Universal medical image auxiliary diagnosis system

Abstract

The invention discloses a universal medical image auxiliary diagnosis system which comprises an ultra-low radioactivity imaging module which adopts a 3D CycleGAN network as a basic skeleton for PET image reconstruction; the epilepsy focus positioning analysis module is based on a self-adaptive medical image segmentation framework in deep learning; the user interface is used for providing a graphical operation interface and supporting multiple interaction modes so as to meet using habits of different doctors; and the data storage and management module is responsible for storage, backup and management of all medical image data and related information in the system, has data retrieval and query functions, and supports remote access and sharing. According to the universal medical image auxiliary diagnosis system, radiation exposure of a patient can be remarkably reduced, the accuracy and reliability of focus positioning can be improved, and powerful technical support is provided for diagnosis and treatment of nervous system diseases such as epilepsy.

Description

Technical Field

[0001] The present invention relates to a medical image-assisted diagnosis system, and more particularly to a universal medical image-assisted diagnosis system. Background Art

[0002] With the rapid development of medical imaging technology, imaging technology has played an important role in clinical diagnosis and treatment. However, the acquisition of medical images is often accompanied by a certain amount of radiation exposure, which poses a certain health risk to patients, especially those who need to undergo medical imaging examinations frequently. How to reduce the radiation exposure of patients while ensuring or even improving the quality of medical images has become a hot issue of common concern in the medical and engineering fields.

[0003] In traditional medical imaging technology, a high radiation dose is often required to obtain high-quality images. However, this not only increases the health risks of patients, but may also cause a series of potential complications. Therefore, the development of technology that can achieve high-quality imaging at ultra-low radioactive doses has become an important direction for the development of medical imaging technology.

[0004] At the same time, as a common neurological disease, the precise location of epilepsy lesions is crucial for the development of treatment plans. Accurate location of epilepsy lesions can not only help doctors understand the patient's condition more accurately, but also provide an important basis for the development of personalized treatment plans. However, as epilepsy lesions often have complex morphology and distribution, traditional medical imaging technology often has difficulty in achieving accurate location.

[0005] Therefore, combining the latest developments in medical imaging technology to develop a medical imaging-assisted diagnosis system that can image at ultra-low radioactive doses and accurately analyze epileptic lesions has important clinical significance and practical value. Summary of the invention

[0006] In view of the shortcomings of the prior art, the object of the present invention is to provide a medical imaging-assisted diagnosis system that can image at extremely low radioactive doses and accurately analyze epileptic lesions.

[0007] To achieve the above object, the present invention provides the following technical solution: a universal medical image-assisted diagnosis system, characterized in that it includes:

[0008] Ultra-low radioactivity imaging module, which uses the 3D CycleGAN network as the basic framework for PET image reconstruction to perform ultra-low radioactivity imaging;

[0009] Epilepsy lesion localization and analysis module: This module automatically processes and analyzes medical images after ultra-low radioactivity imaging based on the adaptive medical image segmentation framework in deep learning. It can accurately identify and locate epilepsy lesions and provide detailed information on the lesions.

[0010] User interface, which is used to provide a graphical operation interface, supports multiple interaction modes to meet the usage habits of different doctors, and supports the import and export of multiple image formats and resolutions;

[0011] Data storage and management module, which is responsible for the storage, backup and management of all medical imaging data and related information in the system. It has data retrieval and query functions and supports remote access and sharing.

[0012] As a further improvement of the present invention, the 3D CycleGAN network used in the ultra-low radioactivity imaging module consists of two mirror-symmetrical generators and two discriminators, which are respectively responsible for generating high-quality reconstructed images from PET scan data and judging whether the generated images are close to the real reference images.

[0013] As a further improvement of the present invention, the specific steps of the ultra-low radioactivity imaging module generating a high-quality reconstructed image from PET scan data are as follows: first, the input low-dose PET data is preprocessed through dual-scale image processing, including weight transfer, weight normalization and weight update processing steps. Subsequently, the dual-scale generator uses the attention mechanism to extract key features and generate a preliminary reconstructed image, i.e., a rough reconstructed PET. The preliminary reconstructed image is then subjected to a multi-scale spatial transformation to capture image features at different scales. Then, with the help of the attention mechanism, the system screens out features that have an important influence on the reconstruction results, and integrates these features into a more refined reconstruction process to generate a finely reconstructed PET.

[0014] As a further improvement of the present invention, the specific steps for the epileptic lesion localization analysis module to identify and locate epileptic lesions and provide detailed information of the lesions are as follows: first, the medical image data is preprocessed to improve the image quality, and then the adaptive medical image segmentation framework network model is trained using the training set, and the loss function and performance indicators of the adaptive medical image segmentation framework network model are monitored to ensure that it can accurately identify and locate epileptic lesions.

[0015] As a further improvement of the present invention, the data storage and management module is a MySQL database, wherein the medical imaging data is stored and managed through a file system and a distributed storage system. The MySQL database has a backup tool mysqldump, which exports all data in the database as an SQL file and stores it. The backup tool mysqldump is provided with a scheduled task, thereby realizing scheduled backup of all data in the database.

[0016] As a further improvement of the present invention, the specific steps for storing data in the data storage and management module are as follows: the user uploads the image file through a POST request and submits the relevant metadata at the same time. After receiving the request, the data storage and management module saves the file to the specified storage path and stores the metadata in the MySQL database; the specific steps for data retrieval and query by the data storage and management module are as follows: the user downloads the specified image file through a GET request, the server retrieves the corresponding file path from the database according to the file ID, and then returns the file to the client.

[0017] As a further improvement of the present invention, it also includes:

[0018] Access control module, which is used to provide different permissions to users of different roles to ensure that each user can only access and operate data and functions within the scope of their permissions;

[0019] Data encryption module, used to encrypt data during data transmission and storage to ensure data confidentiality and integrity;

[0020] Transport layer, which uses TLS / SSL protocol to transmit data;

[0021] Remote access module, used for users to remotely access the database.

[0022] The beneficial effects of the present invention are that the universal medical image-assisted diagnosis system of the present invention can image at extremely low radioactive doses and accurately analyze epileptic lesions. The system can not only significantly reduce the patient's radiation exposure, but also improve the accuracy and reliability of lesion positioning, and provide strong technical support for the diagnosis and treatment of neurological diseases such as epilepsy. The core of the system is from the acquisition of medical images to the generation of the final diagnosis report. This complete process involves multiple key steps. First, medical image data (such as MRI data and PET molecular imaging in Nii format) enters the system and is processed by the ultra-low dose reconstruction module and the NII and Dicom conversion module to ensure the quality and consistency of the image data. Subsequently, the image data enters the intermediary layer and is strictly managed by the data management module to ensure the security and traceability of the data.

[0023] In the image analysis stage, the system uses a variety of advanced technologies and algorithms, such as image segmentation, feature extraction, classification and recognition. In particular, the PET molecular image segmentation module can accurately extract the lesion area and provide strong support for subsequent diagnosis. At the same time, the front-end medical view control module provides users with flexible view operation options, such as window level control, multi-axis section control and color control, so that users can view and analyze medical images more intuitively. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the overall architecture diagram of the universal medical image-assisted diagnosis system;

[0025] Figure 2 This is a flowchart of the full-process integrated PET molecular imaging reconstruction system based on the attention encoder;

[0026] Figure 3 The left side shows the original low-dose image, and the right side shows the reconstructed image;

[0027] Figure 4 The effect of the system on epileptic lesion segmentation;

[0028] Figure 5 This is a schematic diagram of the system user interface;

[0029] Figure 6 A schematic diagram of the user operation guide;

[0030] Figure 7 It is the login interface of the system;

[0031] Figure 8 It is the main interface of this system. DETAILED DESCRIPTION

[0032] The present invention will be further described in detail below in conjunction with the embodiments given in the accompanying drawings. This embodiment obtains a data set containing 200 brains from the MICCAI2022PET Image Reconstruction Global Challenge, which contains the original PET scan data and its corresponding reference images. In order to ensure the effectiveness of model training, the present invention performs necessary preprocessing on the PET scan data, such as denoising, normalization, correction, etc., to eliminate potential data noise, improve data quality, and ensure consistency between data. Next, we divide the data set into a training set, a validation set, and a test set to support subsequent model training, validation, and testing.

[0033] Reference Figure 1 As shown, a universal medical image-assisted diagnosis system for ultra-low radioactivity imaging and epileptic lesion localization analysis in this embodiment includes:

[0034] Ultra-low radioactivity imaging module, such as Figure 2 and Figure 3 As shown, the present invention uses a 3D CycleGAN network as the basic framework for PET image reconstruction. It can reconstruct high-quality medical images at extremely low radioactive doses, and reduce the interference of noise and artifacts through multi-scale and multi-directional reconstruction strategies. The 3D CycleGAN network is a variant of a generative adversarial network (GAN), which is particularly suitable for processing three-dimensional image data. The network consists of two mirror-symmetrical generators (Generators) and two discriminators (Discriminators), which are responsible for generating high-quality reconstructed images from PET scan data and judging whether the generated images are close to the real reference images. In the dual-scale fusion stage, the input low-dose PET data is first preprocessed through dual-scale image processing, including weight transfer, weight normalization, and weight update. Subsequently, the dual-scale generator uses an attention mechanism to extract key features and generate a preliminary reconstructed image, i.e., a rough reconstructed PET, wherein before training the 3D CycleGAN network, the present invention uses a simple interpolation and filtering algorithm to make an initial estimate of the PET image, and then uses an iterative algorithm to train and optimize the 3D CycleGAN network. During the training process, the generator attempts to generate high-quality reconstructed images from PET scan data, while the discriminator determines whether the generated image is close to the real reference image. Through continuous iteration and optimization, the generator gradually learns to extract useful information from the PET scan data and generate reconstruction results that are closer to the real image. At the same time, the present invention incorporates multi-scale and multi-directional reconstruction strategies into the design of the 3D CycleGAN network to reconstruct PET images at different scales and directions. At the same time, the convergence of the 3D CycleGAN network is judged. When the network reaches the convergence condition, the training is stopped and the model parameters are saved. After the final PET image is generated, a post-processing step is performed to further improve the image quality. This includes operations such as filtering and smoothing, which are intended to eliminate residual noise and artifacts in the image and improve the clarity and readability of the image.

[0035] After entering the multi-scale spatial fusion stage, the system performs multi-scale spatial transformation on the initially reconstructed image to capture image features at different scales. Then, with the help of the attention mechanism, the system selects features that have an important impact on the reconstruction results and integrates these features into a more refined reconstruction process to generate a finely reconstructed PET.

[0036] The attention encoder framework plays a core role in the whole process. It is responsible for integrating and transmitting attention information to optimize the reconstruction results. The framework consists of multiple modules, including convolutional blocks, Resnet blocks, forward propagation modules, multi-head self-attention modules, and backward propagation modules, which work together to ensure efficient operation and accurate reconstruction of the system.

[0037] Epilepsy lesion localization analysis module, the present invention systematically collected the original brain data of 140 epilepsy patients from the hospital's positron emission tomography (PET) equipment. These data cover imaging data under different conditions and different imaging conditions. In addition, two professional neuroimaging physicians with rich clinical experience have made detailed labeling (labeling) of the lesions for the brain data of these 140 patients. Finally, the labeled data set is divided into a training set, a validation set, and a test set, such as Figure 4 As shown, it is possible to reconstruct high-quality medical images at extremely low radiation doses and reduce the interference of noise and artifacts through multi-scale and multi-directional reconstruction strategies.

[0038] Based on the adaptive medical image segmentation framework (NNUNet), the present invention selects a network model suitable for processing medical images after ultra-low radioactivity imaging. During the model training process, data preprocessing is first performed to improve image quality, and then the network model is trained using the training set, and the model's loss function and performance indicators (such as Dice coefficient, IoU) are monitored to ensure that it can accurately identify and locate epileptic lesions. The present invention adjusts the model's hyperparameters through the validation set, including learning rate, batch size, and number of iterations. And in this process, grid search or random search and other strategies are used as much as possible to find the best hyperparameter combination to ensure that the model achieves the best effect in identifying epileptic lesions.

[0039] User interface, such as Figure 5 and Figure 8As shown, the React front-end framework and Cornerstone medical front-end framework are mainly used to build the user interface. The present invention utilizes the componentization characteristics of React to split the interface of the medical system into multiple reusable components (such as medical image operation components, medical image display components, etc.). Afterwards, the Cornerstone framework is integrated into the React component to realize the display and processing functions of medical images. Through the API and tools provided by Cornerstone, operations such as loading, displaying, scaling, and rotating medical images can be easily realized. And through the state management mechanism of React (such as Redux or MobX), the state of the medical system (such as the currently used tools, the currently displayed image address, etc.) is uniformly managed to meet the usage habits of different doctors, and support the import and export of multiple image formats and resolutions. In order to ensure that users can give full play to the functions of the system, the system also provides online help documents and training resources. The online help document details the use of each module and plug-in, solutions to common problems, and best practices to help doctors get started quickly and operate efficiently.

[0040] Data storage and management modules, such as Figure 1 As shown in the left column, the medical server is mainly built using the Flask framework of Python, and user information is stored in a MySQL database. The present invention uses the MySQL database to store the metadata of user information and image data, including patient information, image file path and other information. In addition, the image file itself is managed by a file system and a distributed storage system due to its large volume, which can ensure the flexibility of storage and improve the reading speed of data. In order to prevent data loss, a regular backup mechanism is included in the module design. Using the backup tool mysqldump that MySQL carries, all data in the database can be exported as a SQL file and stored in a safe location. By setting a scheduled task (such as Linux cron job), regular automatic backup can be achieved every day. The backup file can be stored in a local server or uploaded to a cloud storage service to further enhance the security of the data.

[0041] The data storage and management module builds a RESTful API service through the Flask framework, providing an interface for uploading, downloading and querying image data. Users can upload image files through POST requests and submit relevant metadata (such as patient ID, image type, etc.). After receiving the request, the server will save the file to the specified storage path and store the metadata in the MySQL database. Users can download the specified image file through GET requests. The server retrieves the corresponding file path from the database based on the file ID and then returns the file to the client.

[0042] Format compatible modules, such as Figure 8 As shown in the left column, this system is compatible with a variety of medical image formats, including but not limited to DICOM (Digital Imaging and Communications in Medicine). It is widely used in CT, MRI, and other imaging equipment. The system has a built-in DICOM parser that can easily import and process DICOM files. At the same time, the system also supports the loading and display of other common image formats (CT, MRI) to meet the needs of different scenarios.

[0043] Access control modules, such as Figure 7 As shown in the figure, the system adopts a role-based access control (RBAC) mechanism. Users with different roles have different permissions, ensuring that each user can only access and operate data and functions within their permissions. Administrators can flexibly configure user roles and permissions through the management interface to ensure the security of the system. The Advanced Encryption Standard (AES) is used for encryption during data transmission and storage to ensure the confidentiality and integrity of the data. The transport layer uses the TLS / SSL protocol to prevent data from being stolen or tampered with during network transmission. Sensitive information stored in the database, such as patient personal information and medical records, is also encrypted and stored to prevent unauthorized access. The system records all doctors' operation logs, including key operations such as login, data access, and data modification, to prevent illegal operations by internal personnel. Log data can be stored for a long time and supports on-demand export and analysis.

[0044] Remote access modules, such as Figure 7 and 8 As shown in the figure, the patient's medical imaging data and medical records can be accessed through the system at a remote location for remote diagnosis. The system provides a wealth of image processing and analysis tools to help doctors complete diagnosis work efficiently. The diagnosis results can be directly fed back to patients and local medical institutions through the system, realizing seamless remote medical services.

[0045] The optimization and upgrade module and system have the ability to continuously optimize and upgrade, and can be continuously improved and perfected according to clinical needs and technological development, supporting software updates and hardware upgrades. The system supports personalized settings, allowing medical institutions to adjust and optimize according to their own special needs to improve the efficiency and accuracy of the workflow. During the system upgrade process, we ensure the integrity and security of the data, provide seamless data migration services, and avoid data loss or damage.

[0046] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A universal medical image-assisted diagnosis system, characterized in that: include: Ultra-low radioactivity imaging module, which uses the 3D CycleGAN network as the basic framework for PET image reconstruction to perform ultra-low radioactivity imaging; Epilepsy lesion localization and analysis module: This module automatically processes and analyzes medical images after ultra-low radioactivity imaging based on the adaptive medical image segmentation framework in deep learning. It can accurately identify and locate epilepsy lesions and provide detailed information on the lesions. User interface, which is used to provide a graphical operation interface, supports multiple interaction modes to meet the usage habits of different doctors, and supports the import and export of multiple image formats and resolutions; Data storage and management module, which is responsible for the storage, backup and management of all medical imaging data and related information in the system. It has data retrieval and query functions and supports remote access and sharing.

2. The universal medical image-assisted diagnosis system according to claim 1, characterized in that: The 3D CycleGAN network used in the ultra-low radioactivity imaging module consists of two mirror-symmetrical generators and two discriminators, which are responsible for generating high-quality reconstructed images from PET scan data and judging whether the generated images are close to the real reference images.

3. The universal medical image-assisted diagnosis system according to claim 2, characterized in that: The specific steps of the ultra-low radioactivity imaging module to generate high-quality reconstructed images from PET scan data are as follows: first, the input low-dose PET data is preprocessed through dual-scale image processing, including weight transfer, weight normalization and weight update processing steps. Subsequently, the dual-scale generator uses the attention mechanism to extract key features and generate a preliminary reconstructed image, i.e., a coarse reconstructed PET. The preliminary reconstructed image is then subjected to a multi-scale spatial transformation to capture image features at different scales. Then, with the help of the attention mechanism, the system screens out features that have an important influence on the reconstruction results, and integrates these features into a more refined reconstruction process to generate a finely reconstructed PET.

4. The universal medical image-assisted diagnosis system according to claim 3, characterized in that: Before training the 3D CycleGAN network, the ultra-low radioactivity imaging module of the present invention uses a simple interpolation and filtering algorithm to perform an initial estimation of the PET image, and then uses an iterative algorithm to train and optimize the 3D CycleGAN network. During the training process, the generator generates a high-quality reconstructed image from the PET scan data, and the discriminator determines whether the generated image is close to the real reference image. The generator and the discriminator perform the above operations repeatedly to continuously iterate and optimize until the generator learns to extract useful information from the PET scan data and generates a reconstruction result that is closer to the real image.

5. The universal medical image-assisted diagnosis system according to claim 4, characterized in that: The specific steps for the epileptic lesion localization analysis module to identify and locate epileptic lesions and provide detailed information about the lesions are as follows: first, the medical image data is preprocessed to improve the image quality, and then the adaptive medical image segmentation framework network model is trained using the training set, and the loss function and performance indicators of the adaptive medical image segmentation framework network model are monitored to ensure that it can accurately identify and locate epileptic lesions.

6. The universal medical image-assisted diagnosis system according to claim 5, characterized in that: The data storage and management module is a MySQL database, wherein the medical imaging data is stored and managed through a file system and a distributed storage system. The MySQL database has a backup tool mysqldump, which exports all data in the database as an SQL file and stores it. The backup tool mysqldump is provided with a scheduled task, thereby realizing scheduled backup of all data in the database.

7. The universal medical image-assisted diagnosis system according to claim 6, characterized in that: The specific steps for storing data in the data storage and management module are as follows: the user uploads the image file through the POST request and submits the relevant metadata at the same time. After receiving the request, the data storage and management module saves the file to the specified storage path and stores the metadata in the MySQL database; the specific steps for data retrieval and query by the data storage and management module are as follows: the user downloads the specified image file through the GET request, the server retrieves the corresponding file path from the database according to the file ID, and then returns the file to the client.

8. The universal medical image-assisted diagnosis system according to claim 7, characterized in that: Also includes: Access control module, which is used to provide different permissions to users with different roles to ensure that each user can only access and operate data and functions within the scope of their permissions; Data encryption module, used to encrypt data during data transmission and storage to ensure data confidentiality and integrity; Transport layer, which uses TLS / SSL protocol to transmit data; Remote access module, used for users to remotely access the database.

Images