to-end learning ability of the model.

[0004] However, existing glaucoma-assisted diagnostic methods based on fundus images still have several limitations. Traditional optic disc region enhancement methods have limited ability to express image edges and multi-scale structural features, making it difficult to achieve robust morphological representation in complex fundus backgrounds, resulting in early lesion features being easily interfered with by noise. Secondly, existing feature extraction methods usually process structural and texture information independently, lacking a collaborative fusion mechanism for heterogeneous features of directional gradients and morphological flow fields, which limits the effective integration of multi-dimensional discriminative information. In addition, most classification models are insufficient in jointly modeling key morphological indicators such as the geometric consistency of the optic disc contour and radial texture symmetry, resulting in the need to improve the discrimination accuracy and generalization ability among healthy, early, and late-stage glaucoma. Therefore, it is urgent to propose a novel image processing-based glaucoma-assisted diagnostic method to improve the robustness of optic disc structural representation, the effectiveness of feature fusion, and the accuracy of diagnostic decisions, providing reliable technical support for clinical glaucoma screening and early diagnosis. Summary of the Invention

[0005] This invention provides a glaucoma-assisted diagnostic method, device, and storage medium based on image processing. Addressing the problems of blurred optic disc region boundaries, unstable structural features, and difficulty in effectively fusing multi-scale morphological information in fundus images, this invention proposes a glaucoma-assisted diagnostic model, which consists of an optic disc region enhancement and representation module, a feature collaborative fusion module, and a geometric morphology joint classification module.

[0006] The technical solution adopted by the present invention to achieve the above objectives specifically includes the following steps:

[0007] S1. Collect optic disc fundus images under different health conditions to construct a glaucoma-assisted diagnostic image dataset;

[0008] S2. Based on glaucoma-assisted diagnostic images, phase consistency values ​​are obtained and phase consistency moments are synthesized to enhance edge maps. The adaptive fractal dimension is calculated by dynamically selecting the optimal box size, and a distance transformation map is generated using the normalized gradient vector and anisotropic modulation factor.

[0009] S3. Construct a morphological context density map based on the mean and standard deviation of the adaptive fractal dimension and distance transformation map. Figure 1 On the one hand, density transformation is captured by a sliding window to generate a multi-scale gradient structure histogram; on the other hand, morphological flow field vectors are calculated to generate a morphological flow field divergence saliency map. The two are then fused together using feature tensors to obtain a joint feature tensor.

[0010] S4. Construct a curvature consistency index based on the edge map enhanced by phase consistency moments, calculate the gradient symmetry measure based on the joint feature tensor, generate a discrimination score by combining the curvature consistency index and the gradient symmetry measure, and perform threshold system classification based on the discrimination score to obtain the glaucoma diagnosis result.

[0011] S5 integrates the optic disc region enhancement and characterization module, the feature synergistic fusion module, and the geometric morphology joint classification module to construct a glaucoma auxiliary diagnostic model;

[0012] S6. The glaucoma auxiliary diagnostic image is input into the glaucoma auxiliary diagnostic model, and the glaucoma diagnostic result is finally output.

[0013] Preferably, the method for constructing a glaucoma-assisted diagnostic image dataset involves acquiring clinically collected optic disc fundus images, uniformly converting the optic disc fundus images to ensure consistency in subsequent processing; performing quality screening on the optic disc fundus images, removing blurry, occluded, and abnormally illuminated samples, and uniformly cropping the remaining images to finally construct the glaucoma-assisted diagnostic image dataset.

[0014] Preferably, in step S2, the optic disc region enhancement and characterization module processes the glaucoma auxiliary diagnostic image to obtain the phase consistency value and synthesize the phase consistency moment enhanced edge map. It obtains the adaptive fractal dimension by dynamically selecting the optimal box size pair and uses the normalized gradient vector and unit direction vector to construct an anisotropic modulation factor to generate a distance transformation map.

[0015] Glaucoma-assisted diagnostic image applications The absolute value of the filter's response is combined with the total response to obtain the phase coherence value. The phase coherence values ​​in all directions are then transformed and synthesized using trigonometric moments to generate a phase coherence moment-enhanced edge map.

[0016] Preferably, the size range with the highest linearity in the double logarithmic coordinate system is found, the optimal box size pair is dynamically determined, and logarithmic operations are performed based on the box size pair to obtain the adaptive fractal dimension.

[0017] Preferably, the normalized gradient vector of each point in the phase-consistency moment enhanced edge map is calculated, an anisotropic modulation factor is constructed based on the unit direction vector of the line connecting the vector and the pixel, the Euclidean distance is weighted using the anisotropic modulation factor, and a distance transformation map is generated by finding the minimum modulation distance to the edge point.

[0018] Furthermore, the visual disk region enhancement and representation module obtains a more continuous and illumination-resistant visual disk edge representation through phase consistency moment enhancement, achieves scale-adaptive characterization of the local and overall structural complexity of the visual disk through adaptive fractal dimension, and characterizes the spatial relationship strength from the visual disk edge to the region interior through anisotropically modulated distance transformation map. In summary, the visual disk region enhancement and representation module, through the synergistic effect of edge enhancement, complexity characterization, and spatial relationship modeling, forms a stable, complete, and scale-adaptive visual disk structural representation, providing high-quality basic feature input for subsequent feature fusion and classification.

[0019] Preferably, in step S3, gradient direction and structure distribution information within a local region are obtained through a sliding window based on the morphological context density map, generating a multi-scale gradient structure histogram; the distance transformation map and adaptive fractal dimension are combined, and the mean and standard deviation of the distance transformation map are combined to obtain the morphological context density map; the weighted gradient response of each directional channel is calculated using directional gradient structure histogram encoding to obtain the multi-scale gradient structure histogram; simultaneously, the morphological flow field vectors in the horizontal and vertical directions of the morphological context density map are calculated to obtain the morphological flow field divergence significance map, and finally... The multi-scale gradient structure histogram and the normalized morphological flow field divergence significance map are fused to obtain the joint feature tensor.

[0020] A morphological context density map is generated based on adaptive fractal dimension and distance transformation graph. By combining the global statistical properties of the distance transformation graph, the ability to express the shape of local structures and spatial context is enhanced, resulting in a morphological context density map.

[0021] Preferably, the distance transformation map and the adaptive fractal dimension are combined, and the morphological context density map is obtained by combining the mean and standard deviation of the distance transformation map. The weighted gradient response of each directional channel is calculated by using directional gradient structure histogram encoding to obtain a multi-scale gradient structure histogram.

[0022] Preferably, a morphological flow field vector is constructed based on the morphological context density map, and the morphological context density map is weighted by calculating the rate of change of the morphological flow field in the horizontal and vertical directions to obtain a morphological flow field divergence significance map.

[0023] Preferably, the weight distribution of the spatial dimension and the normalization calculation of the channel dimension are performed based on the multi-scale gradient structure histogram and the morphological flow field divergence significance map, respectively, and then fused to generate a joint feature tensor.

[0024] Furthermore, the feature synergistic fusion module enhances the correlation between local structures and the global spatial background in the visual disk region through morphological context density maps, thereby solving the problem that single gradient or texture features are difficult to reflect complex contextual relationships; it captures directional changes and local texture gradients at different scales through multi-scale gradient structure histograms, allowing fine-grained structural features to be preserved in low-contrast regions; it highlights boundary abrupt changes, structural separation, and abnormal morphological flow regions through morphological flow field divergence saliency maps, giving abnormal structures higher saliency; and it fuses local gradient statistics with the global morphological flow field through joint feature tensors, forming a more stable and discriminative structural representation than single-source features. In summary, the feature synergistic fusion module can achieve synergistic enhancement and unified representation of multi-source structural information in the visual disk region to support subsequent classification and discrimination.

[0025] Preferably, in step S4, edge pixels are extracted based on the phase consistency moment enhanced edge map and the curvature of each boundary point is calculated. A curvature consistency index is constructed based on the standard deviation and mean of the curvature distribution. A radial texture gradient is constructed based on the joint feature tensor. Radial texture analysis is performed based on the radial texture gradient to obtain a gradient symmetry measure. Finally, the curvature consistency index and the gradient symmetry measure are combined to obtain a discrimination score. The final glaucoma diagnosis category is obtained based on the discrimination score through a threshold system.

[0026] Edge pixels are extracted and their curvature is calculated by enhancing the edge map with phase consistency moments. The curvature consistency index is obtained based on the mean and standard deviation of the curvature.

[0027] Preferably, the texture gradient in each radial direction is calculated in polar coordinates, and a gradient symmetry measure is generated by integrating the difference in radial texture gradients.

[0028] Preferably, a discrimination score is calculated based on the curvature consistency index and gradient symmetry measure, and a diagnostic decision for healthy, early, and late-stage glaucoma is made based on the discrimination score.

[0029] Furthermore, by calculating the curvature consistency index based on the phase consistency moment-enhanced edge map, the smoothness of the contour and the regularity of the local structure can be quantified from the curvature changes of the optic disc boundary, thereby effectively identifying structural abnormalities such as boundary deformation, depression, or irregular expansion caused by glaucoma. Based on the joint feature tensor, a gradient symmetry measure can be constructed, allowing for paired analysis of radial textures in polar coordinates. This highlights the symmetric differences in texture intensity and structural distribution in the optic disc region, making the texture imbalance and directional shift presented in the damaged area of ​​the optic disc more significantly expressed. By constructing a discriminant score based on the curvature consistency index and gradient symmetry measure, and combining it with a threshold system for classification, it is beneficial to uniformly quantify geometric morphological features and texture distribution features, enabling healthy, early, and late-stage glaucoma to form separable discriminant intervals in the numerical space. In summary, the geometric morphology joint classification module, through the synergistic use of contour geometric features and texture symmetry features, achieves multi-dimensional capture and comprehensive judgment of optic disc structural abnormalities, providing a stable, interpretable, and discriminative classification basis for glaucoma staging.

[0030] Preferably, in step S5, the optic disc region enhancement and representation module, the feature fusion module, and the geometric morphology joint classification module are integrated to construct a glaucoma-assisted diagnostic model. The glaucoma-assisted diagnostic image is first processed by the optic disc region enhancement and representation module to obtain a phase consistency moment enhanced edge map, an adaptive fractal dimension, and a distance transformation map. The adaptive fractal dimension and distance transformation map are processed by the feature fusion module to obtain a joint feature tensor. The phase consistency moment enhanced edge map and the joint feature tensor are processed by the geometric morphology joint classification module to finally output the glaucoma diagnostic result.

[0031] Furthermore, a computer-readable storage medium storing a glaucoma auxiliary diagnostic program is characterized in that, when the program is executed by a processor, it performs the following steps:

[0032] The optic disc region enhancement and representation module is invoked to perform phase consistency value extraction, trigonometric moment transformation and edge enhancement on the input optic disc fundus image, generating a phase consistency moment enhanced edge map;

[0033] The adaptive fractal dimension is calculated by dynamically selecting the optimal box size, and a distance transformation map is generated by combining the normalized gradient vector and the anisotropic modulation factor.

[0034] Based on the mean and standard deviation of the adaptive fractal dimension and distance transformation map, a morphological context density map is constructed. A multi-scale gradient structure histogram is generated by capturing the density transformation through a sliding window, and the morphological flow field divergence significance map is calculated. The two are then fused together using feature tensors to obtain a joint feature tensor.

[0035] Curvature consistency index is constructed based on the phase consistency moment enhancement edge map, and a discrimination score is generated by combining the gradient symmetry measure calculated by the joint feature tensor. The diagnosis results of healthy, early or late glaucoma are output through threshold system classification.

[0036] The storage medium is integrated into the medical imaging equipment, and the execution process of the program complies with the medical device software verification standards.

[0037] Furthermore, an electronic device for glaucoma auxiliary diagnosis is characterized by comprising:

[0038] Image acquisition interface, used to receive optic disc fundus image data conforming to the DICOM standard;

[0039] The memory stores the optical eye-assisted diagnostic program and pre-trained threshold system parameters;

[0040] The processor is configured to execute programs and generate intermediate results containing quantitative indicators of the visual structure.

[0041] The medical interaction module is used to associate diagnostic results with the patient's electronic medical record (EMR) and send them to the clinician's workstation via an encrypted transmission protocol;

[0042] The display unit is used to visualize the phase consistency moment enhanced edge map, the morphological flow field divergence significance map, and the final diagnostic classification label;

[0043] The electronic device has passed medical device network security certification, and the output format of the diagnostic results conforms to the HL7FHIR medical data exchange standard.

[0044] In summary, due to the adoption of this technical solution, compared with the prior art, the beneficial effects of this invention are as follows: the optic disc region enhancement and characterization module can obtain a more stable edge structure expression and a more scale-adaptive structural complexity characterization, and improve the integrity of the optic disc region structural features through anisotropic modulation of spatial relationship description; the feature synergistic fusion module can establish a synergistic relationship between multi-scale gradient features, morphological flow field features and context density features, so that local structure, global morphology and spatial saliency information are unified and fused; the geometric morphology joint classification module can combine curvature consistency and texture symmetry to construct a comprehensive discrimination method, making the distinction between healthy, early glaucoma and late glaucoma clearer; the three modules together constitute the geometric morphology joint glaucoma auxiliary diagnostic model, so that the model has stable structural expression ability and higher classification discrimination performance. Attached Figure Description

[0045] Figure 1 This is a step-by-step diagram of an image processing-based glaucoma-assisted diagnostic method.

[0046] Figure 2 This is a structural diagram of a glaucoma-assisted diagnostic model.

[0047] Figure 3 This is a structural diagram of the visual region enhancement and characterization module.

[0048] Figure 4 This is a structural diagram of the feature synergy fusion module.

[0049] Figure 5 This is a structural diagram of the geometric morphology joint classification module.

[0050] Figure 6 Input the glaucoma auxiliary diagnosis model result image into the glaucoma auxiliary diagnosis image. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figures 1-6This invention provides a technical solution: a glaucoma-assisted diagnosis method, device, and storage medium based on image processing, including an optic disc region enhancement and characterization module, a feature synergistic fusion module, and a geometric morphology joint classification module. First, the optic disc region enhancement and characterization module achieves edge enhancement and multi-scale fractal characterization of the glaucoma-assisted diagnosis image. Second, the feature synergistic fusion module extracts and fuses salient features of directional gradient structure and morphological flow field. Finally, the geometric morphology joint classification module analyzes the consistency of optic disc contour curvature and radial texture gradient symmetry to construct a comprehensive discrimination score and complete glaucoma-assisted diagnosis based on a threshold system. Specific steps are as follows: Figure 1 As shown.

[0053] Construct a glaucoma-assisted diagnostic model, the structure of which is as follows: Figure 2 As shown.

[0054] S1. Collect optic disc fundus images under different health conditions to construct a glaucoma-assisted diagnostic image dataset;

[0055] Furthermore, the method for constructing a glaucoma-assisted diagnostic image dataset involves acquiring clinically collected optic disc fundus images containing healthy, early, and late-stage glaucoma conditions. These optic disc fundus images are uniformly converted to PNG format to ensure consistency in subsequent processing. The optic disc fundus images undergo quality screening, and samples with blurriness, occlusion, or abnormal illumination are removed. The remaining images are then uniformly cropped, and the image resolution is set to 1024×768. Finally, a glaucoma-assisted diagnostic image dataset of 1500 images is constructed.

[0056] S2. Based on glaucoma-assisted diagnostic images, phase consistency values ​​are obtained and phase consistency moments are synthesized to enhance edge maps. The adaptive fractal dimension is obtained by dynamically selecting the optimal box size pair, and an anisotropic modulation factor is constructed using normalized gradient vectors and unit direction vectors to generate distance transformation maps.

[0057] Furthermore, the structure diagram of the visual disk region enhancement and representation module is as follows: Figure 3 As shown, the mathematical model for calculating phase consistency values ​​based on glaucoma-assisted diagnostic images is as follows:

[0058] ;

[0059] in, Indicated in scale and direction Lower glaucoma auxiliary diagnostic images medium pixel Phase consistency value, Indicates direction index, Indicates scale index. Indicates the total number of scales. The value is 3. Representing scale and direction of The filter; based on the phase coherence value, trigonometric moment transformation and synthesis are performed to obtain the phase coherence moment enhanced edge map. The mathematical model is as follows:

[0060] ;

[0061] in, This represents the edge map of phase consistency moments. Indicates the total number of directions. The value is 4. Indicated in scale and direction The mathematical model for the basis function angle parameters is as follows:

[0062] .

[0063] Furthermore, a method is proposed to dynamically select the optimal box size to generate an adaptive fractal dimension, finding the most linear scale interval in a double logarithmic coordinate system, thus obtaining a more robust and accurate representation of the view disk complexity. The mathematical model is as follows:

[0064] ;

[0065] in, Represents the adaptive fractal dimension. This indicates the dynamic selection of the optimal box size pair. , and These represent the optimal box size. and Lower Coverage Phase Coherence Moment Enhancement Edge Map Number of boxes required The mathematical model is as follows:

[0066] ;

[0067] in, Indicates that the box size is between and box size Box dimensions between The range of values ​​for the side length is , and These represent the dimensions of the box. and Time-covered phase consistency moment enhancement edge map Number of boxes required Indicate box size The number of boxes at that time The Pearson correlation coefficient is represented by the following mathematical model:

[0068] ;

[0069] in, Representing two sequences of variables respectively, in In the mathematical model, the two variable sequences are different and A sequence of logarithms, and These are respectively represented as the first in the two variable sequences. One element, Let represent the means of the two variable sequences, respectively. This indicates a fixed sequence length, with a value of 191.

[0070] Furthermore, we define the normalized gradient vector, with the mathematical model as follows:

[0071] ;

[0072] in, Represents the normalized gradient vector. and Representing pixels The mathematical model for the central difference approximation gradient in the horizontal and vertical directions is as follows:

[0073] ;

[0074] in, Represents the phase coherence moment enhanced edge map At pixel The value at; Indicates by and The constructed two-dimensional gradient vector; simultaneously, the unit direction vector is constructed, and the mathematical model is as follows:

[0075] ;

[0076] in, Indicates starting from the pixel point Point to target pixel A unit direction vector of length 1; based on the normalized gradient vector and unit direction vector An anisotropic modulation factor is constructed, and then weighted with the Euclidean distance to obtain the distance transformation graph. The mathematical model is as follows:

[0077] ;

[0078] in, Represents the distance transformation graph. Represents the phase coherence moment enhanced edge map The set of all edge pixels in the middle. Indicates in Traverse pixel positions within. This represents the pixel position from which the distance is to be calculated. The anisotropic modulation factor is represented by the mathematical model:

[0079] ;

[0080] in This represents the modulation coefficient, with a value of 0.3.

[0081] S3. Combine the distance transformation map and the adaptive fractal dimension, and combine the mean and standard deviation of the distance transformation map to obtain a morphological context density map. Calculate the weighted gradient response of each directional channel using directional gradient structure histogram encoding to obtain a multi-scale gradient structure histogram. Simultaneously, calculate the morphological flow field vectors in the horizontal and vertical directions of the morphological context density map to obtain a morphological flow field divergence significance map. Finally, The multi-scale gradient structure histogram and the normalized morphological flow field divergence significance map are fused to obtain the joint feature tensor.

[0082] Furthermore, the feature collaborative fusion module structure diagram is as follows: Figure 4 As shown, based on the adaptive fractal dimension and distance transformation graph A morphological context density map is generated that can simultaneously encode local structural shape and global spatial context. The mathematical model is as follows:

[0083] ;

[0084] in, Represents a morphological context density map. Represents the adaptive fractal dimension. Represents the distance transformation graph. and Representing distance transformation graphs respectively The mean and standard deviation, Represented by mathematical constants An exponential function with base 0.

[0085] Furthermore, based on the morphological context density map By capturing the gradient direction and structure distribution of density transformation within a local region through a sliding window, a multi-scale gradient structure histogram is obtained. The mathematical model is as follows:

[0086] ;

[0087] in, Indicates direction The following Centered sliding window The multi-scale gradient structure histogram obtained within the inner [database], Represented by pixels A local neighborhood window centered on the element, with a window size of 7×7. , This indicates the preset total number of directions, with a value of 4. This indicates an indicator function; its value is 1 when its internal condition is true, and 0 otherwise. Represents pixels within the window Morphological context density map The gradient direction angle, The gradient vector, Representing the morphological context density map at the pixel level gradient magnitude at that point Representing the morphological context density map at the pixel level Density value at that location, Indicates direction The angle interval, the mathematical model is:

[0088] .

[0089] Furthermore, a morphological flow field vector is defined, and the rate of change of the morphological flow field vector in the horizontal and vertical directions is calculated. Based on this rate of change, the morphological context density map is weighted to obtain a morphological flow field divergence significance map. The mathematical model is as follows:

[0090] ;

[0091] in, This plot represents the significance of the divergence of the morphological flow field. Indicates at point The defined morphological flow field vector is mathematically modeled as follows:

[0092] ;

[0093] in, Represents a morphological context density map. The gradient vector; Represents the morphological flow field vector Rate of change in the horizontal and vertical directions.

[0094] Furthermore, feature tensor collaborative fusion is performed based on the multi-scale gradient structure histogram. Significance of divergence in morphological flow field to be carried out separately The computation and normalization computation are fused to generate a joint feature tensor. The mathematical model is as follows:

[0095] ;

[0096] in, Represents the joint feature tensor. For pixel index, The value represents the significance of the divergence of the morphological flow field at the corresponding location. express The temperature coefficient, used to control the sharpness of the weight distribution, has a value of 1.5.

[0097] S4. Based on the phase consistency moment enhanced edge map, extract edge pixels and calculate the curvature of each boundary point. Construct a curvature consistency index based on the standard deviation and mean of the curvature distribution. Then, calculate the texture gradient in each radial direction in polar coordinates and obtain the gradient symmetry measure by integrating the difference in radial texture gradients. Finally, combine the curvature consistency index and the gradient symmetry measure to obtain a discrimination score. Based on the discrimination score, complete the diagnostic decision for healthy, early, and late-stage glaucoma.

[0098] Furthermore, the structure diagram of the geometric morphology joint classification module is as follows: Figure 5 As shown, the enhanced edge map of the phase consistency moment. Non-maximum suppression is performed to obtain continuous edge lines with a width of one pixel. Pixels on the edge lines are considered boundary points. The curvature of the boundary points is calculated using the following mathematical model:

[0099] ;

[0100] in, Indicates the first The curvature at each boundary point Indicates the first The coordinates of the boundary points and Let these represent the first derivatives in the horizontal and vertical directions, respectively. The mathematical model is as follows:

[0101] ;

[0102] and Let the second derivatives in the horizontal and vertical directions be represented respectively. The mathematical model is as follows:

[0103] ;

[0104] Based on the curvature of the boundary points Given the standard deviation and mean, the curvature consistency index is calculated using the mathematical model:

[0105] ;

[0106] in, This represents the curvature consistency index. and Representing the boundary curvature The standard deviation and mean.

[0107] Furthermore, based on the joint feature tensor Radial texture gradient analysis is performed, and the mathematical model is as follows:

[0108] ;

[0109] in, Represents the gradient symmetry measure. The radial texture gradient is represented by the following mathematical model:

[0110] ;

[0111] in, The gradient vector of the joint feature tensor Norm, Represents radial distance in polar coordinates. ,calculate The mathematical model is as follows:

[0112] ;

[0113] in, The mathematical model for representing the center of the boundary point is as follows:

[0114] ;

[0115] in, They represent the first The x and y coordinates of each boundary point Indicates the number of boundary points.

[0116] Furthermore, a joint judgment function for set morphology is proposed, which is combined with the aforementioned curvature consistency index. and gradient symmetry measure The discrimination score is calculated using the following mathematical model:

[0117] ;

[0118] in, The discrimination score is used to classify data. The mathematical model is as follows:

[0119] ;

[0120] in, This indicates the diagnosis of glaucoma. and This represents the classification threshold, with a value of [value missing]. , .

[0121] S5 integrates the optic disc region enhancement and representation module, the feature synergistic fusion module, and the geometric morphology joint classification module to construct a glaucoma auxiliary diagnostic model.

[0122] Furthermore, glaucoma auxiliary diagnostic models such as Figure 2 As shown, specifically, the glaucoma-assisted diagnostic image is input into the optic disc region enhancement and representation module, and the output includes a phase consistency moment enhanced edge map, an adaptive fractal dimension, and a distance transformation map. The mathematical model is as follows:

[0123] ;

[0124] in, This represents the edge map of phase consistency moments. Represents the adaptive fractal dimension. Represents the distance transformation graph. This represents an image used for auxiliary diagnosis of glaucoma. This indicates the visual disc region enhancement and characterization module;

[0125] The adaptive fractal dimension and distance transformation graph input feature co-fusion module outputs a joint feature tensor, the mathematical model of which is:

[0126] ;

[0127] in, Represents the joint feature tensor. This indicates a feature-based collaborative fusion module;

[0128] The phase-consistency moment enhanced edge map and joint feature tensor input geometric morphology joint classification module outputs the glaucoma diagnosis category. The mathematical model is as follows:

[0129] ;

[0130] in, This indicates the diagnosis of glaucoma. This represents the geometric morphology joint classification module.

[0131] The glaucoma auxiliary diagnostic image is input into the glaucoma auxiliary diagnostic model, and the final output is the glaucoma diagnostic result.

[0132] Furthermore, glaucoma health status results such as Figure 6As shown, after each glaucoma auxiliary diagnostic image is input into the glaucoma diagnostic model, the model outputs the image's discrimination score and the diagnostic results for healthy, early, and late stages.

[0133] Furthermore, when the glaucoma-assisted diagnostic program is executed by the processor, it calls the optic disc region enhancement and characterization module to process the optic disc fundus image conforming to the DICOM standard. The specific implementation process includes: preprocessing the optic disc fundus image, using Gaussian filtering to eliminate image noise, converting the color image to a grayscale image, and enhancing image contrast through histogram equalization. Phase consistency value extraction uses a Log-Gabor filter bank to perform convolution operations on the image at different scales and directions. The filter center frequency is set to 0.05 to 0.4, with 6 directions and a filter bandwidth of 2 octaves for each direction. Trigonometric moment transformation is achieved by calculating the third moment of the phase consistency value, mapping the phase consistency value to a three-dimensional feature space, and synthesizing the phase consistency moment enhancement edge map using a weighted summation method. The weight coefficients are dynamically adjusted according to the morphological characteristics of the optic disc region. The adaptive fractal dimension calculation employs a dynamic box counting method, with box size pairs ranging from 2 to 32 pixels. By traversing different box size combinations, a straight line is fitted in a double logarithmic coordinate system. The Pearson correlation coefficient is calculated to assess the strength of the linear relationship, and the optimal box size pair with a correlation coefficient greater than 0.95 is selected for fractal dimension calculation. The normalized gradient vector is obtained by calculating the image gray-level gradient, and the anisotropic modulation factor is calculated based on the eigenvalues ​​of the image's local structure tensor. Euclidean distance is weighted to generate a distance transformation map. The morphological context density map construction process includes: calculating the statistical characteristics of the adaptive fractal dimension and the distance transformation map; the mean and standard deviation are calculated in a local region using a sliding window method, with the window size set to 15×15 pixels. The multi-scale gradient structure histogram is generated by statistically analyzing the gradient direction distribution at different scales, with scale parameters set to 3, 5, 7, and 9 pixels. At each scale, the 360-degree directional space is divided into 8 intervals. The morphological flow field vector is obtained by calculating the partial derivatives of the morphological context density map. The rates of change in the horizontal and vertical directions are calculated using the Sobel operator. The morphological context density map is weighted to generate a morphological flow field divergence saliency map. Feature tensor co-fusion uses an outer product operation to fuse the multi-scale gradient structure histogram and the morphological flow field divergence saliency map, generating a three-dimensional joint feature tensor. The curvature consistency index is calculated by extracting edge pixels through non-maximum suppression of the phase consistency moment enhancement edge map, using a second-order differential operator to calculate the curvature value, and obtaining the curvature consistency index based on the statistical characteristics of the curvature distribution. The gradient symmetry measure is calculated in polar coordinates. A polar coordinate system is established with the center of the view disk as the origin. Texture gradient values ​​are calculated in 36 radial directions, and the gradient symmetry measure is obtained by integrating the gradient differences between adjacent directions. The discrimination score is obtained by weighted fusion of curvature consistency index and gradient symmetry measure. The weight parameters are determined by cross-validation, and the thresholds are 0.35 and 0.72. The discrimination score is compared with the threshold to classify healthy, early and late glaucoma.The storage medium is integrated into the optical coherence tomography (OCT) scanner. The program execution process follows the IEC 62304 medical device software lifecycle standard, and the software's functional safety is verified through unit testing, integration testing, and system testing. The glaucoma-assisted diagnostic electronic device is equipped with an image acquisition interface supporting the DICOM 3.0 standard protocol, capable of receiving digital image data from fundus cameras and scanning laser ophthalmoscopes. The memory uses a solid-state drive to store the glaucoma-assisted diagnostic program, and the pre-trained threshold system parameters are stored in non-volatile memory. These parameters include a curvature consistency index threshold of 0.68, a gradient symmetry measure threshold of 0.42, and discrimination score classification thresholds of 0.35 and 0.72. The processor uses an ARM Cortex-A72 architecture with a 2.0GHz clock speed and 4GB of DDR4 memory. During program execution, it generates intermediate results such as adaptive fractal dimension, curvature consistency index, and gradient symmetry measure. The medical interaction module connects to the patient's electronic medical record system via an HL7 V3 standard interface, using AES-256 encryption algorithm for encrypted transmission of diagnostic data, and supports the DICOM structured report output format. The display unit uses a 1920×1080 resolution LCD screen. The visual interface displays the pseudo-color rendering results of the phase consistency moment enhanced edge map, the heat map representation of the morphological flow field divergence significance map, and the diagnostic report including health status classification labels. The electronic equipment is ISO 27001 information security certified, the network communication module supports HTTPS protocol and VPN dedicated channel, and the diagnostic result output format is fully compatible with the HL7FHIR R4 standard, supporting both JSON and XML data exchange formats.

Claims

1. A glaucoma-assisted diagnostic method based on image processing, characterized in that, Collect optic disc fundus images under different health conditions to construct a glaucoma-assisted diagnostic image dataset; Phase consistency values ​​are obtained from glaucoma-assisted diagnostic images and phase consistency moments are synthesized to enhance edge maps. The adaptive fractal dimension is calculated by dynamically selecting the optimal box size, and a distance transformation map is generated using the normalized gradient vector and anisotropic modulation factor. Based on the mean and standard deviation of the adaptive fractal dimension and distance transformation map, a morphological context density map is constructed. On the one hand, the morphological context density map captures the density transformation through a sliding window to generate a multi-scale gradient structure histogram. On the other hand, it calculates the morphological flow field vector to generate a morphological flow field divergence significance map. The two are then fused together using feature tensors to obtain a joint feature tensor. Curvature consistency index is constructed based on the phase consistency moment enhancement edge map, gradient symmetry measure is calculated based on the joint feature tensor, and discrimination score is generated by combining curvature consistency index and gradient symmetry measure. Threshold system classification is performed based on the discrimination score to obtain glaucoma diagnosis results. A glaucoma-assisted diagnostic model is constructed by integrating a visual disc region enhancement and characterization module, a feature synergistic fusion module, and a geometric morphology joint classification module. The glaucoma auxiliary diagnostic image is input into the glaucoma auxiliary diagnostic model, and the final output is the glaucoma diagnostic result.

2. The glaucoma-assisted diagnostic method based on image processing according to claim 1, characterized in that, Phase consistency values ​​are extracted from glaucoma-assisted diagnostic images, and trigonometric moment transformation and synthesis are performed on the phase consistency values ​​to enhance the structural edges of the optic disc region, resulting in a phase consistency moment enhanced edge map. An adaptive fractal dimension is extracted from the phase consistency moment enhanced edge map. The stability of scale selection is improved by dynamically selecting the optimal box size pair and combining the linear relationship in the double logarithmic coordinate system with the Pearson correlation coefficient, thus obtaining the adaptive fractal dimension.

3. The glaucoma-assisted diagnostic method based on image processing according to claim 2, characterized in that, Based on the phase-consistent moment enhancement edge map, a normalized gradient vector and a unit direction vector are constructed. The Euclidean distance is weighted using anisotropic modulation factors to obtain the distance transformation map.

4. The glaucoma-assisted diagnostic method based on image processing according to claim 3, characterized in that, A morphological context density map is generated based on adaptive fractal dimension and distance transformation graph. By combining the global statistical properties of the distance transformation graph, the ability to express the shape of local structures and spatial context is enhanced, resulting in a morphological context density map.

5. The glaucoma-assisted diagnostic method based on image processing according to claim 4, characterized in that, Based on the morphological context density map, gradient direction and structural distribution information within a local region are obtained through a sliding window, generating a multi-scale gradient structure histogram.

6. The glaucoma-assisted diagnostic method based on image processing according to claim 5, characterized in that, A morphological flow field vector is constructed based on the morphological context density map. The morphological context density map is then weighted by calculating the rate of change of the morphological flow field in the horizontal and vertical directions to obtain a morphological flow field divergence significance map.

7. The glaucoma-assisted diagnostic method based on image processing according to claim 6, characterized in that, The weight distribution of the spatial dimension and the normalization calculation of the channel dimension are performed based on the multi-scale gradient structure histogram and the morphological flow field divergence significance map, respectively, and then fused to generate a joint feature tensor.

8. The glaucoma-assisted diagnostic method based on image processing according to claim 7, characterized in that, Edge pixels are extracted and curvature is calculated based on the phase consistency moment enhancement edge map non-maximum suppression. The curvature consistency index is obtained based on the mean and standard deviation of the curvature. The texture gradient in each radial direction is calculated in polar coordinates, and a gradient symmetry measure is generated by integrating the difference in radial texture gradients. The discrimination score is calculated based on the curvature consistency index and gradient symmetry measure, and the diagnosis results of healthy, early and late glaucoma are completed based on the discrimination score.

9. A computer-readable storage medium storing a glaucoma auxiliary diagnostic program thereon, characterized in that, When the program is executed by the processor, it performs the following steps: The optic disc region enhancement and representation module is invoked to perform phase consistency value extraction, trigonometric moment transformation and edge enhancement on the input optic disc fundus image, generating a phase consistency moment enhanced edge map; The adaptive fractal dimension is calculated by dynamically selecting the optimal box size, and a distance transformation map is generated by combining the normalized gradient vector and the anisotropic modulation factor. Based on the mean and standard deviation of the adaptive fractal dimension and distance transformation map, a morphological context density map is constructed. A multi-scale gradient structure histogram is generated by capturing the density transformation through a sliding window, and the morphological flow field divergence significance map is calculated. The two are then fused together using feature tensors to obtain a joint feature tensor. Curvature consistency index is constructed based on the phase consistency moment enhancement edge map, and a discrimination score is generated by combining the gradient symmetry measure calculated by the joint feature tensor. The diagnosis results of healthy, early or late glaucoma are output through threshold system classification. The storage medium is integrated into the medical imaging equipment, and the execution process of the program complies with the medical device software verification standards.

10. An electronic device for glaucoma auxiliary diagnosis, characterized in that, include: Image acquisition interface, used to receive optic disc fundus image data conforming to the DICOM standard; The memory stores the optical eye-assisted diagnostic program and pre-trained threshold system parameters; The processor is configured to execute programs and generate intermediate results containing quantitative indicators of the visual structure. The medical interaction module is used to associate diagnostic results with the patient's electronic medical record (EMR) and send them to the clinician's workstation via an encrypted transmission protocol; The display unit is used to visualize the phase consistency moment enhanced edge map, the morphological flow field divergence significance map, and the final diagnostic classification label; The electronic device has passed medical device network security certification, and the output format of the diagnostic results conforms to the HL7FHIR medical data exchange standard.

Images