An anterior segment disease identification method, system, device and storage medium

Abstract

The application discloses an anterior segment disease identification method, system, device and storage medium. The anterior segment disease identification method comprises the following steps: acquiring an image to be identified; inputting the image to be identified into a trained semi-supervised segmentation model to obtain a first image segmentation result output by the trained semi-supervised segmentation model; extracting a feature vector of the first image segmentation result; inputting the feature vector into a trained support vector machine model to obtain a disease classification result output by the trained support vector machine model; selecting a corresponding high-precision disease diagnosis model from a high-precision disease diagnosis model set based on the disease classification result, and taking the high-precision disease diagnosis model as a high-precision disease diagnosis model to be used; determining a second image segmentation result from the first image segmentation result based on the disease classification result; inputting the second image segmentation result and the image to be identified into the high-precision disease diagnosis model to be used to obtain an anterior segment disease identification result of the image to be identified, and the efficiency and accuracy of diagnosis are improved.

Description

Technical Field

[0001] This application relates to the technical field of anterior segment disease identification, and in particular to a method, system, device and storage medium for anterior segment disease identification. Background Technology

[0002] With the development of artificial intelligence, computer-aided diagnosis plays an important role in disease screening and diagnosis. Currently, two methods are commonly used to diagnose eye diseases: One method involves processing images of the human eye using optical devices to obtain diagnostic images, which are then used by clinicians based on their experience to determine if the user has an eye disease. The other method uses AR glasses to obtain the amplitude or phase spectrum of the human eye image, which is then used by a diagnostic device to automatically diagnose whether the user has an eye disease. However, the former relies heavily on subjective judgment, and different doctors may arrive at inconsistent results; moreover, the latter, while eliminating the need for manual diagnosis, cannot precisely distinguish between various eye diseases. Summary of the Invention

[0003] This application aims to at least address the technical problems existing in the prior art. To this end, this application proposes a method, system, device, and storage medium for identifying anterior segment diseases, which can improve the efficiency and accuracy of diagnosis.

[0004] A first aspect of this application provides a method for identifying anterior segment diseases, comprising the following steps: Acquire an image to be identified, wherein the image to be identified is an anterior segment optical coherence tomography image to be classified; The image to be identified is input into a trained semi-supervised segmentation model to obtain the first image segmentation result output by the trained semi-supervised segmentation model, wherein the trained semi-supervised segmentation model is a U-shaped convolutional neural network architecture. Extract the feature vector of the first image segmentation result; The feature vector is input into the trained support vector machine model to obtain the disease classification result output by the trained support vector machine model, wherein the disease classification result is glaucoma, corneal disease and / or lens disease; Based on the disease classification results, a corresponding high-precision disease diagnosis model is selected from the high-precision disease diagnosis model set as a high-precision disease diagnosis model to be used. The high-precision disease diagnosis model set includes a high-precision disease diagnosis model for glaucoma, a high-precision disease diagnosis model for corneal diseases, and a high-precision disease diagnosis model for lens diseases. All of the high-precision disease diagnosis models adopt the ResNet50 network architecture. Based on the disease classification results, a second image segmentation result is determined from the first image segmentation result; The second image segmentation result and the image to be identified are input into the high-precision disease diagnosis model to obtain the anterior segment disease identification result of the image to be identified.

[0005] The anterior segment disease identification method according to the embodiments of this application has at least the following beneficial effects: This method acquires an image to be identified, inputs it into a trained semi-supervised segmentation model to obtain a first image segmentation result output by the trained semi-supervised segmentation model, and extracts the feature vector of the first image segmentation result. The feature vector is then input into a trained support vector machine model to obtain a disease classification result output by the trained support vector machine model. Based on the disease classification result, a corresponding high-precision disease diagnosis model is selected from a set of high-precision disease diagnosis models as the high-precision disease diagnosis model to be used. Based on the disease classification result, a second image segmentation result is determined from the first image segmentation result. The second image segmentation result and the image to be identified are then input into the high-precision disease diagnosis model to be used to obtain the anterior segment disease identification result of the image to be identified. This application combines the preliminary disease classification result with targeted high-precision disease diagnosis to achieve rapid initial screening and high-precision diagnosis of various ophthalmic diseases, improving the efficiency and accuracy of diagnosis, thereby achieving efficient clinical auxiliary decision-making.

[0006] According to some embodiments of this application, the first image segmentation result includes a plurality of structural segmentation images, and the extraction of the feature vector of the first image segmentation result includes: Determine the mean gray level, maximum gray level, standard deviation gray level, and gray level entropy of each of the structured segmented images; The texture energy of each of the structured segmentation images is calculated using the gray-level co-occurrence matrix, wherein the texture energy is a constant used to characterize the uniformity of the gray-level distribution of the structured segmentation image; By combining the mean gray level, the maximum gray level, the standard deviation of gray level, the gray level entropy, and the texture energy, a multidimensional feature vector is obtained for each of the structure segmentation images; The feature vectors of the first image segmentation result are obtained by normalizing all the multidimensional feature vectors using the Min-Max normalization method.

[0007] According to some embodiments of this application, before inputting the feature vector into the trained support vector machine model, the method further includes: Construct a first training dataset, wherein the first training dataset includes historical multidimensional feature vectors of historical structure segmentation images and historical disease classification results, wherein the historical disease classification results include glaucoma, corneal diseases and / or lens diseases; An initial support vector machine model is constructed by inputting the first training dataset into the initial support vector machine model and training the initial support vector machine model using the One-vs-Rest method until a first preset maximum number of iterations is reached, thereby obtaining the trained support vector machine model.

[0008] According to some embodiments of this application, before inputting the image to be recognized into the trained semi-supervised segmentation model, the method further includes: Construct a second training dataset, wherein the second training dataset includes labeled historical anterior segment optical coherence tomography (OCT) images and unlabeled historical anterior segment OCT images, wherein the label value of the labeled historical anterior segment OCT images is the segmentation result of the historical anterior segment OCT images, and the segmentation result includes a segmentation category, wherein the segmentation category includes anterior chamber and iris; Construct an initial semi-supervised deep neural network model based on Mean-Teacher, wherein the initial semi-supervised deep neural network model based on Mean-Teacher includes a student model and a teacher model, and the student model and the teacher model adopt the same U-shaped convolutional neural network architecture; Input the second training dataset into the initial semi-supervised deep neural network model to obtain the first total loss value output by the initial semi-supervised deep neural network model; Based on the first total loss value, the initial semi-supervised deep neural network model is iteratively updated until the second preset maximum number of iterations is reached, thereby obtaining the trained semi-supervised segmentation model.

[0009] According to some embodiments of this application, the step of inputting the second training dataset into the initial semi-supervised deep neural network model to obtain the first total loss value output by the initial semi-supervised deep neural network model includes: The labeled historical anterior segment optical coherence tomography images are input into the teacher model to calculate the first supervised loss value of the labeled historical anterior segment optical coherence tomography images in the initial semi-supervised deep neural network model using the mean square error loss function. The unlabeled historical anterior segment optical coherence tomography image is input into the initial semi-supervised deep neural network model to calculate the first consistency loss value of the unlabeled historical anterior segment optical coherence tomography image in the initial semi-supervised deep neural network model through the mean square error loss function. The unlabeled historical anterior segment optical coherence tomography image is input into the initial semi-supervised deep neural network model to determine the first uncertainty loss value of the unlabeled historical anterior segment optical coherence tomography image using the Monte Carlo method; The first total loss value is calculated based on the first supervision loss value, the first consistency loss value, and the first uncertainty loss value.

[0010] According to some embodiments of this application, the first total loss value is calculated using the following formula:

[0011] in, This is the first total loss value. The first supervised loss value, This is the first consistency loss value. This is the first uncertain loss value. To preset the balance coefficient value, This is the first weighting coefficient value. This is the value of the second weighting coefficient.

[0012] According to some embodiments of this application, the method further includes: Based on the image to be identified and the segmentation result of the first image, the two-dimensional parameter measurement results are determined, wherein the two-dimensional parameter measurement results include the anterior chamber angle opening distance, the anterior chamber angle crypt area, and the trabecular meshwork-iris space area; Based on the image to be identified, a three-dimensional point cloud model of the eye structure is constructed, and the volume parameter measurement results in the three-dimensional point cloud model of the eye structure are determined. The volume parameter measurement results include the anterior chamber volume, iris volume, and trabecular meshwork-iris space volume. Based on the disease classification results, the anterior segment disease identification results, the two-dimensional parameter measurement results, the three-dimensional point cloud model of the eye structure, and the volume parameter measurement results, an auxiliary diagnostic report for anterior segment diseases is generated. Send the anterior segment disease auxiliary diagnostic report to the user.

[0013] A second aspect of this application provides an anterior segment disease identification system, the anterior segment disease identification system comprising: The data acquisition module is used to acquire the image to be identified, wherein the image to be identified is an anterior segment optical coherence tomography image to be classified; The first image segmentation module is used to input the image to be identified into a trained semi-supervised segmentation model to obtain the first image segmentation result output by the trained semi-supervised segmentation model, wherein the trained semi-supervised segmentation model is a U-shaped convolutional neural network architecture. The feature extraction module is used to extract the feature vector of the first image segmentation result; The disease classification module is used to input the feature vector into the trained support vector machine model to obtain the disease classification result output by the trained support vector machine model, wherein the disease classification result is glaucoma, corneal disease and / or lens disease; The model selection module is used to select a corresponding high-precision disease diagnosis model from the high-precision disease diagnosis model set based on the disease classification results, as a high-precision disease diagnosis model to be used. The high-precision disease diagnosis model set includes high-precision disease diagnosis models for glaucoma, corneal diseases, and lens diseases. All the high-precision disease diagnosis models adopt the ResNet50 network architecture. The second image segmentation module is used to determine the second image segmentation result from the first image segmentation result based on the disease classification result; The disease identification module is used to input the second image segmentation result and the image to be identified into the high-precision disease diagnosis model to obtain the anterior segment disease identification result of the image to be identified.

[0014] This system acquires an image to be identified, inputs it into a trained semi-supervised segmentation model to obtain a first image segmentation result output by the trained semi-supervised segmentation model, and extracts the feature vector of the first image segmentation result. The feature vector is then input into a trained support vector machine model to obtain a disease classification result output by the trained support vector machine model. Based on the disease classification result, a corresponding high-precision disease diagnosis model is selected from a set of high-precision disease diagnosis models as the high-precision disease diagnosis model to be used. Based on the disease classification result, a second image segmentation result is determined from the first image segmentation result. The second image segmentation result and the image to be identified are then input into the high-precision disease diagnosis model to be used to obtain the anterior segment disease identification result of the image to be identified. This application, by combining preliminary disease classification results with targeted high-precision disease diagnosis, achieves rapid initial screening and high-precision diagnosis of various ophthalmic diseases, improving the efficiency and accuracy of diagnosis, thereby realizing efficient clinical auxiliary decision-making.

[0015] A third aspect of this application provides an anterior segment disease identification electronic device, including at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enables the at least one control processor to perform the aforementioned anterior segment disease identification method.

[0016] A fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the aforementioned anterior segment disease identification method.

[0017] It should be noted that the beneficial effects of the second to fourth aspects of this application compared with the prior art are the same as the beneficial effects of the aforementioned anterior segment disease recognition system compared with the prior art, and will not be described in detail here.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart illustrating an embodiment of the anterior segment disease identification method provided in this application; Figure 2 These are schematic diagrams of three-dimensional reconstruction of ophthalmic images in different shooting modes, representing embodiments of the anterior segment disease identification method provided in this application. Figure 3 This is a schematic diagram of the structure of an embodiment of the anterior segment disease recognition system provided in this application; Figure 4 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application. Detailed Implementation

[0020] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0021] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0022] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0023] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0024] With the development of artificial intelligence, computer-aided diagnosis plays an important role in disease screening and diagnosis. Currently, two methods are commonly used to diagnose eye diseases: One method involves processing images of the human eye using optical devices to obtain diagnostic images, which are then used by clinicians based on their experience to determine if the user has an eye disease. The other method uses AR glasses to obtain the amplitude or phase spectrum of the human eye image, which is then used by a diagnostic device to automatically diagnose whether the user has an eye disease. However, the former relies heavily on subjective judgment, and different doctors may arrive at inconsistent results; moreover, the latter, while eliminating the need for manual diagnosis, cannot precisely distinguish between various eye diseases.

[0025] To address the aforementioned technical deficiencies, embodiments of this application provide a method, system, device, and storage medium for identifying anterior segment diseases.

[0026] Please see Figure 1 This is a flowchart illustrating a method for identifying anterior segment diseases provided in an embodiment of this application. The method is applied to an electronic device, such as a server. Figure 1 As shown, this anterior segment disease identification method includes: Step S101: Obtain the image to be identified, wherein the image to be identified is an anterior segment optical coherence tomography image to be classified; In step S101, the above-mentioned acquisition of the image to be identified can be achieved by using anterior segment optical coherence tomography to acquire the initial image to be identified using a planar scan method or a ring scan method, and then using automatic pixel pitch conversion and resolution interpolation adjustment to calibrate the image spatial scale to obtain the calibrated image to be identified. The calibrated image to be identified is then stored according to a pre-set format standard based on actual needs to obtain the image to be identified.

[0027] This application provides a consistent data foundation for subsequent anterior segment structure segmentation and eye disease diagnosis by acquiring anterior segment optical coherence tomography images with uniform format and stable quality, thus ensuring the accuracy and stability of subsequent analysis.

[0028] Step S102: Input the image to be recognized into the trained semi-supervised segmentation model to obtain the first image segmentation result output by the trained semi-supervised segmentation model, wherein the trained semi-supervised segmentation model is a U-shaped convolutional neural network architecture. Step S103: Extract the feature vector of the first image segmentation result; Step S104: Input the feature vector into the trained support vector machine model to obtain the disease classification result output by the trained support vector machine model, wherein the disease classification result is glaucoma, corneal disease and / or lens disease; Step S105: Based on the disease classification results, select the corresponding high-precision disease diagnosis model from the high-precision disease diagnosis model set as the high-precision disease diagnosis model to be used. The high-precision disease diagnosis model set includes high-precision disease diagnosis models for glaucoma, corneal diseases, and lens diseases. All high-precision disease diagnosis models adopt the ResNet50 network architecture. Step S106: Based on the disease classification results, determine the second image segmentation result from the first image segmentation result; In step S106, the above-mentioned determination of the second image segmentation result from the first image segmentation result based on the disease classification result can be the image segmentation result corresponding to the disease classification result selected from the first image segmentation result as the second image segmentation result.

[0029] Step S107: Input the second image segmentation result and the image to be identified into the high-precision disease diagnosis model to obtain the anterior segment disease identification result of the image to be identified.

[0030] The above-mentioned anterior segment disease identification results can be as follows: if the disease classification result is glaucoma, the anterior segment disease identification result includes the determination of whether the glaucoma is open or closed; if the disease classification result is corneal disease, the anterior segment disease identification result includes the determination of the degree of opacity and keratoconus; if the disease classification result is lens disease, the anterior segment disease identification result includes the determination of lens age, thickness, cataract area and severity level.

[0031] This method acquires an image to be identified, inputs it into a trained semi-supervised segmentation model to obtain a first image segmentation result output by the trained semi-supervised segmentation model, and extracts the feature vector of the first image segmentation result. The feature vector is then input into a trained support vector machine model to obtain a disease classification result output by the trained support vector machine model. Based on the disease classification result, a corresponding high-precision disease diagnosis model is selected from a set of high-precision disease diagnosis models as the high-precision disease diagnosis model to be used. Based on the disease classification result, a second image segmentation result is determined from the first image segmentation result. The second image segmentation result and the image to be identified are then input into the high-precision disease diagnosis model to be used to obtain the anterior segment disease identification result of the image to be identified. This application combines the preliminary disease classification result with targeted high-precision disease diagnosis to achieve rapid initial screening and high-precision diagnosis of various ophthalmic diseases, improving the efficiency and accuracy of diagnosis, thereby achieving efficient clinical auxiliary decision-making.

[0032] In some embodiments, the first image segmentation result includes several structural segmentation images, and the feature vector of the first image segmentation result is extracted, including: Step S201: Determine the mean gray level, maximum gray level, standard deviation gray level, and gray level entropy of each structure segmentation image; Step S202: Calculate the texture energy of each structure segmentation image using the gray-level co-occurrence matrix, where the texture energy is a constant used to characterize the uniformity of the gray-level distribution of the structure segmentation image; Step S203: Combine the mean gray level, maximum gray level, standard deviation gray level, gray level entropy, and texture energy to obtain a multidimensional feature vector for each structure segmentation image; Specifically, the segmentation result of the first image (which can be a corneal image, anterior chamber image, or lens image) is used as the analysis object. Multi-dimensional morphological and optical features, such as the mean gray level, maximum gray level, standard deviation gray level, gray level entropy, and texture energy, are extracted from the first image segmentation result, as follows: Gray mean ( It can reflect the overall gray level within a structural region and is a key indicator for judging tissue density or transparency. The calculation formula is:

[0033] Where N is the total number of pixels in the target structure (corneal image, anterior chamber image, or lens image). For pixels The grayscale value (grayscale range 0 to 255).

[0034] Maximum grayscale value ( This can reflect the highest gray value within a structural region, used to identify localized high-reflectivity anomalies (such as high-gray-value points in corneal scars and foci of lens opacity). The calculation formula is:

[0035] Texture energy ( The texture energy of each structure segmentation image is calculated using the gray-level co-occurrence matrix and the following formula:

[0036] in, Let be the co-occurrence probability of gray level i and gray level j in the gray-level co-occurrence matrix.

[0037] Extract the grayscale standard deviation and grayscale entropy of each structural segmentation image.

[0038] Step S204: Normalize all multidimensional feature vectors using the Min-Max normalization method to obtain the feature vector of the first image segmentation result.

[0039] This application provides a quantitative basis for subsequent identification of anterior segment diseases by automatically extracting and quantifying clinical parameters of ocular structures, thereby improving the efficiency and accuracy of diagnosis.

[0040] In some embodiments, before inputting the feature vectors into the trained support vector machine model, the method further includes: Step S301: Construct the first training dataset, wherein the first training dataset includes historical multidimensional feature vectors of historical structure segmentation images and historical disease classification results, wherein the historical disease classification results include glaucoma, corneal diseases and / or lens diseases; Step S302: Construct an initial support vector machine model. Input the first training dataset into the initial support vector machine model to train the initial support vector machine model using the One-vs-Rest method until the first preset maximum number of iterations is reached, and obtain the trained support vector machine model.

[0041] The first preset maximum number of iterations can be a value preset according to actual needs.

[0042] Specifically, the above support vector machine model uses radial basis function (RBF) as kernel function, expands to multi-class classification through one-vs-Rest (one-to-the-others) method, and trains the initial support vector machine model by optimizing the penalty coefficient and kernel parameters through 5-fold cross-validation until the first preset maximum number of iterations is reached, thus obtaining the trained support vector machine model.

[0043] This application outputs disease classification results through a trained support vector machine model, achieving a high degree of coupling with the previous segmentation results, ensuring the accuracy and stability of feature extraction, and providing an efficient and reliable preliminary basis for subsequent eye disease identification and diagnostic report generation.

[0044] In some embodiments, before inputting the image to be recognized into the trained semi-supervised segmentation model, the method further includes: Step S401: Construct a second training dataset, wherein the second training dataset includes labeled historical anterior segment optical coherence tomography images and unlabeled historical anterior segment optical coherence tomography images. The label value of the labeled historical anterior segment optical coherence tomography images is the segmentation result of the historical anterior segment optical coherence tomography images. The segmentation result includes the segmentation category, which includes the anterior chamber and the iris. Step S402: Construct an initial semi-supervised deep neural network model based on Mean-Teacher, wherein the initial semi-supervised deep neural network model based on Mean-Teacher includes a student model and a teacher model, and the student model and the teacher model adopt the same U-shaped convolutional neural network architecture. Step S403: Input the second training dataset into the initial semi-supervised deep neural network model to obtain the first total loss value output by the initial semi-supervised deep neural network model; Step S404: Based on the first total loss value, iteratively update the initial semi-supervised deep neural network model until the second preset maximum number of iterations is reached, and obtain the trained semi-supervised segmentation model.

[0045] The second preset maximum number of iterations can be a value preset according to actual needs.

[0046] Specifically, this application employs a semi-supervised deep neural network model based on a Mean-Teacher framework to efficiently and accurately segment AS-OCT images (anterior segment optical coherence tomography images). This semi-supervised framework, through the collaborative training of the student and teacher models, achieves efficient utilization of both limited labeled and extensive unlabeled data, addressing the core issues of high labeling costs and data scarcity in medical images. Both the student and teacher models utilize the same U-Shape Network (U-Net) architecture, with parameters independent but iteratively synchronized: the student model receives input images with random noise or data augmentation for prediction, optimizing its parameters by minimizing the supervision loss of labeled data and the consistency loss of unlabeled data; the teacher model dynamically updates its parameters using an Exponential Moving Average (EMA) mechanism, with the update formula as follows:

[0047] in, For teacher model weights, For student model weights, λ represents the number of training rounds, and λ is the EMA decay parameter used to maintain parameter stability in order to generate high-quality pseudo-labels.

[0048] The training of the semi-supervised deep neural network model based on Mean-Teacher (Teacher-Student) uses the Adam optimizer, and the training epochs can be up to 100. The teacher model that performs best on the test set during the training epochs can be selected as the final model.

[0049] In some embodiments, inputting the second training dataset into the initial semi-supervised deep neural network model to obtain the first total loss value output by the initial semi-supervised deep neural network model includes: Step S501: Input the labeled historical anterior segment optical coherence tomography images into the teacher model, and calculate the first supervised loss value of the labeled historical anterior segment optical coherence tomography images in the initial semi-supervised deep neural network model through the mean square error loss function. Step S502: Input the unlabeled historical anterior segment optical coherence tomography image into the initial semi-supervised deep neural network model, so as to calculate the first consistency loss value of the unlabeled historical anterior segment optical coherence tomography image in the initial semi-supervised deep neural network model through the mean square error loss function. Step S503: Input the unlabeled historical anterior segment optical coherence tomography image into the initial semi-supervised deep neural network model to determine the first uncertainty loss value of the unlabeled historical anterior segment optical coherence tomography image using the Monte Carlo method. Step S504: Calculate the first total loss value based on the first supervision loss value, the first consistency loss value, and the first uncertainty loss value.

[0050] Specifically, Monte Carlo dropout is used to approximate Bayesian uncertainty. Multiple dropout layers are added to the teacher network and T forward propagations are performed to obtain T softmax probability maps. (Where t∈[1,T], c is the segmentation category, such as anterior chamber or iris); using prediction entropy as the uncertainty measure, the uncertainty map for each category is calculated as shown in the following formula:

[0051] in, For uncertainty graphs, To predict entropy.

[0052] When the segmentation category is iris, the pixel uncertainty value is set to the maximum value. and minimum value Divide the data into k equally spaced intervals, and count the number of pixels in the first α high-uncertainty intervals. Image-level uncertainty scores were calculated using a weighted average. :

[0053] in, β is the median of the i-th interval, and β is the average number of small structure pixels for normalization, which is preset according to actual needs. When the segmentation category is iris, the first uncertainty loss value is calculated using the following formula:

[0054] Wherein, τ is a first threshold preset according to actual needs. The teacher model's predicted value for the image. The student model's predicted value for the image. This is the first uncertain loss value.

[0055] When the segmentation category is anterior chamber, a pixel-level uncertainty correction method is used to construct the Boolean matrix. (The pixel uncertainty value is lower than the second threshold preset according to actual needs) hour, ),based on The element-wise product of the difference from the prediction is calculated using the following formula to determine the first uncertainty loss value:

[0056] The formula for calculating the first consistency loss value of the unlabeled historical anterior segment optical coherence tomography images in the initial semi-supervised deep neural network model using the mean squared error loss function is as follows: j∈[1,C]; Where C represents the total number of segmentation categories. Let be the predicted value for the j-th segmentation category in the teacher model. This is the predicted value for the j-th segmentation category in the student model.

[0057] This application achieves accurate segmentation of multiple structures such as the cornea, iris, and lens by introducing an uncertainty-driven label adaptation and selection strategy and designing a small-scale consistency loss constraint, thereby enhancing the segmentation performance of small-area structures.

[0058] In some embodiments, the first total loss value is calculated using the following formula:

[0059] in, This is the first total loss value. The first supervised loss value, This is the first consistency loss value. This is the first uncertain loss value. This is a preset balance coefficient value set according to actual needs. The first weighting coefficient value is preset according to actual needs. This is a second weighting coefficient value pre-set according to actual needs.

[0060] This application enables weighted collaborative optimization of multi-target loss, thereby improving recognition accuracy.

[0061] In some embodiments, the method further includes: Step S601: Based on the segmentation results of the image to be identified and the first image, determine the two-dimensional parameter measurement results, wherein the two-dimensional parameter measurement results include the anterior chamber angle opening distance, the anterior chamber angle crypt area, and the trabecular meshwork-iris space area; In step S601, the two-dimensional parameter measurement result determined above based on the image to be identified and the first image segmentation result can be the two-dimensional parameter measurement result extracted from the image to be identified and the first image segmentation result.

[0062] Step S602: Based on the image to be identified, construct a three-dimensional point cloud model of the eye structure and determine the volume parameter measurement results in the three-dimensional point cloud model of the eye structure. The volume parameter measurement results include the anterior chamber volume, iris volume, and trabecular meshwork-iris space volume. In step S602, the above-mentioned three-dimensional point cloud model of the eye structure is constructed based on the image to be identified, and the volume parameter measurement results in the three-dimensional point cloud model of the eye structure are determined. The three-dimensional point cloud model of the eye structure is constructed based on the image to be identified and the scanning method of the image to be identified, and the volume parameter measurement results in the three-dimensional point cloud model of the eye structure are extracted. The scanning method is either a planar scanning method or a circular scanning method.

[0063] Step S603: Based on the disease classification results, anterior segment disease identification results, two-dimensional parameter measurement results, three-dimensional point cloud model of eye structure and volume parameter measurement results, generate an anterior segment disease auxiliary diagnosis report; In step S603, the above-mentioned generation of anterior segment disease auxiliary diagnosis report based on disease classification results, anterior segment disease identification results, two-dimensional parameter measurement results, three-dimensional point cloud model of eye structure and volume parameter measurement results can be achieved by integrating disease classification results, anterior segment disease identification results, two-dimensional parameter measurement results, three-dimensional point cloud model of eye structure and volume parameter measurement results according to a report template pre-set according to actual needs, to obtain an anterior segment disease auxiliary diagnosis report.

[0064] Step S604: Send the anterior segment disease auxiliary diagnosis report to the user.

[0065] This application generates an anterior segment disease auxiliary diagnostic report based on disease classification results, anterior segment disease identification results, two-dimensional parameter measurement results, three-dimensional point cloud model of eye structure, and volume parameter measurement results. It can comprehensively and accurately reflect the actual physiological characteristics of the eyeball to the user, improve the user experience, and increase the efficiency and accuracy of diagnosis, thereby achieving efficient clinical auxiliary decision-making.

[0066] Additionally, refer to Figure 3One embodiment of this application provides an anterior segment disease recognition system, including a data acquisition module 1100, a first image segmentation module 1200, a feature extraction module 1300, a disease classification module 1400, a model selection module 1500, a second image segmentation module 1600, and a disease recognition module 1700, wherein: The data acquisition module 1100 is used to acquire the image to be identified, wherein the image to be identified is an anterior segment optical coherence tomography image to be classified; The first image segmentation module 1200 is used to input the image to be recognized into the trained semi-supervised segmentation model and obtain the first image segmentation result output by the trained semi-supervised segmentation model. The trained semi-supervised segmentation model is a U-shaped convolutional neural network architecture. The feature extraction module 1300 is used to extract the feature vector of the first image segmentation result; The disease classification module 1400 is used to input the feature vector into the trained support vector machine model and obtain the disease classification result output by the trained support vector machine model, wherein the disease classification result is glaucoma, corneal disease and / or lens disease; The model selection module 1500 is used to select a corresponding high-precision disease diagnosis model from the high-precision disease diagnosis model set based on the disease classification results, as a high-precision disease diagnosis model to be used. The high-precision disease diagnosis model set includes high-precision disease diagnosis models for glaucoma, corneal diseases, and lens diseases. All high-precision disease diagnosis models adopt the ResNet50 network architecture. The second image segmentation module 1600 is used to determine the second image segmentation result from the first image segmentation result based on the disease classification result; The disease identification module 1700 is used to input the second image segmentation result and the image to be identified into the high-precision disease diagnosis model to obtain the anterior segment disease identification result of the image to be identified.

[0067] This system acquires an image to be identified, inputs it into a trained semi-supervised segmentation model to obtain a first image segmentation result output by the trained semi-supervised segmentation model, and extracts the feature vector of the first image segmentation result. The feature vector is then input into a trained support vector machine model to obtain a disease classification result output by the trained support vector machine model. Based on the disease classification result, a corresponding high-precision disease diagnosis model is selected from a set of high-precision disease diagnosis models as the high-precision disease diagnosis model to be used. Based on the disease classification result, a second image segmentation result is determined from the first image segmentation result. The second image segmentation result and the image to be identified are then input into the high-precision disease diagnosis model to be used to obtain the anterior segment disease identification result of the image to be identified. This application, by combining preliminary disease classification results with targeted high-precision disease diagnosis, achieves rapid initial screening and high-precision diagnosis of various ophthalmic diseases, improving the efficiency and accuracy of diagnosis, thereby realizing efficient clinical auxiliary decision-making.

[0068] It should be noted that the system embodiments described above are based on the same inventive concept as the method embodiments described above. Therefore, the relevant content of the method embodiments described above is also applicable to the system embodiments described above, and will not be repeated here.

[0069] Figure 4 A schematic diagram of the hardware structure for anterior segment disease recognition provided in an embodiment of this application is shown.

[0070] An anterior segment disease recognition device may include a processor 301 and a memory 302 storing computer program instructions.

[0071] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0072] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory.

[0073] In some embodiments, memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this disclosure.

[0074] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the anterior segment disease identification methods in the above embodiments.

[0075] In one example, the anterior segment disease recognition device may further include a communication interface 303 and a bus 310. Wherein, as Figure 4 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 310 and complete communication with each other.

[0076] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0077] Bus 310 includes hardware, software, or both, that couples components of an anterior segment disease recognition device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 310 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0078] This anterior segment disease recognition device can execute the anterior segment disease recognition method in this application embodiment based on a three-dimensional design model, thereby achieving a combination of... Figure 1 and Figure 3 The methods and systems for identifying anterior segment diseases are described.

[0079] Furthermore, in conjunction with the anterior segment disease identification method in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the anterior segment disease identification methods in the above embodiments.

[0080] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0081] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0082] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0083] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and 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, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0084] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for identifying anterior segment diseases, characterized in that, The method for identifying anterior segment diseases includes: Acquire an image to be identified, wherein the image to be identified is an anterior segment optical coherence tomography image to be classified; The image to be identified is input into a trained semi-supervised segmentation model to obtain the first image segmentation result output by the trained semi-supervised segmentation model, wherein the trained semi-supervised segmentation model is a U-shaped convolutional neural network architecture. Extract the feature vector of the first image segmentation result; The feature vector is input into the trained support vector machine model to obtain the disease classification result output by the trained support vector machine model, wherein the disease classification result is glaucoma, corneal disease and / or lens disease; Based on the disease classification results, a corresponding high-precision disease diagnosis model is selected from the high-precision disease diagnosis model set as a high-precision disease diagnosis model to be used. The high-precision disease diagnosis model set includes a high-precision disease diagnosis model for glaucoma, a high-precision disease diagnosis model for corneal diseases, and a high-precision disease diagnosis model for lens diseases. All of the high-precision disease diagnosis models adopt the ResNet50 network architecture. Based on the disease classification results, a second image segmentation result is determined from the first image segmentation result; The second image segmentation result and the image to be identified are input into the high-precision disease diagnosis model to obtain the anterior segment disease identification result of the image to be identified.

2. The method for identifying anterior segment diseases according to claim 1, characterized in that, The first image segmentation result includes several structural segmentation images, and the extraction of the feature vector of the first image segmentation result includes: Determine the mean gray level, maximum gray level, standard deviation gray level, and gray level entropy of each of the structured segmented images; The texture energy of each of the structured segmentation images is calculated using the gray-level co-occurrence matrix, wherein the texture energy is a constant used to characterize the uniformity of the gray-level distribution of the structured segmentation image; By combining the mean gray level, the maximum gray level, the standard deviation of gray level, the gray level entropy, and the texture energy, a multidimensional feature vector is obtained for each of the structure segmentation images; The feature vectors of the first image segmentation result are obtained by normalizing all the multidimensional feature vectors using the Min-Max normalization method.

3. The method for identifying anterior segment diseases according to claim 1, characterized in that, Before inputting the feature vector into the trained support vector machine model, the method further includes: Construct a first training dataset, wherein the first training dataset includes historical multidimensional feature vectors of historical structure segmentation images and historical disease classification results, wherein the historical disease classification results include glaucoma, corneal diseases and / or lens diseases; An initial support vector machine model is constructed by inputting the first training dataset into the initial support vector machine model and training the initial support vector machine model using the One-vs-Rest method until a first preset maximum number of iterations is reached, thereby obtaining the trained support vector machine model.

4. The method for identifying anterior segment diseases according to claim 3, characterized in that, Before inputting the image to be recognized into the trained semi-supervised segmentation model, the method further includes: Construct a second training dataset, wherein the second training dataset includes labeled historical anterior segment optical coherence tomography (OCT) images and unlabeled historical anterior segment OCT images, wherein the label value of the labeled historical anterior segment OCT images is the segmentation result of the historical anterior segment OCT images, and the segmentation result includes a segmentation category, wherein the segmentation category includes anterior chamber and iris; Construct an initial semi-supervised deep neural network model based on Mean-Teacher, wherein the initial semi-supervised deep neural network model based on Mean-Teacher includes a student model and a teacher model, and the student model and the teacher model adopt the same U-shaped convolutional neural network architecture; Input the second training dataset into the initial semi-supervised deep neural network model to obtain the first total loss value output by the initial semi-supervised deep neural network model; Based on the first total loss value, the initial semi-supervised deep neural network model is iteratively updated until the second preset maximum number of iterations is reached, thereby obtaining the trained semi-supervised segmentation model.

5. The method for identifying anterior segment diseases according to claim 4, characterized in that, The step of inputting the second training dataset into the initial semi-supervised deep neural network model to obtain the first total loss value output by the initial semi-supervised deep neural network model includes: The labeled historical anterior segment optical coherence tomography images are input into the teacher model to calculate the first supervised loss value of the labeled historical anterior segment optical coherence tomography images in the initial semi-supervised deep neural network model using the mean square error loss function. The unlabeled historical anterior segment optical coherence tomography image is input into the initial semi-supervised deep neural network model to calculate the first consistency loss value of the unlabeled historical anterior segment optical coherence tomography image in the initial semi-supervised deep neural network model through the mean square error loss function. The unlabeled historical anterior segment optical coherence tomography image is input into the initial semi-supervised deep neural network model to determine the first uncertainty loss value of the unlabeled historical anterior segment optical coherence tomography image using the Monte Carlo method; The first total loss value is calculated based on the first supervision loss value, the first consistency loss value, and the first uncertainty loss value.

6. The method for identifying anterior segment diseases according to claim 5, characterized in that, The first total loss value is calculated using the following formula: in, This is the first total loss value. The first supervised loss value, This is the first consistency loss value. This is the first uncertain loss value. To preset the balance coefficient value, This is the first weighting coefficient value. This is the value of the second weighting coefficient.

7. The method for identifying anterior segment diseases according to claim 1, characterized in that, The method for identifying anterior segment diseases also includes: Based on the image to be identified and the segmentation result of the first image, the two-dimensional parameter measurement results are determined, wherein the two-dimensional parameter measurement results include the anterior chamber angle opening distance, the anterior chamber angle crypt area, and the trabecular meshwork-iris space area; Based on the image to be identified, a three-dimensional point cloud model of the eye structure is constructed, and the volume parameter measurement results in the three-dimensional point cloud model of the eye structure are determined. The volume parameter measurement results include the anterior chamber volume, iris volume, and trabecular meshwork-iris space volume. Based on the disease classification results, the anterior segment disease identification results, the two-dimensional parameter measurement results, the three-dimensional point cloud model of the eye structure, and the volume parameter measurement results, an auxiliary diagnostic report for anterior segment diseases is generated. Send the anterior segment disease auxiliary diagnostic report to the user.

8. A system for identifying anterior segment diseases, characterized in that, The anterior segment disease identification system includes: The data acquisition module is used to acquire the image to be identified, wherein the image to be identified is an anterior segment optical coherence tomography image to be classified; The first image segmentation module is used to input the image to be identified into a trained semi-supervised segmentation model to obtain the first image segmentation result output by the trained semi-supervised segmentation model, wherein the trained semi-supervised segmentation model is a U-shaped convolutional neural network architecture. The feature extraction module is used to extract the feature vector of the first image segmentation result; The disease classification module is used to input the feature vector into the trained support vector machine model to obtain the disease classification result output by the trained support vector machine model, wherein the disease classification result is glaucoma, corneal disease and / or lens disease; The model selection module is used to select a corresponding high-precision disease diagnosis model from the high-precision disease diagnosis model set based on the disease classification results, as a high-precision disease diagnosis model to be used. The high-precision disease diagnosis model set includes high-precision disease diagnosis models for glaucoma, corneal diseases, and lens diseases. All the high-precision disease diagnosis models adopt the ResNet50 network architecture. The second image segmentation module is used to determine the second image segmentation result from the first image segmentation result based on the disease classification result; The disease identification module is used to input the second image segmentation result and the image to be identified into the high-precision disease diagnosis model to obtain the anterior segment disease identification result of the image to be identified.

9. A device for identifying anterior segment diseases, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform an anterior segment disease identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform a method for identifying anterior segment diseases as described in any one of claims 1 to 7.

Images