Deep learning assisted tumor radiotherapy dose setting method and Web application system

Abstract

The invention discloses a deep learning assisted tumor radiotherapy dose setting method and system. The method comprises the following steps: selecting a to-be-analyzed tumor medical image; converting the tumor medical image into a visual pixel index image by using a tumor sub-region division model, and dividing pixels in the visual pixel index image into tumor sub-regions with high, medium and low risk levels; setting a radiotherapy dose, and calling the tumor subregion radiotherapy result prediction model to carry out radiotherapy reaction prediction so as to generate a radiotherapy reaction prediction result; and setting a tumor radiotherapy dose corresponding to the tumor medical image according to the radiotherapy reaction prediction result. According to the method, the tumor subregion division model based on the wavelet clustering Wavelluster algorithm is introduced, and compared with manual focus sketching, the efficiency is higher; and the radiotherapy result is predicted through the tumor subregion radiotherapy result prediction model, so that the accuracy and the efficiency of setting the radiotherapy dose are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the intersection of artificial intelligence and medical image processing technology, and more specifically, to a method for deep learning-assisted setting of tumor radiotherapy dose and a Web application system. Background Art

[0002] Malignant tumors are increasingly threatening human health. Radiotherapy (abbreviated as radiotherapy) is one of the three main means of treating malignant tumors. Currently, about 70% of cancer patients need to receive radiotherapy for different purposes at different stages of the disease. The implementation of radiotherapy technology depends on equipment and doctor experience. The location, size and shape of the tumor are determined by imaging equipment such as CT and MR (magnetic resonance imaging). The radiotherapy doctor marks the area to be irradiated on the patient's body surface or on the computer based on the above imaging results, and then a special medical physicist formulates a corresponding radiotherapy plan to treat the patient.

[0003] Traditional lesion delineation requires the radiotherapist to manually delineate using professional software, which usually includes: tumor boundary: the doctor manually determines the tumor boundary based on imaging features (such as shape and density); target area division: the target area is divided into high-risk and low-risk areas based on different characteristics of the tumor. Multi-layer delineation: repeated delineation on images at different levels to ensure the accuracy of the tumor volume. Manual delineation can be guaranteed by the doctor's experience to ensure a certain degree of accuracy, but there are problems of low efficiency and strong subjectivity.

[0004] In addition, the setting of radiotherapy dose is also very important. Radiotherapy planning mainly includes dose calculation and tuning. Doctors need to use the radiotherapy planning system to calculate the required radiotherapy dose according to the outlined target area, and then work with physicists to adjust the radiotherapy plan to ensure that the maximum dose is concentrated on the tumor while minimizing damage to surrounding normal tissues. The current dose setting is mainly based on the doctor's combination of the known safe irradiation dose that normal tissues can accept (irradiated volume dose and maximum dose, etc.), the patient's individual conditions and the doctor's personal experience to set the radiotherapy dose, but there is a strong problem of subjectivity and a high risk of error.

[0005] In summary, the traditional tumor medical imaging delineation and radiotherapy dose setting process is highly dependent on the professional judgment and experience of doctors. Although these methods are effective to a certain extent, they may be affected by subjective factors, are time-consuming, and have low efficiency. Therefore, a more automated and intelligent solution is urgently needed to improve the efficiency and accuracy of the tumor radiotherapy process. Summary of the invention

[0006] In view of the defects of the prior art, the present invention provides a method and a Web application system for setting tumor radiotherapy doses assisted by deep learning, which effectively solves the problems of long tumor sub-region delineation time, long dose setting cycle and low prediction accuracy in traditional methods.

[0007] To achieve the above objectives, on the one hand, the present invention provides a method for setting tumor radiotherapy dose by deep learning assistance, the method comprising:

[0008] Step S1, selecting a tumor medical image to be analyzed;

[0009] Step S2, converting the tumor medical image into a visualized pixel index image using a tumor sub-region division model, and dividing the pixels in the visualized pixel index image into tumor sub-regions of three risk levels: high, medium, and low;

[0010] Step S3, setting radiotherapy doses for tumor sub-regions of different risk levels respectively, and calling a tumor sub-region radiotherapy result prediction model to perform radiotherapy response prediction to generate a radiotherapy response prediction result;

[0011] Step S4, setting the tumor radiotherapy dose corresponding to the tumor medical image according to the radiotherapy response prediction result.

[0012] In some of the embodiments, in step S2, the tumor subregion division model is a tumor subregion division model based on a wavelet clustering Wavecluster algorithm, and the tumor subregion division model analyzes the imaging genomics characteristics of the tumor medical image and performs risk level division of tumor subregions based on the wavelet clustering Wavecluster algorithm.

[0013] In some of the embodiments, in step S3, the tumor sub-region radiotherapy result prediction model is a pre-trained PatchGAN-Unet radiotherapy result prediction model.

[0014] In some of the embodiments, in step S3, the radiotherapy response prediction is specifically: for any pixel of the tumor in the tumor medical image, according to the radiotherapy dose set for the sub-region to which the pixel belongs, predicting the standard uptake value of the pixel after radiotherapy.

[0015] In some embodiments, the method further comprises:

[0016] Step S5: persistently storing the radiotherapy response prediction result.

[0017] In some embodiments, step S5 specifically includes:

[0018] Encrypting the radiotherapy response prediction result;

[0019] Upload the encrypted data to the specified bucket of the Object Storage Service (OSS) through HTTP request;

[0020] Persistently store the unique object URL returned by OSS.

[0021] On the other hand, the present invention also discloses a Web application system for deep learning-assisted tumor radiotherapy dose setting, including: a user interface, a gateway, an application subsystem, a database and a cloud server;

[0022] The user interface is used to provide a user interaction front-end page accessed through a browser, and the user interface communicates securely with the gateway through the HTTP protocol;

[0023] The gateway is used to process all HTTP requests in and out of the Web application system and exchange data with the application subsystem;

[0024] The application subsystem includes a data processing module, a patient information management module, a tumor sub-region division module, a radiotherapy dose setting module and a radiotherapy result prediction module;

[0025] The database is used to store patient information, tumor medical images and radiotherapy setting data;

[0026] The cloud server is used to host and run applications, provide computing resources and storage space;

[0027] The cloud server is connected to the database, and when running the user interface, the gateway and the application subsystem, the method for setting tumor radiotherapy dose assisted by deep learning described in the above embodiment is adopted.

[0028] In some of the embodiments, the user interface includes a patient information management interface, a tumor sub-region division interface, and a radiotherapy dose setting and outcome prediction interface;

[0029] The patient information management interface, the tumor sub-region division interface and the radiotherapy dose setting and result prediction interface are connected to the patient information management module, the tumor sub-region division module and the radiotherapy dose setting and result prediction module respectively through the data processing module;

[0030] The patient information management interface is used to perform dual authentication and authorization with the hospital PACS system through the patient information management module to obtain patient information and tumor medical images;

[0031] The tumor sub-region division interface is used to divide the tumor medical image into three tumor sub-regions of high, medium and low risk levels through the tumor sub-region division module;

[0032] The radiotherapy dose setting and result prediction interface is used to provide dose input boxes for three tumor sub-regions: high, medium and low, so as to input the set radiotherapy dose into the radiotherapy dose setting module, generate radiotherapy response prediction results through the radiotherapy result prediction module, and update and store the radiotherapy response prediction results.

[0033] In some of the embodiments, the radiotherapy dose setting and result prediction interface has a dose range prompt.

[0034] In some of the embodiments, the data processing module is used to perform Salt value encryption processing on the data of the Web application system, and store the encrypted data together with the Salt value in a database;

[0035] The patient information management module is used to perform dual authentication and authorization on users in conjunction with the hospital PACS system to obtain patient information and tumor medical images;

[0036] The tumor sub-region division module is connected to the patient information management module to convert the tumor medical image into a visualized pixel index image using the tumor sub-region division model, and divide the pixels in the visualized pixel index image into tumor sub-regions of three risk levels: high, medium and low;

[0037] The radiotherapy dose setting module is connected to the tumor sub-region division module to set the radiotherapy doses for the tumor sub-regions of the three risk levels of high, medium and low respectively;

[0038] The radiotherapy dose prediction module is connected to the radiotherapy dose setting module to call the tumor sub-region radiotherapy result prediction model to predict the radiotherapy response of the tumor sub-regions with the three risk levels of high, medium and low according to different radiotherapy doses to generate radiotherapy response prediction results, and update and store the radiotherapy response prediction results.

[0039] Compared with the prior art, the above invention has the following advantages or beneficial effects:

[0040] (1) The present invention introduces a tumor sub-region division model based on the wavelet clustering Wavecluster algorithm, which is more efficient than manually outlining lesions. Furthermore, combined with a deep learning algorithm, it can generate radiotherapy response prediction results for different radiotherapy doses based on the PatchGAN-Unet radiotherapy result prediction model, and then the tumor radiotherapy dose corresponding to the tumor medical image can be personalized according to the radiotherapy response prediction results, thereby significantly improving the accuracy and efficiency of radiotherapy dose setting and reducing the workload of doctors.

[0041] (2) The present invention can perform dual authentication with the hospital PACS system, so that the Web application system can obtain the patient's medical images and clinical data in real time, quickly respond to clinical needs, and ensure the timeliness and effectiveness of radiotherapy plans. OSS is used for medical image storage, combined with strict access control and data encryption measures, to effectively ensure the security and privacy of patient information.

[0042] (3) The present invention provides an intuitive visual user interface that enables doctors to easily view tumor distribution, risk level, and predicted radiotherapy results. It supports multiple settings of radiotherapy doses to compare predicted results. Doctors can choose the best plan based on the patient's specific situation, thereby achieving personalized treatment and improving the scientificity and reliability of clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention and its features and advantages will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following accompanying drawings.

[0044] Figure 1 A flow chart of the method for deep learning-assisted setting of tumor radiotherapy dose in the present invention;

[0045] Figure 2 A schematic diagram of the software architecture of the Web application system in the present invention;

[0046] Figure 3 A schematic diagram of the deployment architecture of the Web application system in the present invention;

[0047] Figure 4 is a schematic diagram of an application subsystem in the present invention;

[0048] Figure 5 It is the patient information retrieval interface of the user interface of the present invention;

[0049] Figure 6 The tumor medical image selection interface of the user interface of the present invention;

[0050] Figure 7 It is a tumor sub-region division interface of the user interface of the present invention;

[0051] Figure 8 This is the tumor radiation dose setting and radiotherapy response result prediction interface of the user interface in the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0053] The terms "comprises", "including" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or apparatuses.

[0054] In the following detailed description, many specific details are set forth to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that well-known control algorithms are not shown in detail to avoid obscuring the main purpose of the present invention.

[0055] Embodiment 1:

[0056] like Figure 1 As shown, this embodiment provides a method for deep learning-assisted setting of tumor radiotherapy dose, which can be used to set lung tumor radiotherapy dose or other tumor radiation dose as needed; specifically, the method includes the following steps:

[0057] Step S1, select the tumor medical image to be analyzed; specifically, in this embodiment, the patient information management module is integrated with the hospital PACS system to ensure data consistency and real-time performance. A dual identity authentication mechanism is used to ensure that only authorized users can access sensitive patient information. The user can quickly retrieve detailed information about the relevant patient by entering the patient's ID number. The basic information to be provided includes ID photo, name, ID number, gender, age, chief complaint, current medical history and preliminary diagnosis to ensure that the doctor can fully understand the patient's background before treatment. The user needs to carefully check the current medical history and preliminary diagnosis records to ensure the accuracy and completeness of the information. When necessary, the user can update the information, and the system automatically records all changes to ensure the real-time and traceability of the data. All medical images and basic image information related to the patient are retrieved from the database according to the selected patient ID number to ensure that the doctor obtains comprehensive imaging data. The basic information includes the image shooting date, the patient's physical condition (such as weight, complications, etc.) and the doctor's remarks to help the user better understand the image background. The user browses the retrieved medical images and selects the tumor medical images to be analyzed as the tumor medical images to be analyzed in combination with the basic information.

[0058] Furthermore, the storage of medical images retrieved from the above database uses OSS for persistent storage. The storage and management of medical images are realized through OSS, and service interaction is carried out through the RESTful API provided by OSS, which supports image upload, download and management operations. According to the selected OSS, the corresponding software development kit (SDK) is selected to simplify service calls and ensure seamless connection with the backend Spring Boot framework.

[0059] Step S2, using the tumor sub-region division model to convert the tumor medical image into a visual pixel index image through a clustering algorithm, and divide the pixels in the visual pixel index image into three tumor sub-regions of high, medium and low risk levels, so that doctors can quickly identify the key treatment areas.

[0060] Specifically, the tumor sub-regional division model analyzes the radiomics features of tumor medical images based on the wavelet clustering Wavecluster algorithm and divides the risk levels of tumor sub-regions. The tumor sub-regional division model is based on the wavelet clustering Wavecluster algorithm, which can effectively analyze the radiomics features and perform sub-regional division. Among them, the radiomics features are image feature information extracted from the region of interest (ROI) of the image by using radiological imaging methods combined with automated algorithms. This information can be deeply analyzed through statistical analysis and data mining, and the key information obtained by the analysis will further assist in the classification and grading of image sub-regions. Radiomics feature data usually have multidimensionality (non-linear and multi-scale) and large-scale characteristics, and are suitable for the use of the grid-based Wavecluster wavelet fast clustering algorithm. The algorithm has the characteristics of non-necessity of clustering preset and low sensitivity to data flow order, and can comprehensively analyze the tumor image structure. It is the core of the model's accurate division of sub-regions.

[0061] Step S3, setting radiotherapy doses for tumor sub-regions of different risk levels, that is, setting radiotherapy doses for tumor sub-regions of high, medium and low risk levels, and using a tumor sub-region radiotherapy result prediction model to predict radiotherapy responses for tumor sub-regions of high, medium and low risk levels according to different radiation doses to generate radiotherapy response prediction results;

[0062] Specifically, the user interface can provide input boxes for radiotherapy doses of high, medium and low sub-regions. Users can input different radiotherapy doses based on clinical experience. The input box design is intuitive and easy to use, and supports dose range prompts, thereby reducing the difficulty of user operation. The tumor sub-region radiotherapy result prediction model is a pre-trained PatchGAN-Unet radiotherapy result prediction model, which uses big data analysis technology to improve the accuracy of prediction. Specifically, after the user enters the radiotherapy dose, the pre-trained PatchGAN-Unet radiotherapy result prediction model is called through the Flask framework to automatically generate the PatchGAN-Unet radiotherapy response result prediction, saving time and improving efficiency. Users can combine the original sub-region map with the radiotherapy result prediction map for in-depth analysis to evaluate the radiotherapy effect. According to the prediction results, users can flexibly adjust the radiotherapy dosage settings of the three sub-regions until the results meet clinical expectations to ensure the maximum effectiveness of the treatment.

[0063] Among them, the PatchGAN-Unet radiotherapy result prediction refers to predicting the standard uptake value (SUV value) of any pixel of the tumor in the tumor medical image (PET / CT image) before radiotherapy, based on the radiotherapy dose set by the user for the sub-region to which the pixel belongs, in order to achieve the purpose of predicting the effect of radiotherapy. Among them, SUV (Standardized Uptake Value) is a quantitative indicator used in radioactive medical imaging, mainly used in PET (positron emission tomography) imaging. The SUV calculation formula is SUV is used to evaluate the uptake of radioactive tracers by tumor tissue. The SUV value reflects the metabolic activity of tumor cells. The higher the SUV value, the stronger the activity of tumor cells. By comparing the changes in SUV values ​​at different time points, the treatment effect and disease progression can be evaluated.

[0064] Among them, the above-mentioned Flask framework is deeply compatible with the Python language used in the tumor sub-regional radiotherapy outcome prediction model. The framework is used to encapsulate the prediction service of the deep learning model so that it can be called as an independent service. The services based on this framework are restricted to intranet access by default to protect the model and data from unauthorized access, and peripheral requests are managed through the gateway.

[0065] Step S4, setting the tumor radiotherapy dose corresponding to the tumor medical image according to the radiotherapy response prediction result.

[0066] Step S5: persistently storing the radiotherapy response prediction results.

[0067] The step S5 specifically includes:

[0068] The radiotherapy response prediction results (specifically, the prediction result graph generated by the tumor sub-region radiotherapy result prediction model) are encrypted.

[0069] Upload the encrypted data to the specified storage bucket of OSS through HTTP request.

[0070] Persistently store the unique object URL returned by OSS.

[0071] In this embodiment, during the process of persistent storage of the radiotherapy response prediction results, metadata such as the set sub-regional dose can be stored together to ensure data integrity and retrieval efficiency; specifically, the storage content includes the original sub-regional image, the radiotherapy result prediction image and the sub-regional radiotherapy dose.

[0072] In step S5, the relational database MySQL is used in persistent storage to ensure structured storage and efficient query of data. The data model in the database follows the principle of normalization, and multiple tables are designed to reduce data redundancy and improve data consistency. The tables in the database include: user table (Users), patient information table (Patients), medical image table (Medical_Images), tumor sub-region table (Tumor_Subtargets), and radiotherapy prediction record table (Radiation_Records).

[0073] Specifically, the user table (Users) is used to store user identity information and supports user verification and permission management. Specific fields include: user_id (primary key, uniquely identifies the user), username (user name), password_hash (password hash value), salt (Salt value), role (user role, such as doctor, administrator), last_login (last login time).

[0074] The patient information table (Patients) is used to store basic patient information for easy retrieval and management by doctors. The specific fields include: patient_id (primary key, uniquely identifies the patient), user_id (foreign key, associated with the user table), name (patient name), id_number (ID number), gender (gender), age (age), chief complaint (chief complaint), medical_history (current medical history), initial_diagnosis (initial diagnosis), photo (file path of ID photo).

[0075] The Medical_Images table is used to store medical image information related to patients for easy retrieval and display. The specific fields include: image_id (primary key, uniquely identifies the image), patient_id (foreign key, associated with the patient table), image_path (storage path of the image file), capture_date (image shooting date), patient's physical condition during capture (patient's physical condition during capture), doctor_notes (doctor's notes).

[0076] The tumor sub-region table (Tumor_Subtargets) is used to store detailed information about each tumor sub-region to support subsequent radiotherapy dose setting. Specific fields include: subtarget_id (primary key, uniquely identifies the sub-region), image_id (foreign key, associated with the medical image table), risk_level (risk level, high, medium, low), segmentation_data (segmentation data, storing pixel information of the tumor region).

[0077] The radiation therapy prediction record table (Radiation_Records) is used to store the radiation therapy dose settings and corresponding prediction results, so as to facilitate the evaluation and improvement of treatment effects. The specific fields include: record_id (primary key, uniquely identifies the record), patient_id (foreign key, associated with the patient table), subtarget_id (foreign key, associated with the tumor sub-region table), radiation_dose (radiation therapy dose), predicted_response (predicted radiation therapy results), and timestamp (recording time).

[0078] Embodiment 2:

[0079] like Figures 2 to 4 As shown, this embodiment also provides a Web application system for deep learning-assisted tumor radiotherapy dose setting. Specifically, the Web application system includes: a user interface (UI, user interface layer), a gateway (gateway layer), an application subsystem (application layer), a database (service layer) and a cloud server. The application subsystem includes a data processing module, a patient information management module, a tumor sub-region division module, a radiotherapy dose setting module and a radiotherapy result prediction module. The data processing module is used to perform Salt value encryption processing on the data of the Web application system and store the encrypted data together with the Salt value in the database.

[0080] The above user interface is used to provide a user interaction front-end page accessed through a browser (such as a web browser), and the user interface communicates securely with the gateway through the HTTP protocol. The user interface is built based on the Vue3 framework, uses JavaScript to process data logic, uses HTML5 and CSS to set page styles, and provides a smooth and responsive user experience; the user interface is used by doctors and medical personnel.

[0081] Specifically, the user interface includes a patient information management interface, a tumor sub-region division interface, and a radiotherapy dose setting and result prediction interface; the patient information management interface, the tumor sub-region division interface, and the radiotherapy dose setting and result prediction interface are connected to the patient information management module, the tumor sub-region division module, and the radiotherapy dose setting and result prediction module through the data processing module respectively; the patient information management interface is used to perform dual authentication and authorization with the hospital PACS system through the patient information management module to obtain patient information and tumor medical images (that is, the patient information management module can be used to perform information interaction with the hospital PACS system through the patient information management module to perform user login and patient information retrieval, and to display and browse the patient's medical images); the tumor sub-region division interface is used to divide the tumor medical image into three tumor sub-regions of high, medium and low risk levels through the tumor sub-region division module; the radiotherapy dose setting and result prediction interface is used to provide dose input boxes of high, medium and low tumor sub-regions to input the set radiotherapy dose to the radiotherapy dose setting module (that is, the user is allowed to input the radiotherapy dose for the sub-regions of different risk levels; The user sets the range of sub-regional radiotherapy dose recommendations according to the system prompts, enters the radiotherapy doses for three sub-regions (high, medium and low), and after generating the radiotherapy response prediction results through the radiotherapy result prediction module, displays the radiotherapy response prediction results predicted by the deep learning model, and updates and stores the radiotherapy response prediction results.

[0082] The above gateway is used to process all HTTP requests in and out of the Web application system and perform key access control on sensitive requests.

[0083] Encrypt and authenticate the front-end post request, and exchange data with the application subsystem through the internal communication protocol.

[0084] like Figure 3As shown, the patient information management module is used to perform dual authentication and authorization on users in combination with the hospital PACS system to obtain patient information and tumor medical images; that is, the patient information management module allows authorized users to retrieve and manage patient information, medical images and other data in the hospital PACS system; users log in to the Web application through the dual authentication mechanism, and enter the patient ID number to retrieve patient information, and can manage patient information; according to the patient ID number, relevant tumor medical images and basic image information can be retrieved from the database; specifically, the patient information management module includes a patient information management unit, a patient basic information and condition retrieval unit, and a patient medical image query unit. The tumor sub-region division module is connected to the patient information management module to use the tumor sub-region division model to convert the tumor medical image into a visualized pixel index image, and divide the pixels in the visualized pixel index image into three tumor sub-regions of high, medium and low risk levels; specifically, the tumor sub-region division module includes a training and test data management unit and a tumor sub-region division unit. The radiotherapy dose setting module is connected to the tumor sub-region division module to set the radiotherapy dose for the tumor sub-regions of the three risk levels of high, medium and low respectively; specifically, the radiotherapy dose setting module includes a radiotherapy dose setting unit and a radiotherapy effect prediction unit; the above-mentioned radiotherapy dose prediction module is connected to the radiotherapy result prediction module to call the tumor sub-region radiotherapy result prediction model to predict the radiotherapy response of the tumor sub-regions of the three risk levels of high, medium and low according to different radiotherapy doses to generate radiotherapy response prediction results, and update and store the radiotherapy response prediction results; specifically, the radiotherapy result prediction module includes a training and test data management unit, a prediction model training unit and a tumor sub-region radiotherapy effect prediction unit.

[0085] The above database is used to store patient information, tumor medical images and radiotherapy setting data, and the above database uses Mysql to store structured data. The structured data in the database includes: User table (Users): stores user identity information and permissions. Patient information table (Patients): stores basic information of patients; Medical image table (Medical_Images): stores metadata and storage path of medical images; Tumor sub-region table (Tumor_Subtargets): stores detailed information of tumor sub-regions; Radiotherapy prediction record table (Radiation_Records): stores radiotherapy dose settings and prediction results. The database is combined with OSS to store unstructured data. The unstructured data includes unstructured data such as patient medical image files and prediction result images generated by deep learning models. The database stores the specific storage path of the above unstructured data in OSS.

[0086] The above-mentioned cloud server is used to host and run applications (the above-mentioned user interface, gateway and application subsystem are all applications), provide computing resources and storage space; the cloud server is connected to the database, and when running the above-mentioned user interface, gateway and application subsystem, implements the method of deep learning-assisted setting of tumor radiotherapy dose in the above-mentioned embodiment 1, thereby providing efficient tumor radiotherapy dose setting support.

[0087] The present invention provides functional units such as portal login, patient basic information query, patient condition data query, tumor slice map (tumor medical image) retrieval, tumor sub-region radiotherapy dose setting, deep learning assisted radiotherapy response prediction, etc. The Web application system uses Vue3 and Springboot frameworks to build the front end and back end, the Flask framework calls the deep learning model, and MySQL and OSS store structured and unstructured data respectively. The application integrates the PatchGAN-Unet model, which is based on a large number of lung tumor slice image training. This deep learning model can predict tumor radiotherapy responses at different doses, help medical professionals personalize radiotherapy doses, improve treatment effects, and reduce side effects, and has significant clinical application value.

[0088] The following is a specific implementation method. All data in the interface screenshots and charts involved are simulated data and are only used to illustrate the embodiments and effects of the present invention. These simulated data do not reflect actual clinical data or user data, but are used to demonstrate a specific implementation method of the present invention.

[0089] The patient information management interface of the user interface includes a patient information retrieval interface and an oncology medical image selection interface: FIG5 shows the patient information retrieval interface of the present invention. The doctor accesses the Web application system through a Web browser and logs in using the credentials of the hospital PACS system. After logging in, the user is allowed to retrieve patient information by entering the patient's ID number. The interface is divided into two parts: the left side displays the patient's ID photo, and the right side displays the patient's basic information and related diagnosis and treatment conditions. The user can search for other patient information by entering the ID number, or click the "Next" button to enter the next page.

[0090] The figure shows the tumor medical image selection interface, which is the tumor medical image selection interface of the present invention. In this interface, the doctor can select the patient's tumor image from the left, and the system will automatically display the selected image on the right. After the selection is completed, click the "Next" button, and the system will enter the tumor sub-region division interface.

[0091] Figure 7The figure shows the tumor sub-region division interface of the present invention. In this interface, the present invention divides the tumor medical image selected by the user in the previous interface into three sub-regions of high, medium and low risk levels represented by pixel graphs through the wavelet clustering algorithm through the tumor sub-region division model based on the wavelet clustering Wavecluster algorithm, and displays the division results. The doctor clicks the "Next" button to enter the radiotherapy dose setting and result prediction interface.

[0092] Figure 8 Shown is the radiotherapy dose setting and result prediction interface of the present invention. According to the division results of the tumor sub-regions and the radiotherapy dose values ​​input by the doctor, the present invention calls the pre-trained PatchGAN-Unet tumor sub-region radiotherapy result prediction model to further analyze and diagnose the patient's condition. The doctor can set different doses for the three risk levels of sub-regions in the middle radiotherapy dose area, and click the "Predict" button to predict the treatment effect. The prediction results will be displayed on the right side of the interface. The doctor can continuously adjust the dose value until the most suitable personalized and precise radiotherapy dose distribution for the patient is determined, achieving more accurate and faster personalized treatment. All prediction records, including images and dose settings, are encrypted and stored in the database for future query and analysis.

[0093] The above-mentioned Web application system is developed based on the Vue3 progressive framework and Spring Boot framework, and uses MySQL and OSS for persistent storage of user data and medical image data, with good scalability and user experience.

[0094] The above-mentioned Web application system uses advanced clustering algorithms to efficiently divide complex multi-dimensional medical images into sub-regions. Combined with the user's setting of radiation doses for each sub-region after division, the application uses a pre-trained deep learning model to quickly predict the radiotherapy results of different sub-regions and store these prediction results persistently.

[0095] In summary, the present invention integrates data acquisition, analysis and dose setting into a Web application system, simplifies the clinical workflow, reduces the cumbersome manual operations in traditional methods, and improves the efficiency of tumor radiotherapy plan formulation.

[0096] Although a specific embodiment is described above, the present invention is not limited thereto. For example, the deep learning model can be replaced by other types of machine learning models, the sub-regional division algorithm and the radiotherapy result prediction algorithm can be replaced by other types of model algorithms according to different application scenarios, and the storage solutions for structured and unstructured data can be replaced by other types of data storage solutions. In addition, the user interface can be customized and optimized according to different needs.

[0097] Those skilled in the art should understand that those skilled in the art can implement variations by combining the prior art and the above embodiments, which will not be described in detail here. Such variations do not affect the essential content of the present invention, and will not be described in detail here.

[0098] The above describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the above-mentioned specific embodiments, and the devices and structures that are not described in detail should be understood to be implemented in a common manner in the art; any technician familiar with the art can use the above-disclosed methods and technical contents to make many possible changes and modifications to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, or modify them into equivalent embodiments of equivalent changes, which does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention are still within the scope of protection of the technical solutions of the present invention.

[0099] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

Claims

1. A method for setting tumor radiotherapy dose assisted by deep learning, characterized in that: The method comprises: Step S1, selecting a tumor medical image to be analyzed; Step S2, converting the tumor medical image into a visualized pixel index image using a tumor sub-region division model, and dividing the pixels in the visualized pixel index image into tumor sub-regions of three risk levels: high, medium, and low; Step S3, setting radiotherapy doses for tumor sub-regions of different risk levels respectively, and calling a tumor sub-region radiotherapy result prediction model to perform radiotherapy response prediction to generate a radiotherapy response prediction result; Step S4, setting the tumor radiotherapy dose corresponding to the tumor medical image according to the radiotherapy response prediction result.

2. The method for setting tumor radiotherapy dose by deep learning assistance according to claim 1, characterized in that: In the step S2, the tumor sub-region division model analyzes the radiomics features of the tumor medical image based on the wavelet clustering Wavecluster algorithm and divides the risk levels of the tumor sub-regions.

3. The method for setting tumor radiotherapy dose by deep learning assistance according to claim 1, characterized in that: In step S3, the tumor sub-region radiotherapy result prediction model is a pre-trained PatchGAN-Unet radiotherapy result prediction model.

4. The method for setting tumor radiotherapy dose by deep learning assistance according to claim 1, characterized in that: In step S3, the radiotherapy response prediction is specifically: for any pixel of the tumor in the tumor medical image, according to the radiotherapy dose set for the sub-region to which the pixel belongs, predicting the standard uptake value of the pixel after radiotherapy.

5. The method for setting tumor radiotherapy dose by deep learning assistance according to claim 1, characterized in that: The method further comprises: Step S5: persistently storing the radiotherapy response prediction result.

6. The method for setting tumor radiotherapy dose by deep learning assistance according to claim 5, characterized in that: The step S5 specifically includes: Encrypting the radiotherapy response prediction result; Upload the encrypted data to the specified bucket of the object storage service through HTTP request; Persistently store the unique object URL returned by the object storage service.

7. A Web application system for deep learning-assisted tumor radiotherapy dose setting, characterized in that: include: User interface, gateway, application subsystem, database and cloud server; The user interface is used to provide a user interaction front-end page accessed through a browser, and the user interface communicates securely with the gateway through the HTTP protocol; The gateway is used to process all HTTP requests in and out of the Web application system and exchange data with the application subsystem; The application subsystem includes a data processing module, a patient information management module, a tumor sub-region division module, a radiotherapy dose setting module and a radiotherapy result prediction module; The database is used to store patient information, tumor medical images and radiotherapy setting data; The cloud server is used to host and run applications, provide computing resources and storage space; The cloud server is connected to the database, and implements the method according to any one of claims 1 to 6 when running the user interface, the gateway and the application subsystem.

8. The Web application system for deep learning-assisted tumor radiotherapy dose setting according to claim 7, characterized in that: The user interface includes a patient information management interface, a tumor sub-region division interface, and a radiotherapy dose setting and result prediction interface; The patient information management interface, the tumor sub-region division interface and the radiotherapy dose setting and result prediction interface are connected to the patient information management module, the tumor sub-region division module and the radiotherapy dose setting and result prediction module respectively through the data processing module; The patient information management interface is used to perform dual authentication and authorization with the hospital PACS system through the patient information management module to obtain patient information and tumor medical images; The tumor sub-region division interface is used to divide the tumor medical image into three tumor sub-regions of high, medium and low risk levels through the tumor sub-region division module; The radiotherapy dose setting and result prediction interface is used to provide dose input boxes for three tumor sub-regions: high, medium and low, so as to input the set radiotherapy dose into the radiotherapy dose setting module, generate radiotherapy response prediction results through the radiotherapy result prediction module, and update and store the radiotherapy response prediction results.

9. The Web application system for deep learning-assisted tumor radiotherapy dose setting according to claim 8, characterized in that: The radiotherapy dose setting and result prediction interface has a dose range prompt.

10. The Web application system for deep learning-assisted tumor radiotherapy dose setting according to claim 7, characterized in that: The data processing module is used to perform Salt value encryption processing on the data of the Web application system, and store the encrypted data together with the Salt value in the database; The patient information management module is used to perform dual authentication and authorization on users in conjunction with the hospital PACS system to obtain patient information and tumor medical images; The tumor sub-region division module is connected to the patient information management module to convert the tumor medical image into a visualized pixel index image using the tumor sub-region division model, and divide the pixels in the visualized pixel index image into tumor sub-regions of three risk levels: high, medium and low; The radiotherapy dose setting module is connected to the tumor sub-region division module to set the radiotherapy doses for the tumor sub-regions of the three risk levels of high, medium and low respectively; The radiotherapy dose prediction module is connected to the radiotherapy dose setting module to call the tumor sub-region radiotherapy result prediction model to predict the radiotherapy response of the tumor sub-regions with the three risk levels of high, medium and low according to different radiotherapy doses to generate radiotherapy response prediction results, and update and store the radiotherapy response prediction results.

Images