Fundus image feature extraction method and system based on morphological perception state propagation

Abstract

The present application relates to the technical field of medical image processing, and provides an eye fundus image feature extraction method and system based on shape perception state propagation to solve the problem of poor accuracy of medical image analysis results, which comprises extracting shallow texture features, middle layer structure features and high layer semantic features from the eye fundus image; extracting shape parameters from the high layer semantic features and mapping them into corresponding channel weights, and fusing the channel weights to generate shape guide weights; remodeling the high layer semantic features into corresponding sequences along a set direction, reorganizing them into spatial features, and then enhancing them; based on the shape perception features and the high layer semantic features, high layer fusion features, middle layer fusion features and shallow layer fusion features are obtained through residual fusion or sampling residual fusion, and finally the eye fundus image features are obtained through weighted fusion. It can make the model self-adaptively strengthen the morphological structure coupling representation with pathological significance under the condition of single modal input.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method and system for extracting fundus image features based on morphological perception state propagation. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Accurate diagnosis of fundus diseases heavily relies on pathological morphological clues in medical imaging. OCT (Optical Coherence Tomography) and OCTA (Optical Coherence Tomography Angiography) provide high-dimensional representations of the retinal microstructure and vascular network, respectively. However, existing single-modal feature extraction methods suffer from insufficient scale adaptability and lack of morphological priors. Fundus lesions exhibit significant scale variations, and standard convolutional neural networks with fixed receptive fields struggle to capture both local details and global context: shallow networks easily lose macroscopic patterns, while deep networks annihilate sub-pixel lesions due to downsampling. Strategies such as dilated convolution are incompatible with the non-uniform topology of blood vessels, easily introducing artifacts. Clinical diagnosis relies on quantifiable morphological indicators, but existing models implicitly compress morphological information into a high-dimensional feature space, lacking explicit mapping to clinical parameters. Semantic breaks exist when fusing handcrafted morphological features with deep features, and morphological indicators are not converted into dynamic regulatory signals embedded in the feature evolution process, leading to a disconnect between model decision-making logic and clinical cognition. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for extracting fundus image features based on morphological perception state propagation. This method can use morphological features as a general feature extraction framework for dynamically controlling the injection of signals into the state propagation and nonlinear enhancement processes, enabling the model to adaptively enhance the coupled representation of pathologically significant morphological structures under single-modal input conditions.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for extracting fundus image features based on morphological perception state propagation.

[0006] A method for extracting fundus image features based on morphological perception state propagation, comprising: Extracting superficial texture features, mid-layer structural features, and high-layer semantic features from fundus images; Morphological parameters are extracted from high-level semantic features and mapped to corresponding channel weights. The channel weights are then fused to generate morphological guidance weights. The high-level semantic features are reshaped into corresponding sequences along a set direction. The corresponding sequences are then normalized by layers and propagated to a state. During the state propagation process, morphological guidance weights are introduced as dynamic control signals to inject into the state evolution process. Finally, the sequences are recombined into spatial features through cross-fusion operations. Spatial features are enhanced by morphology-aware learnable nonlinear transformations to obtain morphology-aware features. The morphological perception features and high-level semantic features are residually fused to obtain high-level fused features. These high-level fused features are then sampled and residually fused with mid-level structural features to obtain mid-level fused features. The mid-level fused features are then sampled and residually fused with shallow texture features to obtain shallow fused features. The high-level fused features, mid-level fused features, and shallow fused features are then weighted and fused to obtain fundus image features.

[0007] In one implementation, the morphological parameters include vessel tortuosity and fractal dimension.

[0008] As one implementation method, morphological parameters are mapped to corresponding channel weights using learnable functions; , ; in: For vectorization operations; , , and All are learnable projection matrices; For the Sigmoid function; To correct the linear unit; Input parameters; For the curvature of blood vessels; It is the fractal dimension; For the channel weight of blood vessel tortuosity; The channel weights are for fractal dimension.

[0009] As one implementation method, morphological guidance weights are introduced as dynamic control signals into the state evolution process during state propagation, and their corresponding characterization is as follows: ; ; in: Represents the Hadamard product; To convert a vector into a diagonal matrix; Here is the state transition matrix. For the input projection matrix, To output the projection matrix, For the state dimension; Let t be the latent state guided by morphological prior dynamics. Let be the hidden state at time t; The corresponding sequence is reconstructed from high-level semantic features along a defined direction; This is the output of the state-space model; Weights are assigned based on shape.

[0010] As one implementation method, the process of enhancing spatial features through morphology-aware learnable nonlinear transformation is as follows: ; ; in, For morphological perception features; Learnable nonlinear transformation functions for shape perception; This is a spatial context-enhanced feature representation enhanced by double convolution; and It is a linear projection matrix; It is a set of activation functions after morphological modulation; To correct the linear unit; , For the convolution kernel tensor; Spatial features; and This is the dilated convolution bias vector; This is a two-dimensional convolution operation.

[0011] As one implementation method, the set of activation functions after morphological modulation The first in Each component is set as follows: : ; ; in, The number of grid points; For the first The learnable coefficients of dynamic modulation corresponding to each grid point; For the first The learnable coefficients corresponding to each grid point; For the first The B-spline function corresponding to each grid point; Let the order be the spline order. This is the scaling factor; for The One component; Weighting based on shape; For input parameters.

[0012] As one implementation method, high-level semantic features are unfolded into a sequence along four directions: left to right, top to bottom, right to left, and bottom to top.

[0013] A second aspect of the present invention provides a fundus image feature extraction system based on morphological perception state propagation.

[0014] A fundus image feature extraction system based on morphological perception state propagation, comprising: The multi-layer feature extraction module is used to extract shallow texture features, mid-layer structural features and high-layer semantic features from fundus images; The morphological guidance weight generation module is used to extract morphological parameters from high-level semantic features and map them to corresponding channel weights, and then merge the channel weights to generate morphological guidance weights. The spatial feature reorganization module is used to reshape high-level semantic features into corresponding sequences along a set direction, normalize the corresponding sequences, and then propagate the state. During the state propagation process, morphological guidance weights are introduced as dynamic control signals to inject into the state evolution process, and then the spatial features are reorganized through cross-fusion operations. The morphological sensing feature enhancement module is used to enhance spatial features through a morphological sensing learnable nonlinear transformation to obtain morphological sensing features. The fundus image feature generation module is used to perform residual fusion of morphological perception features and high-level semantic features to obtain high-level fused features, and then sample these features and perform residual fusion with mid-level structural features to obtain mid-level fused features; after sampling the mid-level fused features, perform residual fusion with shallow texture features to obtain shallow fused features; and finally, perform weighted fusion of the high-level fused features, mid-level fused features, and shallow fused features to obtain fundus image features.

[0015] A third aspect of the present invention provides a computer-readable storage medium.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the fundus image feature extraction method based on morphological perception state propagation as described above.

[0017] A fourth aspect of the present invention provides an electronic device.

[0018] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the fundus image feature extraction method based on morphological perception state propagation as described above.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention extracts multi-layer features from fundus images, targets high-level semantic features, extracts morphological parameters from these features and maps them to corresponding channel weights, and then fuses them to generate morphological guidance weights. During state propagation, morphological guidance weights are introduced as dynamic control signals injected into the state evolution process. Through cross-fusion operations, the corresponding sequences of high-level semantic features reshaped along a set direction are reconstructed to form spatial features. The spatial features are enhanced by morphological-aware learnable nonlinear transformation to obtain morphological-aware features. Based on morphological-aware features and high-level semantic features, high-level fusion features, mid-level fusion features, and shallow-level fusion features are obtained through residual fusion or post-sampling residual fusion. Finally, fundus image features are obtained through weighted fusion. Through the dual mechanism of morphological modulation state propagation and morphological-aware nonlinear enhancement, deep collaborative modeling of morphological and structural features in single-modality fundus images is achieved. It has universality and can be adapted to single medical images of different modalities such as OCT and OCTA. The output features can be directly transferred to various downstream diagnostic tasks, and the computational complexity is linear, meeting the real-time requirements of clinical practice. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a flowchart of the fundus image feature extraction method based on morphological perception state propagation according to an embodiment of the present invention; Figure 2 These are fundus images and their corresponding fundus image features according to embodiments of the present invention; Figure 3 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, 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 invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] Figure 1 This is a flowchart of a fundus image feature extraction method based on morphological perception state propagation, according to an embodiment of the present invention. Figure 1 As shown in the figure, the fundus image feature extraction method based on morphological perception state propagation in this invention includes the following steps: Step S101: Extract shallow texture features, mid-level structural features, and high-level semantic features from fundus images.

[0026] Let the input fundus image be ,in and These are the space height and width, respectively. This refers to the number of channels. In this embodiment, a three-stage convolutional encoder is used to extract multi-scale local features: ; ; ; in, It represents shallow texture features, preserves the microstructure of the retinal surface (such as the texture of the nerve fiber layer and the primary outline of microvessels), and has rich spatial details but low semantic abstraction. It represents the characteristics of the middle layer structure, encodes the morphology of the middle layer of the retina (such as the outer plexiform layer boundary and microaneurysm clusters), and balances local details with regional context. It represents high-level semantic features and characterizes the overall topology of lesions (such as CNV neovascular network skeleton and Drusen spatial distribution pattern), providing a stable semantic basis for morphological feature extraction; Represents a two-dimensional convolution operation; For the convolution kernel tensor, ; For bias vectors, ; The dimension corresponds to the hierarchical features; For layer normalization function, This is to correct the linear unit. This effectively captures abnormalities in the retinal laminar structure (such as retinal thickening in DR) and local vascular texture patterns, providing a basic representation for subsequent morphological modeling.

[0027] It should be noted here that, in addition to the three-stage convolutional encoder, the ResNet backbone network can also be used to extract the outputs of conv2_x, conv3_x, and conv4_x as... , , We utilize residual connections to enhance gradient propagation.

[0028] Step S102: Extract morphological parameters from high-level semantic features and map them to corresponding channel weights, then fuse the channel weights to generate morphological guidance weights.

[0029] In this embodiment, the morphological parameters include blood vessel tortuosity. and fractal dimension .

[0030] Specifically, the Hessian matrix is ​​calculated using image intensity. The vascular skeleton is extracted through Hessian matrix eigenvalue analysis, and then a binary vascular mask is obtained. , ; in For spatial coordinates, For image intensity, These are the eigenvalues ​​of the Hessian matrix. Blood vessel pixels satisfy... ,and Based on this, a binary vascular mask is generated. .

[0031] Along the center line of the blood vessel Calculate curvature, i.e., the tortuosity of blood vessels. ; ; in, This is the arc length parameter.

[0032] Fractal properties of vascular networks were calculated using box counting: ; in, To cover the vascular network Scale grid number.

[0033] The morphological parameters are mapped to corresponding channel weights using learnable functions; , ; in: For vectorization operations; , , and All are learnable projection matrices; For the Sigmoid function; To correct the linear unit; Input parameters; For the curvature of blood vessels; It is the fractal dimension; For the channel weight of blood vessel tortuosity; The channel weights are for fractal dimension.

[0034] Weighting of blood vessel curvature channels With fractal dimension channel weights Fusion, generating form-guided weights : This explicit encoding of clinically validated morphological indicators provides pathological priors for subsequent state propagation.

[0035] Step S103: Reshape the high-level semantic features into corresponding sequences along the set direction, normalize the corresponding sequences by layer and then propagate the state. In the process of state propagation, introduce morphological guiding weights as dynamic control signals to inject into the state evolution process, and then reorganize them into spatial features through cross-fusion operations.

[0036] Will Remodeling into a sequence After layer normalization, state propagation is performed; among which, Traditional state-space models update hidden states using the following equation. : ; in, Here is the state transition matrix. For the input projection matrix, To output the projection matrix, For the state dimension; A morphological modulation mechanism is introduced, in which morphological guiding weights are injected as dynamic control signals into the state evolution process during state propagation. The corresponding characterization is as follows: ; ; in: Represents the Hadamard product; To convert a vector into a diagonal matrix; Here is the state transition matrix. For the input projection matrix, To output the projection matrix, For the state dimension; Let t be the latent state guided by morphological prior dynamics. Let be the hidden state at time t; The corresponding sequence is reconstructed from high-level semantic features along a defined direction; This is the output of the state-space model; Weights are assigned based on shape.

[0037] To capture multi-directional spatial dependencies, a four-directional cross-directional scanning mechanism (CDS) is designed: high-level semantic features are unfolded into sequences along four directions: left to right, top to bottom, right to left, and bottom to top. After propagation through the aforementioned morphological modulation states, they are recombined into spatial features through a cross-merge operation. This mechanism achieves linear complexity. Its global dependency modeling is significantly better than that of Transformer. Complexity.

[0038] Step S104: Enhance the spatial features through a morphology-aware learnable nonlinear transformation to obtain morphology-aware features.

[0039] against The design shape undergoes adaptive nonlinear transformation. Traditional KANs utilize learnable B-spline functions. Parameterized activation function: ; A morphology-aware mechanism is introduced to dynamically modulate the B-spline coefficients based on morphological features, thereby enabling the nonlinear transformation to adaptively respond to morphological characteristics: enhancing the nonlinear intensity in high curvature regions to capture details of vascular tortuosity, and smoothing the transformation in low fractal dimension regions to suppress noise interference.

[0040] Activation function after morphological modulation The first in Each component is set as follows: : ; ; in, The number of grid points; For the first The learnable coefficients of dynamic modulation corresponding to each grid point; For the first The learnable coefficients corresponding to each grid point; For the first The B-spline function corresponding to each grid point; Let the order be the spline order. This is the scaling factor; for The One component; Weighting based on shape; For input parameters.

[0041] To enhance spatial information interaction, a two-layer dilated convolution is applied beforehand: ; The process of enhancing spatial features through morphologically perceptible learnable nonlinear transformation is as follows: ; in, For morphological perception features; Learnable nonlinear transformation functions for shape perception; This is a spatial context-enhanced feature representation enhanced by double convolution; and It is a linear projection matrix; It is a set of activation functions after morphological modulation; To correct the linear unit; , For the convolution kernel tensor; Spatial features; and This is the dilated convolution bias vector; This is a two-dimensional convolution operation.

[0042] Step S105: Perform residual fusion of morphological perception features and high-level semantic features to obtain high-level fused features, sample them, and then perform residual fusion with mid-level structural features to obtain mid-level fused features; sample the mid-level fused features and then perform residual fusion with shallow texture features to obtain shallow fused features; perform weighted fusion of high-level fused features, mid-level fused features, and shallow fused features to obtain fundus image features.

[0043] The residual fusion mechanism is designed to preserve shallow detail features: ; ; ; in This is a bilinear upsampling operation; Characterized by high-level integration; This is a mid-level fusion feature; This is a shallow fusion feature.

[0044] Introducing learnable parameters That is, the weights corresponding to the fused features, to achieve adaptive weighting of multi-scale features: ; initial Automatic optimization during training. Final output. As a highly discriminative feature representation, it can be directly input into downstream task modules (such as global average pooling + fully connected layers for classification).

[0045] To enhance feature discriminativeness, a contrastive learning loss is employed. Constrained feature space distribution: ; in For the first The feature vector of each sample For its positive samples (different enhanced views of the same category). For cosine similarity, This refers to the temperature parameter.

[0046] Table 1 presents a performance comparison between the present invention and traditional models. Analysis of Table 1 shows that the morphological state propagation method for fundus image feature extraction based on morphological perception state propagation provided in this embodiment of the invention has a linear computational complexity compared to standard Transformer and CNN models; moreover, it performs well in terms of inference time, memory usage, and average accuracy. The fundus image feature extraction method based on morphological perception state propagation provided in this embodiment of the invention is universal, adaptable to single medical images of different modalities such as OCT and OCTTA, and its output features can be directly transferred to various downstream diagnostic tasks. Furthermore, its computational complexity is linear, meeting the real-time requirements of clinical practice.

[0047] Table 1. Performance comparison results of different models;

[0048] In Table 1, the average accuracy (Acc) is one of the most basic and widely used performance metrics in machine learning and artificial intelligence evaluation, representing the proportion of samples that the model correctly predicts out of the total number of samples.

[0049] The standard Transformer is a deep learning model architecture based on the self-attention mechanism, proposed by Google.

[0050] CNN (Deep Stacked) refers to a deeply stacked convolutional neural network. It constructs a deep neural network by alternately stacking multiple convolutional layers and pooling layers to achieve hierarchical feature extraction from grid data such as images. This deep structure can gradually abstract high-level features of object parts and even complete semantics from low-level features such as edges and textures, and is the core foundation of modern computer vision tasks.

[0051] The fundus image feature extraction system based on morphology-aware state propagation provided in this embodiment of the invention can be implemented in software. The fundus image feature extraction system based on morphology-aware state propagation includes the following software modules: The multi-layer feature extraction module is used to extract shallow texture features, mid-layer structural features and high-layer semantic features from fundus images; The morphological guidance weight generation module is used to extract morphological parameters from high-level semantic features and map them to corresponding channel weights, and then merge the channel weights to generate morphological guidance weights. The spatial feature reorganization module is used to reshape high-level semantic features into corresponding sequences along a set direction, normalize the corresponding sequences, and then propagate the state. During the state propagation process, morphological guidance weights are introduced as dynamic control signals to inject into the state evolution process, and then the spatial features are reorganized through cross-fusion operations. The morphological sensing feature enhancement module is used to enhance spatial features through a morphological sensing learnable nonlinear transformation to obtain morphological sensing features. The fundus image feature generation module is used to perform residual fusion of morphological perception features and high-level semantic features to obtain high-level fused features, and then sample these features and perform residual fusion with mid-level structural features to obtain mid-level fused features; after sampling the mid-level fused features, perform residual fusion with shallow texture features to obtain shallow fused features; and finally, perform weighted fusion of the high-level fused features, mid-level fused features, and shallow fused features to obtain fundus image features.

[0052] It should be noted that each module in the fundus image feature extraction system based on morphology-aware state propagation in this embodiment corresponds one-to-one with each step in the fundus image feature extraction method based on morphology-aware state propagation in the above embodiment, and their specific implementation processes are the same, so they will not be repeated here.

[0053] The structure of the electronic device according to an embodiment of the present invention will be described in detail below. Figure 3 This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of the present invention. It can be understood that... Figure 3 The diagram shows only an exemplary structure of the electronic device, not the entire structure. Some or all of the structures shown may be implemented as needed.

[0054] The electronic device provided in this embodiment of the invention includes: at least one processor 301, a memory 302, a user interface 303, and at least one network interface 304. The various components in the morphologically-aware state propagation-based fundus image feature extraction system are coupled together via a bus system 305. It can be understood that the bus system 305 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 305 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 3 The general designated all buses as Bus System 305.

[0055] The user interface 303 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0056] It is understood that memory 302 can be volatile memory or non-volatile memory, or both. In this embodiment of the invention, memory 302 is capable of storing data to support the operation of the terminal. Examples of this data include any computer programs used to operate on the terminal, such as operating systems and applications. The operating system includes various system programs, such as framework layers, core library layers, driver layers, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications.

[0057] In some embodiments, the fundus image feature extraction system based on morphology-aware state propagation provided in this invention can be implemented using a combination of hardware and software. For example, the fundus image feature extraction system based on morphology-aware state propagation provided in this invention can be a processor in the form of a hardware decoding processor, programmed to execute the fundus image feature extraction method based on morphology-aware state propagation provided in this invention. For instance, the processor in the form of a hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0058] As an example, processor 301 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0059] As an example of the hardware implementation of the fundus image feature extraction system based on morphology-aware state propagation provided in this embodiment of the invention, the device provided in this embodiment of the invention can be directly executed by a processor 301 in the form of a hardware decoding processor. For example, it can be executed by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the fundus image feature extraction method based on morphology-aware state propagation provided in this embodiment of the invention.

[0060] The memory 302 in this embodiment of the invention is used to store various types of data to support the operation of the fundus image feature extraction system based on morphological perception state propagation, or to store data for execution. Figure 1 The program code for the method shown. Examples of this data include: any executable instructions for operation on a morphology-aware state propagation-based fundus image feature extraction system, such as executable instructions that can be included in the executable instructions to implement the morphology-aware state propagation-based fundus image feature extraction method of the embodiments of the present invention.

[0061] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including functions for executing... Figure 1 The program code for the method shown. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by the central processing unit, it performs the various functions defined in the apparatus of this application.

[0062] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting fundus image features based on morphological perception state propagation, characterized in that, include: Extracting superficial texture features, mid-layer structural features, and high-layer semantic features from fundus images; Morphological parameters are extracted from high-level semantic features and mapped to corresponding channel weights. The channel weights are then fused to generate morphological guidance weights. The high-level semantic features are reshaped into corresponding sequences along a set direction. The corresponding sequences are then normalized by layers and propagated to a state. During the state propagation process, morphological guidance weights are introduced as dynamic control signals to inject into the state evolution process. Finally, the sequences are recombined into spatial features through cross-fusion operations. Spatial features are enhanced by morphology-aware learnable nonlinear transformations to obtain morphology-aware features. The morphological perception features and high-level semantic features are residually fused to obtain high-level fused features. These high-level fused features are then sampled and residually fused with mid-level structural features to obtain mid-level fused features. The mid-level fused features are then sampled and residually fused with shallow texture features to obtain shallow fused features. The high-level fused features, mid-level fused features, and shallow fused features are then weighted and fused to obtain fundus image features.

2. The fundus image feature extraction method based on morphological perception state propagation as described in claim 1, characterized in that, The morphological parameters include vessel tortuosity and fractal dimension.

3. The fundus image feature extraction method based on morphological perception state propagation as described in claim 2, characterized in that, The morphological parameters are mapped to corresponding channel weights using learnable functions; , ； in: For vectorization operations; , , and All are learnable projection matrices; For the Sigmoid function; To correct the linear unit; Input parameters; For the curvature of blood vessels; It is the fractal dimension; For the channel weight of blood vessel tortuosity; The channel weights are for fractal dimension.

4. The fundus image feature extraction method based on morphological perception state propagation as described in claim 1, characterized in that, In the state propagation process, morphological guidance weights are introduced as dynamic control signals injected into the state evolution process, and their corresponding characterization is as follows: ； ； in: Represents the Hadamard product; To convert a vector into a diagonal matrix; Here is the state transition matrix. For the input projection matrix, To output the projection matrix, For the state dimension; Let t be the latent state guided by morphological prior dynamics. Let be the hidden state at time t; The corresponding sequence is reconstructed from high-level semantic features along a defined direction; This is the output of the state-space model; Weights are assigned based on shape.

5. The fundus image feature extraction method based on morphological perception state propagation as described in claim 1, characterized in that, The process of enhancing spatial features through morphologically perceptible learnable nonlinear transformation is as follows: ； ； in, For morphological perception features; Learnable nonlinear transformation functions for shape perception; This is a spatial context-enhanced feature representation enhanced by double convolution; and It is a linear projection matrix; It is a set of activation functions after morphological modulation; To correct the linear unit; , For the convolution kernel tensor; Spatial features; and This is the dilated convolution bias vector; This is a two-dimensional convolution operation.

6. The fundus image feature extraction method based on morphological perception state propagation as described in claim 5, characterized in that, Activation function set after morphological modulation The first in Each component is set as follows: : ； ； in, The number of grid points; For the first The learnable coefficients of dynamic modulation corresponding to each grid point; For the first The learnable coefficients corresponding to each grid point; For the first The B-spline function corresponding to each grid point; Let the order be the spline order. This is the scaling factor; for The One component; Weighting based on shape; For input parameters.

7. The fundus image feature extraction method based on morphological perception state propagation as described in claim 1, characterized in that, The high-level semantic features are unfolded into a sequence along four directions: left to right, top to bottom, right to left, and bottom to top.

8. A fundus image feature extraction system based on morphological perception state propagation, characterized in that, The method is implemented using the steps of the fundus image feature extraction method based on morphological perception state propagation as described in any one of claims 1-7, including: The multi-layer feature extraction module is used to extract shallow texture features, mid-layer structural features and high-layer semantic features from fundus images; The morphological guidance weight generation module is used to extract morphological parameters from high-level semantic features and map them to corresponding channel weights, and then merge the channel weights to generate morphological guidance weights. The spatial feature reorganization module is used to reshape high-level semantic features into corresponding sequences along a set direction, normalize the corresponding sequences, and then propagate the state. During the state propagation process, morphological guidance weights are introduced as dynamic control signals to inject into the state evolution process, and then the spatial features are reorganized through cross-fusion operations. The morphological sensing feature enhancement module is used to enhance spatial features through a morphological sensing learnable nonlinear transformation to obtain morphological sensing features. The fundus image feature generation module is used to perform residual fusion of morphological perception features and high-level semantic features to obtain high-level fused features, and then sample these features and perform residual fusion with mid-level structural features to obtain mid-level fused features; after sampling the mid-level fused features, perform residual fusion with shallow texture features to obtain shallow fused features; and finally, the fundus image features are obtained by weighted fusion of the high-level fused features, mid-level fused features, and shallow fused features.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps in the fundus image feature extraction method based on morphological perception state propagation as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the fundus image feature extraction method based on morphological perception state propagation as described in any one of claims 1-7.

Images