Cornea conus detection system based on BS architecture

Abstract

The invention relates to the field of medical data processing, and particularly discloses a BS architecture-based keratoconus detection system, which comprises a database module, a data read-write module, a data processing module, a data identification module and an account management module, the data processing module is used for mapping a cornea image exported from the three-dimensional anterior segment analyzer into a two-dimensional target depth map of a plurality of channels, and cutting, normalizing and splicing the depth map to meet the input requirement of an image recognition algorithm; the data identification module contains an image identification algorithm, the cornea image data in the database are trained and modeled through the algorithm, and features are extracted and fused through an attention mechanism, so that cornea conus diagnosis is performed on the cornea image derived from the three-dimensional anterior segment analyzer. The method can assist in medical diagnosis of the keratoconus and perform centralized processing, management and storage on cornea image data and identification results.

Description

Technical Field

[0001] The present invention relates to the field of medical data processing, and particularly to a keratoconus detection system based on the BS architecture. Background Art

[0002] Keratoconus is an eye disease in which the cornea bulges outwards and gradually thins, presenting a conical shape; this condition may lead to severe myopia and astigmatism, and when the condition worsens, it may cause acute corneal edema and eventually form scars, resulting in a sharp decline in vision; this disease usually appears in adolescence and gradually worsens over time; although obvious keratoconus can be easily identified, in the early stages of the disease, its symptoms are not obvious, so early diagnosis may be somewhat difficult.

[0003] Currently in the medical field, doctors must invest a great deal of time and effort in analyzing corneal topographies to measure the shape of a patient's cornea, and then give guidance on whether further tests are needed or treatment should be sought; this process requires doctors to have extensive professional knowledge and excellent clinical experience in order to make an accurate evaluation, and its efficiency is relatively low.

[0004] Therefore, in order to improve the detection efficiency, a keratoconus detection system is needed. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a keratoconus detection system based on the BS architecture, which is used to assist in the medical diagnosis of keratoconus and centrally process, manage, and store corneal image data and recognition results.

[0006] To achieve the above purpose, the present invention particularly provides a keratoconus detection system based on the BS architecture. This system is deployed in a local server on the internal network, and various terminals communicating with the local server connect to the system through a browser web page for data interaction; this system includes a database module, a data reading and writing module, a data processing module, a data recognition module, and an account management module;

[0007] The database module is used to store system data, and the system data includes corneal image data, patient information data, determination result data, and trained detection model data;

[0008] The data reading and writing module is used to read the system data in the database module or write the system data into the database module;

[0009] The data processing module is used to map the corneal images exported from a three-dimensional anterior segment analyzer into two-dimensional target depth maps of several channels, and perform cropping, normalization, and stitching operations on the depth maps to meet the input requirements of the image recognition algorithm;

[0010] The data recognition module contains an image recognition algorithm, which trains and models the corneal image data in the database through this algorithm, and extracts and fuses features through an attention mechanism, so as to diagnose keratoconus for the corneal images exported from a three-dimensional anterior segment analyzer;

[0011] The account management module is used to determine the verification information to determine the access rights of the account.

[0012] As a further improvement to the technical solution of the present invention, the steps of training and modeling the corneal image data in the database include:

[0013] S1. Divide the data processed by the data processing module into three parts: a training set, a validation set, and a test set according to a set ratio;

[0014] S2. Extract features and perform cross-channel feature fusion on the corneal image data in the sample set;

[0015] S3. Identify the fused features, and perform backpropagation update on the model parameters through the recognition results and loss function to obtain the final target image recognition algorithm model.

[0016] As a further improvement to the technical solution of the present invention, in step S1, the sample data set is divided according to a ratio of 6:2:2. The training set and the validation set are divided in a 4-fold cross manner, and the average error of multiple modelings is obtained during model validation to determine appropriate model hyperparameters; after the hyperparameters are determined, the training set and the validation set are merged to form the final training set, and the final model is trained. Finally, the generalization ability of the model is detected through the test set.

[0017] As a further improvement to the technical solution of the present invention, in step S2, the feature extraction is performed by using multiple convolutions and skip connections. While ensuring the backpropagation of gradients and accelerating the model training speed, the corneal features are extracted to obtain a higher-dimensional representation of the features.

[0018] As a further improvement to the technical solution of the present invention, in step S2, the feature fusion performs cross-channel feature fusion on the sample features based on the self-attention mechanism. When realizing feature fusion, the position information of the features, including relative position and absolute position, will also be fused to connect the cross-channel feature information with the position information and obtain the correlation relationship between features and between features and positions.

[0019] As a further improvement to the technical solution of the present invention, in step S3, the fused features are input into the softmax model and two probability values are output. The category corresponding to the maximum output probability is the sample category recognized by the model. According to the error between the sample category recognized by the model and the sample label, the cross-entropy loss function is used to perform backpropagation update on the model parameters. After iterating several times until convergence, the model parameters are fixed to obtain the final target image recognition algorithm model;

[0020] The mathematical expression of the cross-entropy loss function L is:

[0021] Probability of the class.

[0023] As a further improvement to the technical solution of the present invention, the corneal image data includes the anterior corneal surface height, the posterior corneal surface height, the anterior corneal surface curvature, the posterior corneal surface curvature, the diopter of the entire cornea, the corneal thickness, and the anterior chamber depth.

[0024] As a further improvement to the technical solution of the present invention, in the account management module, the access rights of the account are divided into ordinary accounts, administrator accounts, and expert accounts;

[0025] Ordinary accounts can upload corneal image data to the database module, and at the same time use the data recognition module to perform keratoconus determination on specific corneal data and initiate a corneal data review request for corneal images that are difficult to determine;

[0026] Administrator accounts are responsible for maintaining the system and receiving corneal data review requests from ordinary accounts;

[0027] Expert accounts are used to receive review invitations from administrators and make manual judgments on corneal data and recognition results.

[0028] Compared with the prior art, the present invention has the following beneficial technical effects:

[0029] A keratoconus detection system based on the BS architecture provided by the present invention, when in use, the user accesses the system through the IP address on the web page, provides verification information such as account and password to compare with the information stored on the server side, and after determining the account access rights, can batch upload information such as patient data, corneal images, and annotation data exported from the three-dimensional anterior segment analyzer to the database module of the system. By calling the data processing module, the data recognition module and combining with expert manual judgment, a correct and reliable keratoconus diagnosis result can be obtained. It can be seen from this that the present invention can provide reliable guidance for the diagnosis of keratoconus, and at the same time uniformly collect and manage the data information of the entire diagnosis process, which is convenient for realizing fast data retrieval and download, and provides more comprehensive and accurate data support for data analysis, mining, and exploration research.

[0030] Advantages of additional aspects of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute a limitation to the present invention.

[0032] Figure 1 is the application flow chart of the present invention;

[0033] Figure 2 is the flow chart of the training and modeling stage of the present invention;

[0034] Figure 3 is the state diagram when data is uploaded in the present invention;

[0035] Figure 4 is the state diagram when the detection result is displayed in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Embodiment

[0038] This embodiment provides a keratoconus detection system based on the BS architecture, which is applicable to the recognition, processing and management of corneal image data; the system is deployed in a local server on the internal network, and various terminals communicating with the local server are connected to the system through a browser web page for data interaction.

[0039] The BS architecture is a computer architecture developed based on Internet technology, which is an optimization and innovation of the traditional CS architecture. In the BS architecture, the user's operation interface is implemented through a web browser, but some basic data processing tasks are undertaken by the front end (i.e., the browser). However, most important business logics are executed by the back-end server, which forms what we call a three-tier architecture. The BS architecture is a network architecture model that emerged with the rise of Web technology, and the Web browser is its most important client application. The advantage of this model is that it integrates the core parts of the client and the back-end system, making the development, maintenance, and use of the system more convenient. Just install a browser (such as Netscape Navigator or Internet Explorer) on the client device and install the corresponding database service (such as Oracle, Sybase, Informix, or SQL Server) on the server side, and data-database interaction can be achieved through the Web server, thus greatly reducing the total cost of ownership (TCO) for users during the usage process.

[0040] The system includes a database module, a data reading and writing module, a data processing module, a data recognition module, and an account management module.

[0041] The database module is used to store system data, and the system data includes corneal image data, patient information data, determination result data, and trained detection model data.

[0042] The data reading and writing module is used to read the system data in the database module or write the system data into the database module.

[0043] The data processing module is used to map the corneal images exported from the three-dimensional anterior segment analyzer into two-dimensional target depth maps of several channels, and perform cropping, normalization, and stitching operations on the depth maps to meet the input requirements of the image recognition algorithm.

[0044] The data recognition module contains an image recognition algorithm. Through this algorithm, training and modeling are performed on the corneal image data in the database, and features are extracted and fused through the attention mechanism, so as to diagnose keratoconus for the corneal images exported from the three-dimensional anterior segment analyzer.

[0045] The account management module is used to determine the verification information to determine the access rights of the account.

[0046] The corneal image data includes the anterior corneal surface height, posterior corneal surface height, anterior corneal surface curvature, posterior corneal surface curvature, refractive power of the entire cornea, corneal thickness, and anterior chamber depth.

[0047] In the account management module, the access rights of accounts are divided into ordinary accounts, administrator accounts, and expert accounts;

[0048] Ordinary accounts can upload corneal image data to the database module, and at the same time use the data recognition module to perform keratoconus determination on specific corneal data, and initiate a corneal data review request for corneal images that are difficult to determine; the administrator account is responsible for system maintenance and receiving corneal data review requests from ordinary accounts; the expert account is used to receive review invitations from the administrator and make manual judgments on corneal data and recognition results.

[0049] As Figure 1 shown, after the system is converted into visualization software, its main application process is as follows:

[0050] First, the user accesses the management platform system through the IP address on the web page and provides verification information such as the account and password. The server side will determine the verification information to determine whether the access right of the account is an ordinary account, a management account, or an expert account.

[0051] Second, ordinary accounts batch upload information such as patient data, corneal images, and annotation data exported from the three-dimensional anterior segment analyzer to the database module of the system; the corneal image data includes seven types of two-dimensional depth maps of corneal anterior surface height, corneal posterior surface height, corneal anterior surface curvature, corneal posterior surface curvature, refractive power of the entire cornea, corneal thickness, and anterior chamber depth; click Figure 3 the square area, and select the file to be uploaded in the pop-up selection box.

[0052] After the file upload is completed, the server side will perform the following operations: the system will automatically execute the data processing module to process the uploaded seven types of two-dimensional target depth corneal image data in excel format; first, perform visual area cropping to discard data irrelevant to corneal diagnosis; then normalize the values of the two-dimensional depth maps to make the units and scales of different features unified; finally, splice the seven typical features of the cornea in the depth direction to synthesize an image with seven channels and store it in the database in csv format.

[0053] Third, click the detection button to obtain the detection result below, as Figure 4 shown. When determining keratoconus, ordinary accounts input specific corneal image data into the data recognition module to obtain the algorithm determination result; if there are doubts about the determination result, an application can be made for manual review and determination of this sample. After receiving the review request, the administrator account will allocate the sample to the corresponding expert account. After the expert makes a manual determination through the expert account, the determination result will be sent to the ordinary account.

[0054] Fourth, the general account receives the judgment results returned by the data recognition module or the expert account, exports the patient information, corneal image data, judgment results, etc. of this diagnosis, and stores the information in the database at the same time for other users to retrieve and view.

[0055] In this embodiment, the steps of training and modeling the corneal image data in the database include:

[0056] S1. Divide the data processed by the data processing module into three parts: a training set, a validation set, and a test set according to a set ratio;

[0057] S2. Extract features and perform cross-channel feature fusion on the corneal image data in the sample set;

[0058] S3. Identify the fused features, and perform backpropagation update on the model parameters through the recognition results and the loss function to obtain the final target image recognition algorithm model.

[0059] In step S1, the sample data set is divided according to a ratio of 6:2:2. The training set and the validation set are divided in a 4-fold cross-validation manner. The average error of multiple modelings is obtained during model validation to determine appropriate model hyperparameters; after the hyperparameters are determined, the training set and the validation set are combined to form the final training set, and the final model is trained. Finally, the generalization ability of the model is detected through the test set.

[0060] In step S2, the feature extraction is performed by using multiple convolutions and skip connections. While ensuring the backpropagation of gradients and accelerating the model training speed, the corneal features are extracted to obtain a higher-dimensional representation of the features.

[0061] In step S2, the feature fusion performs cross-channel feature fusion on the sample features based on the self-attention mechanism. When performing feature fusion, the position information of the features, including relative position and absolute position, is also fused to associate the cross-channel feature information with the position information and obtain the correlation relationship between features and between features and positions.

[0062] In step S3, the fused features are input into the softmax model and two probability values are output. The category corresponding to the maximum output probability is the sample category recognized by the model; according to the error between the sample category recognized by the model and the sample label, the cross-entropy loss function is used to perform backpropagation update on the model parameters. After iterating several times to reach convergence, the model parameters are fixed to obtain the final target image recognition algorithm model;

[0063] The mathematical expression of the cross-entropy loss function L is:

[0064]

[0065] Among them, y i represents the label of sample i, where the positive class is 1 and the negative class is 0; p i represents the probability that sample i is predicted as the positive class.

[0066] Finally, it should be noted that specific examples are used in this article to elaborate on the principles and implementation methods of the present invention. The descriptions of the above embodiments are only used to help understand the core idea of the present invention. Without departing from the principles of the present invention, several improvements and modifications can still be made to the present invention, and these improvements and modifications also fall within the protection scope of the present invention.

Claims

1. A keratoconus detection system based on the BS architecture, which is deployed in a local server on the intranet, and various terminals communicating with the local server connect to the system through a browser web page for data interaction; characterized in that: The system includes a database module, a data reading and writing module, a data processing module, a data recognition module, and an account management module; The database module is used to store system data, and the system data includes corneal image data, patient information data, determination result data, and trained detection model data; The data reading and writing module is used to read the system data in the database module or write the system data into the database module; The data processing module is used to map the corneal images exported from the three-dimensional anterior segment analyzer into two-dimensional target depth maps of several channels, and perform cropping, normalization, and stitching operations on the depth maps to meet the input requirements of the image recognition algorithm; The data recognition module contains an image recognition algorithm, which trains and models the corneal image data in the database through this algorithm, and extracts and fuses features through an attention mechanism, so as to diagnose keratoconus for the corneal images exported from the three-dimensional anterior segment analyzer; The account management module is used to determine the verification information to determine the access rights of the account.

2. The keratoconus detection system based on the BS architecture according to claim 1, characterized in that: The steps of training and modeling the corneal image data in the database include: S1. Divide the data processed by the data processing module into three parts: a training set, a validation set, and a test set according to a set ratio; S2. Extract features and perform cross-channel feature fusion on the corneal image data in the sample set; S3. Identify the fused features, and perform backpropagation update on the model parameters through the recognition results and loss function to obtain the final target image recognition algorithm model.

3. The keratoconus detection system based on the BS architecture according to claim 2, characterized in that: In step S1, the sample data set is divided according to a ratio of 6:2:

2. The training set and the validation set are divided in a 4-fold cross manner, and the average error of multiple modelings is obtained during model validation to determine appropriate model hyperparameters; after the hyperparameters are determined, the training set and the validation set are merged to form the final training set, and the final model is trained. Finally, the generalization ability of the model is detected through the test set.

4. The keratoconus detection system based on the BS architecture according to claim 3, characterized in that: In step S2, the feature extraction is performed by using multiple convolutions and skip connections. While ensuring the backpropagation of gradients and accelerating the model training speed, the corneal features are extracted to obtain a higher-dimensional representation of the features.

5. The keratoconus detection system based on the BS architecture according to claim 4, characterized in that: In step S2, the feature fusion performs cross-channel feature fusion on the sample features based on the self-attention mechanism. When realizing the feature fusion, the position information of the features, including relative position and absolute position, will also be fused to connect the cross-channel feature information with the position information and obtain the correlation relationship between features and between features and positions.

6. The keratoconus detection system based on the BS architecture according to claim 5, characterized in that: In step S3, after inputting the fused features into the softmax model, two probability values are output, and the category corresponding to the maximum output probability is the sample category recognized by the model; according to the error between the sample category recognized by the model and the sample label, the cross-entropy loss function is used to perform backpropagation updates on the model parameters. After iterating several times until convergence, the model parameters are fixed, and the final target image recognition algorithm model is obtained; The mathematical expression of the cross-entropy loss function L is: where y i represents the label of sample i, with the positive class being 1 and the negative class being 0; p i represents the probability that sample i is predicted as the positive class.

7. A keratoconus detection system based on a BS architecture according to any one of claims 1 to 6, characterized in that: The corneal image data includes the anterior corneal surface height, posterior corneal surface height, anterior corneal surface curvature, posterior corneal surface curvature, refractive power of the entire cornea, corneal thickness, and anterior chamber depth.

8. A keratoconus detection system based on a BS architecture according to any one of claims 1 to 6, characterized in that: In the account management module, the access rights of the account are divided into ordinary accounts, administrator accounts, and expert accounts; The ordinary account uploads the corneal image data to the database module, and at the same time uses the data recognition module to perform keratoconus determination on specific corneal data, and initiates a corneal data review request for corneal images that are difficult to determine; The administrator account maintains the system and receives corneal data review requests from ordinary accounts; The expert account receives the review invitation from the administrator and makes a manual judgment on the corneal data and the recognition results.

Images