An optical coherence tomography-based ophthalmic disease diagnosis system

Abstract

This invention discloses an ophthalmic disease diagnostic system based on optical coherence tomography (OCT), comprising: an image data acquisition module for acquiring OCT image data; a feature extraction and similarity calculation module for extracting OCT image feature representations and layered structural features of the macular region, calculating macular structural similarity information between different OCT image data, and constructing a structural similarity alignment matrix reflecting the structural similarity among all OCT images; simultaneously constructing an association matrix representing the correspondence between OCT images and disease categories; and a model training module that introduces the structural similarity alignment matrix and association matrix as joint supervision signals into the objective function, adaptively learning to obtain an ophthalmic disease diagnostic model; the ophthalmic disease diagnostic model outputs a classification diagnostic result representing the probability that the input OCT image data belongs to various ophthalmic diseases. This invention achieves automatic identification and classification of ophthalmic diseases, with high diagnostic accuracy and good robustness.

Description

Technical Field

[0001] This invention relates to the field of computer-aided disease diagnosis technology, and in particular to an ophthalmic disease diagnosis system based on optical coherence tomography. Background Technology

[0002] The macula is a crucial structure in the retina responsible for central and fine visual functions, and its diseases are highly prevalent and harmful. Abnormal changes in the macula can easily lead to various retinal diseases, particularly impacting the development and progression of age-related macular degeneration, causing significant damage to visual function. Currently, optical coherence tomography (OCT) is commonly used clinically to image and analyze the macula and retinal structures. OCT provides high-resolution tomographic images of various retinal layers and has become an important tool for the diagnosis and follow-up of macular-related diseases.

[0003] Traditional OCT image-based interpretation of ophthalmic diseases relies heavily on human experience, resulting in low diagnostic efficiency, high subjectivity, and insufficient result stability. Meanwhile, existing automated analysis methods still fall short in feature representation efficiency, adaptability to macular laminar structures, and the ability to differentiate between multiple disease categories, making it difficult to balance diagnostic accuracy with computational efficiency. This limits the further application of OCT images in the automated diagnosis of ophthalmic diseases.

[0004] Currently, there are various OCT image analysis methods based on computational models, including methods based on graph topological similarity, semi-supervised learning, traditional machine learning, and graph neural networks. However, these methods still have shortcomings. While they have improved the classification performance of OCT images to some extent, systematic comparative studies on multi-class OCT image classification are still limited. In particular, there is a lack of research on unified and standardized evaluation of various deep learning models on multiple public datasets, making it difficult to comprehensively evaluate the generalization performance and applicability of different methods. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an ophthalmic disease diagnostic system based on optical coherence tomography (OCT). This invention simultaneously uses deep feature learning and structural constraint mechanisms to model and analyze OCT images, thereby achieving automatic identification and classification of ophthalmic diseases, exhibiting high diagnostic accuracy and good robustness.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an ophthalmic disease diagnostic system based on optical coherence tomography, comprising: The image data acquisition module is configured to acquire OCT image data for the diagnosis of ophthalmic diseases and to preprocess the OCT image data. The feature extraction and similarity calculation module is configured to: extract OCT image feature representations based on OCT image data, extract the layered structure features of the macular region in the OCT image data, calculate the macular structure similarity information between different OCT image data, and construct a structural similarity alignment matrix reflecting the entire OCT image; at the same time, obtain disease category label information based on OCT image data and construct an association matrix representing the correspondence between OCT images and disease categories. The model training module is configured to: construct an objective function based on structural constraint learning, and introduce the structural similarity alignment matrix and the correlation matrix as joint supervision signals into the objective function to adaptively learn and obtain an ophthalmic disease diagnosis model; The diagnostic prediction module is configured to: acquire OCT image data of the patient to be diagnosed as input to the ophthalmic disease diagnostic model, perform forward propagation processing, and output a classification diagnostic result indicating the probability of the input OCT image data belonging to various ophthalmic diseases.

[0007] As a further technical solution, the preprocessing of OCT image data includes at least one of image normalization, noise suppression, and retinal region extraction.

[0008] As a further technical solution, disease category label information can be obtained by combining the manual annotation results or the prior attributes of the OCT image data.

[0009] As a further technical solution, the similarity information of macular structure between different OCT image data is calculated. The specific method is as follows: based on the Mahalanobis distance, the feature covariance matrix is ​​introduced to normalize and correct the feature expression space, and the Gaussian kernel function is used to shrink the mapping relationship of the feature expression space to calculate the similarity information of macular structure between different OCT image data.

[0010] As a further technical solution, the similarity information of the macular structure is calculated, specifically represented as follows: ; in, For the first Page and the first Results of macular structure similarity measurement between OCT image data. and Based on the first Page and the first OCT image feature representation extracted from OCT image data. Let be the characteristic covariance matrix.

[0011] As a further technical solution, the specific expression of the objective function based on structural constraint learning is as follows: ; in, The objective function for learning based on structural constraints is... For the classification supervision loss function, The structural constraint feature loss function, and These are the weight hyperparameters used in adaptive optimization.

[0012] As a further technical solution, the classification supervision loss function is used to dominate the overall accuracy of disease diagnosis prediction, and it is expressed as follows: ;in, Indicates the true label, This represents the model's predicted output.

[0013] As a further technical solution, the structural constraint feature loss function is used to enable the model to learn and maintain the consistency of the underlying anatomical structure, and it is expressed as follows: ; in, The elements in the structural similarity alignment matrix represent the macular structural similarity information. and They represent the first The and the first Feature vectors output from OCT image data through a deep convolutional neural network model This represents the squared Euclidean distance between two eigenvectors.

[0014] As a further technical solution, the objective function is optimized and updated using a gradient descent algorithm until convergence is obtained to obtain a trained ophthalmic disease diagnostic model. The optimization and update process is represented as follows: ; in, and These are all network model parameters. For learning rate, The objective function is the learning objective function based on structural constraints.

[0015] As a further technical solution, the ophthalmic disease diagnosis model takes the OCT image data of the patient to be diagnosed as input, and outputs a classification diagnosis result representing the probability of the input OCT image data belonging to various ophthalmic diseases. Its mapping expression is as follows: ; in, and Both are mapping functions. For the OCT image data of the patient to be diagnosed, This is used to represent the classification and diagnostic results of the probability of the input OCT image data belonging to various ophthalmic diseases.

[0016] One or more technical solutions of the present invention have the following beneficial effects: This invention abandons manual, experience-based interpretation and instead employs automated processing based on model algorithms. This solves the problems of low diagnostic efficiency, high subjectivity, and insufficient result stability, providing consistent and objective diagnostic results, thereby significantly improving diagnostic efficiency and stability. Furthermore, through the supervisory signal of a "structural similarity alignment matrix," the model is constrained in the feature space, resulting in images with similar macular structures having more similar feature representations. This enhances the ability to capture lesion morphological features and helps to more accurately distinguish different ophthalmic diseases in complex scenarios.

[0017] This invention uses a unified similarity constraint learning framework to jointly optimize the structural feature representation of OCT image data and disease classification results. Compared with traditional methods that rely on a single feature or empirical interpretation, this framework can provide more robust diagnostic performance and effectively improve the recognition accuracy of multiple types of ophthalmic diseases. It has good generalization ability and can be used for auxiliary clinical screening and diagnosis of different types of ophthalmic diseases. Attached Figure Description

[0018] 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.

[0019] Figure 1 This is a schematic diagram of the structure of an ophthalmic disease diagnostic system based on optical coherence tomography according to the present invention; Figure 2 This is a schematic diagram illustrating the workflow of an ophthalmic disease diagnostic system based on optical coherence tomography (OCT) according to the present invention. Figure 3 This is a schematic diagram illustrating the extraction of macular structural features and the calculation of similarity association topology based on OCT image data according to the present invention. Detailed Implementation

[0020] 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 in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] Example 1 like Figure 1and Figure 2 As shown, this embodiment provides an ophthalmic disease diagnostic system based on optical coherence tomography (OCT). This system integrates deep feature learning from OCT image data with a constraint mechanism based on macular structure similarity to achieve automated and high-precision diagnosis of various ophthalmic diseases. The system specifically includes the following modules: The image data acquisition module is configured to acquire OCT image data for the diagnosis of ophthalmic diseases and to preprocess the OCT image data. The feature extraction and similarity calculation module is configured to: extract OCT image feature representations based on OCT image data, extract the layered structure features of the macular region in the OCT image data, calculate the macular structure similarity information between different OCT image data, and construct a structural similarity alignment matrix reflecting the entire OCT image; at the same time, obtain disease category label information based on OCT image data and construct an association matrix representing the correspondence between OCT images and disease categories. The model training module is configured to: construct an objective function based on structural constraint learning, and introduce the structural similarity alignment matrix and the correlation matrix as joint supervision signals into the objective function to adaptively learn and obtain an ophthalmic disease diagnosis model; The diagnostic prediction module is configured to: acquire OCT image data of the patient to be diagnosed as input to the ophthalmic disease diagnostic model, perform forward propagation processing, and output a classification diagnostic result indicating the probability of the input OCT image data belonging to various ophthalmic diseases.

[0022] In the image data acquisition module, OCT image data for the diagnosis of ophthalmic diseases is first acquired, and the OCT image data is preprocessed, including at least one of image normalization, noise suppression, and retinal region extraction.

[0023] In the feature extraction and similarity calculation module, based on the preprocessed OCT image data, the OCT image feature representation and the layered structure features of the macular region in the OCT image data are first extracted. Based on the OCT image feature representation and the layered structure features of the macular region, the macular structure similarity information (structural similarity) between different OCT image data is calculated. Then, based on the macular structure similarity information, a structural similarity alignment matrix reflecting the structural similarity among all OCT images is constructed. .

[0024] In this embodiment, based on preprocessed OCT image data, core structural features reflecting the health status of the macular region are extracted. To overcome the problem of varying scales of local lesions (such as cystic edema and drusen) in OCT image data and to enhance morphological correlation, arbitrary lesions caused by reflection of light from different biological tissue layers are represented. Page and the first For each OCT image data set, the corresponding deep network feature vectors are constructed and obtained. and (Image feature representation).

[0025] Furthermore, based on the Mahalanobis distance, a feature covariance matrix is ​​introduced to normalize and correct the feature representation space, and a Gaussian kernel function is used to shrink the mapping relationship of the feature representation space. The similarity information of the macular structure between different OCT image data is then calculated, specifically as follows: ; in, For the first Page and the first Results of macular structure similarity measurement between OCT image data. and Based on the first Page and the first OCT image feature representation extracted from OCT image data. Let be the characteristic covariance matrix.

[0026] Using this method to calculate macular structure similarity information not only integrates the distribution density of the data itself, but also weakens the consistency contamination interference caused by the difference in background noise of different scanners to a certain extent, which greatly improves the fidelity of comparing similar features of multidimensional complex fundus data.

[0027] Meanwhile, in the feature extraction and similarity calculation module, disease category label information is obtained based on the manual annotation results corresponding to the OCT image data, and an association matrix representing the correspondence between OCT images and disease categories is constructed.

[0028] The correspondence between OCT image data and disease categories can be directly obtained from the OCT image dataset, where each OCT image corresponds to a pre-labeled disease category. The OCT image dataset contains multiple OCT image samples and various ophthalmic disease categories. For ease of description, a set is used. To represent a set of OCT image data, use the set. This represents a set of disease categories.

[0029] To accurately represent the relationship between OCT image data and disease categories, a matrix is ​​used. This represents the association matrix between OCT image data and disease categories. Specifically, when the... OCT image data Corresponding to disease category At that time, the elements in the matrix The value is set to When OCT image data does not correspond to the disease category, The value is set to .

[0030] Therefore, the correlation matrix The The line represents the first The feature vectors of association between OCT images and all disease categories, and the matrix The The column represents all OCT images and the first one. The feature vectors relating to different disease categories. For example... Figure 3 As shown, the OCT image dataset covers a variety of common ophthalmic diseases, providing a reliable data foundation for the research and implementation of automatic diagnostic methods for ophthalmic diseases based on OCT images.

[0031] In the model training module, a structural similarity alignment matrix containing implicit topological relationships between features is obtained. and related networks Subsequently, a unified constraint optimization solution framework was established based on structural similarity properties: if two test or training OCT tomographic images are extremely similar in feature space (i.e., ... If the value is close to 1, then the classification label decision results corresponding to these two samples should also have a very high probability of belonging to the same disease category in the final prediction space.

[0032] The algorithm utilizes a supervised learning strategy to adaptively learn and update the corresponding discriminant function and its weight parameters. Once the unknown lesion image during the testing process is mapped to a probability by the model output, and then passed through a multi-classifier (e.g., Softmax), the system outputs a vector of associated probability values. The eye disease name corresponding to the dimension position index of the maximum associated probability is extracted as the final diagnostic label, thus completing the entire auxiliary diagnostic function based on pathological morphology and correlation constraints.

[0033] In the model training module, the objective function is learned based on structural constraints. The specific expression is: ; in, The objective function for learning based on structural constraints is... For the classification supervision loss function, The structural constraint feature loss function, and These are the weight hyperparameters used in adaptive optimization.

[0034] The classification-supervised loss function is used to dominate the overall accuracy of disease diagnosis prediction, and it is expressed as follows: ;in, Indicates the true label, This represents the model's predicted output.

[0035] The structural constraint feature loss function is used to enable the model to learn and maintain the consistency of the underlying anatomical structure, and it is expressed as: ; in, The elements in the structural similarity alignment matrix represent the macular structural similarity information. and They represent the first The and the first Feature vectors output from OCT image data through a deep convolutional neural network model This represents the squared Euclidean distance between two eigenvectors.

[0036] The physical meaning of the structural constraint feature loss function lies in the result of the macular structural similarity measurement. The higher the OCT image, the closer the distance between the two images in the feature space is forced, thus prompting the model to learn and maintain the consistency of the underlying anatomical structure.

[0037] The objective function is optimized and updated using the gradient descent algorithm until convergence is obtained, resulting in a trained ophthalmic disease diagnostic model. The optimization and update process is represented as follows: ; in, and These are all network model parameters. For learning rate, The objective function is the learning objective function based on structural constraints.

[0038] The ophthalmic disease diagnostic model takes the OCT image data of the patient to be diagnosed as input, and outputs a classification diagnostic result representing the probability of the input OCT image data belonging to various ophthalmic diseases. Its mapping expression is as follows: ; in, and Both are mapping functions. For the OCT image data of the patient to be diagnosed, This is used to represent the classification and diagnostic results of the probability of the input OCT image data belonging to various ophthalmic diseases.

[0039] To objectively evaluate the performance of this embodiment's model compared to the baseline and other models using only residual architectures or graph topology networks, comparative statistics were designed on publicly available ophthalmic disease datasets: all OCT images used for experimental validation were standardized to... .

[0040] A fair experiment was conducted, and the optimizer was configured with adaptive convergence SGD. The results show that fusing similarity constraints with statistical correction of the covariance matrix can greatly reduce the ambiguity of feature decisions and stabilize the robustness of the model when the amount of data is limited or the category features are coupled (see the comparison of comprehensive performance parameters in Table 1, where the proposed joint model with introduced collaborative constraints achieves a significant improvement in accuracy).

[0041] Table 1 compares the prediction accuracy of the newly proposed constraint validation model with its baseline architecture on the core dataset.

[0042] Various modifications and variations of this invention will be apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An optical coherence tomography-based ophthalmic disease diagnosis system, characterized by, include: The image data acquisition module is configured to acquire OCT image data for the diagnosis of ophthalmic diseases and to preprocess the OCT image data. The feature extraction and similarity calculation module is configured to: extract OCT image feature representation based on OCT image data, extract the layered structure features of the macular region in OCT image data, calculate the macular structure similarity information between different OCT image data, and construct a structural similarity alignment matrix that reflects the total number of OCT images; Simultaneously, based on the OCT image data, disease category label information is obtained, and an association matrix representing the correspondence between OCT images and disease categories is constructed; The model training module is configured to: construct an objective function based on structural constraint learning, and introduce the structural similarity alignment matrix and the correlation matrix as joint supervision signals into the objective function to adaptively learn and obtain an ophthalmic disease diagnosis model; The diagnostic prediction module is configured to: acquire OCT image data of the patient to be diagnosed as input to the ophthalmic disease diagnostic model, perform forward propagation processing, and output a classification diagnostic result indicating the probability of the input OCT image data belonging to various ophthalmic diseases.

2. The optical coherence tomography-based ophthalmic disease diagnosis system of claim 1, wherein, The preprocessing of OCT image data includes at least one of image normalization, noise suppression, and retinal region extraction.

3. The optical coherence tomography-based ophthalmic disease diagnosis system of claim 1, wherein, By combining the manual annotation results or the prior attributes of the established database corresponding to the OCT image data, disease category label information is obtained.

4. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 1, characterized in that, The macular structure similarity information between different OCT image data is calculated using the following method: based on the Mahalanobis distance, a feature covariance matrix is ​​introduced to normalize and correct the feature representation space, and a Gaussian kernel function is used to shrink the mapping relationship of the feature representation space to calculate the macular structure similarity information between different OCT image data.

5. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 4, characterized in that, The calculation of the macular structure similarity information is specifically represented as follows: ； in, For the first Page and the first Results of macular structure similarity measurement between OCT image data. and Based on the first Page and the first OCT image feature representation extracted from OCT image data. Let be the characteristic covariance matrix.

6. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 1, characterized in that, The specific expression for the objective function based on structural constraint learning is as follows: ； in, The objective function for learning based on structural constraints is... For the classification supervision loss function, The structural constraint feature loss function, and These are the weight hyperparameters used in adaptive optimization.

7. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 6, characterized in that, The classification-supervised loss function is used to dominate the overall accuracy of disease diagnosis prediction, and it is expressed as follows: ;in, Indicates the true label, This represents the model's predicted output.

8. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 6, characterized in that, The structural constraint feature loss function is used to enable the model to learn and maintain the consistency of the underlying anatomical structure, and it is expressed as follows: ； in, The elements in the structural similarity alignment matrix represent the macular structural similarity information. and They represent the first The and the first Feature vectors output from OCT image data through a deep convolutional neural network model This represents the squared Euclidean distance between two eigenvectors.

9. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 1, characterized in that, The objective function is optimized and updated using the gradient descent algorithm until convergence is obtained, resulting in a trained ophthalmic disease diagnostic model. The optimization and update process is represented as follows: ； in, and These are all network model parameters. For learning rate, The objective function is the learning objective function based on structural constraints.

10. The ophthalmic disease diagnostic system based on optical coherence tomography as described in claim 1, characterized in that, The ophthalmic disease diagnostic model takes the OCT image data of the patient to be diagnosed as input, and outputs a classification diagnostic result representing the probability of the input OCT image data belonging to various ophthalmic diseases. Its mapping expression is as follows: ； in, and Both are mapping functions. For the OCT image data of the patient to be diagnosed, This is used to represent the classification and diagnostic results of the probability of the input OCT image data belonging to various ophthalmic diseases.

Images