Space segmentation radiotherapy auxiliary system based on multi-modal image

Abstract

The technical scheme of the invention discloses a spatial segmentation radiotherapy auxiliary system based on a multi-modal image, and the system is characterized in that the system comprises a deformation registration module of the multi-modal image; a target area sketching and high-dose small ball target area sketching module; and a cooperation module. The system disclosed by the invention is used for guiding target region sketching through a multi-modal image deformation registration technology, supporting automatic / manual design of a high-dose small ball target region, and cooperating with a treatment planning system (TPS) to complete a staged treatment process. The system disclosed by the invention finally evaluates the overall treatment effect through deformation registration and dose superposition, and is suitable for precise SFRT treatment planning of equipment such as an accelerator, a TOMO, a cyber knife and the like.

Description

Technical Field

[0001] The present invention relates to a spatially fractionated radiation therapy (SFRT) auxiliary planning system. Background Art

[0002] Spatially fractionated radiation therapy (SFRT) needs to balance tumor killing and normal tissue protection under complex dose distribution, and requires more accurate anatomical and functional imaging fusion. PET / CT (such as FDG, PSMA) can identify high metabolic / high proliferation areas in the tumor and guide the dose hotspot layout of SFRT. MRI (such as DCE-MRI, DWI) distinguishes tumor necrosis and active areas to avoid over-irradiation in inactive areas. Based on pre-treatment multimodal imaging (CT+PET+MRI), heterogeneous areas such as hypoxia and vascular abnormalities in the tumor can be identified. During treatment, tumor regression or functional changes can be monitored through image registration, and the dose fractionation strategy of SFRT can be dynamically adjusted. In different radiotherapy centers, SFRT is usually based on conventional radiotherapy pathways and lacks deep support for the use of multimodal images, resulting in poor accuracy in the delineation of high-dose small ball targets in the target area and within the target area. In terms of high-dose small ball target delineation, there is a lack of automated means and it relies entirely on manual operation, which is inefficient and subjective, and the delineation results of different doctors vary greatly.

[0003] In terms of dose calculation, all processes are usually completed within a single software system, lacking efficient collaboration with other TPS systems, making it difficult to fully utilize the advantages of different systems. At the same time, during and after the treatment, the conversion of physical dose to EQD2, deformation registration, and dose superposition are relatively complex operations. Existing methods often cannot achieve accurate calculation and effective integration, affecting the accurate assessment of treatment prognosis. Summary of the invention

[0004] The purpose of the present invention is to solve the following problems existing in the existing radiotherapy technology: low accuracy in delineating the target area and the high-dose small ball target area inside the target area: relying on single-modality images and manual operations, errors are prone to occur, affecting the treatment effect; low efficiency: manual delineation of high-dose small ball target areas is time-consuming and labor-intensive, and cannot meet the needs of clinical rapid treatment; lack of collaboration: the dose calculation system is closed and cannot work with other TPS, limiting the optimization of the technology; inaccurate dose processing: the dose conversion, alignment and superposition calculations during the treatment process are inaccurate, making it difficult to accurately assess the treatment prognosis.

[0005] In order to achieve the above object, the technical solution of the present invention is to disclose a spatial segmentation radiotherapy auxiliary system based on multimodal images, which is characterized by comprising:

[0006] The deformation registration module for multi-modal images is used to import multi-modal images before the start of treatment and after the first-stage treatment, fuse images of different modalities, achieve precise registration of multi-modal images, and obtain precise fused images;

[0007] The target area delineation and high-dose small ball target area delineation module is used to, before the start of treatment and after the first-stage treatment, on the precise fused images output by the deformation registration module, complete the delineation of the outer contour target area by manual delineation method, and complete the delineation of the high-dose small ball target area by manual delineation method or automatic delineation method, where:

[0008] When using the manual delineation method to delineate the outer contour target area, the doctor uses the delineation tool provided by the target area delineation and high-dose small ball target area delineation module to delineate the target area, and the target area delineation and high-dose small ball target area delineation module provides auxiliary reference lines and marks based on the precise fused images to help the doctor more accurately determine the target area boundary and complete the delineation of the outer contour target area;

[0009] When using the manual delineation method to delineate the high-dose small ball target area:

[0010] Set the desired small ball radius value;

[0011] Set the small ball coordinates;

[0012] The target area delineation and high-dose small ball target area delineation module generates a small ball structure based on the set radius value and small ball coordinates;

[0013] The target area delineation and high-dose small ball target area delineation module calculates the distance between the currently generated small ball and other already generated small balls, judges whether it meets the pre-set geometric requirements. If it does not meet the requirements, re-set the desired small ball radius value and small ball coordinates, and the target area delineation and high-dose small ball target area delineation module regenerates the small ball structure based on the updated parameters and judges again until it meets the requirements;

[0014] Save the small ball structure that meets the requirements;

[0015] When using the automatic delineation method to delineate the high-dose small ball target area:

[0016] Set the small ball radius and the number of generated small balls;

[0017] The target area delineation and high-dose small ball target area delineation module randomly generates 10 coordinate points within the delineation area and judges whether the geometric requirements are met between these coordinate points. If not, re-generate the coordinate points until the generated coordinate points meet the requirements;

[0018] For the coordinate points that meet the requirements, the target area delineation and high-dose small ball target area delineation module generates small ball structures according to each coordinate point and the set radius, and saves the generated small ball structures;

[0019] A collaboration module, which is used to collaborate with different TPS systems before the treatment starts and after the first-stage treatment. Based on the high-dose small ball target area and the outer contour target area obtained by the target area delineation and high-dose small ball target area delineation module, the TPS system calculates the doses for the first-stage treatment of the irradiation of the high-dose small ball and the conventional radiotherapy of the second-stage target area in SFRT treatment.

[0020] Preferably, the multi-modal images include DICOM image files of PET / CT, MRI, and simulated CT. For multi-modal images such as DICOM image files of PET / CT, MRI, and simulated CT, the images are fused to achieve precise registration of the multi-modal images and obtain the precise fused images.

[0021] Preferably, the deformation registration module includes

[0022] An image preprocessing unit, which is used to preprocess the imported multi-modal images;

[0023] A noise removal unit, which is used to perform noise removal and contrast enhancement operations on the images processed by the image preprocessing unit;

[0024] A feature extraction unit, which is used to extract the features of the images processed by the noise removal unit and provide image feature information for subsequent registration;

[0025] A registration unit, which is used to fuse the multi-modal images processed by the noise removal unit based on the image feature information obtained by the feature extraction unit, achieve precise registration of the multi-modal images, and obtain the precise fused images.

[0026] Preferably, the preprocessing is to adjust the pixel value ranges of images of different modalities to a unified standard.

[0027] Preferably, the noise removal unit:

[0028] Uses the CLAHE algorithm of OpenCV to enhance the contrast of the images processed by the image preprocessing unit;

[0029] After converting the images from the BGR color space to the LAB space, performs contrast enhancement processing on the luminance channel and then converts back to the BGR space.

[0030] Preferably, the feature extraction unit:

[0031] Uses OpenCV for edge detection and texture feature extraction;

[0032] Perform keypoint detection and deep learning feature extraction using the SIFT, SURF algorithms or the CNN of PyTorch.

[0033] Preferably, the registration unit includes:

[0034] A coarse registration subunit, which is used to initially align the multimodal images processed by the noise removal unit using affine transformation based on the image feature information obtained by the feature extraction unit, eliminate large-scale displacements and rotations, and then perform multi-scale registration to gradually refine the registration accuracy;

[0035] A fine registration subunit, which is used to perform deformation registration on the images processed by the coarse registration subunit using the B-spline transformation of SimpleITK. Convert the reference image and the image to be registered into SimpleITK image objects, initialize the B-spline transformation, and set the relevant parameters of the image registration method, and then execute the registration process to obtain the final transformation matrix, and resample the images processed by the coarse registration subunit to achieve precise registration.

[0036] Preferably, the cooperation module includes:

[0037] A data export unit, which is used to export the outer contour target area obtained by the target area delineation and the high-dose small ball target area delineation module, as well as the target area structure of the high-dose small ball target area and the CT image together;

[0038] An interface unit, which is an interface with different TPS systems and is used for:

[0039] Send the data exported by the data export unit to the TPS system. During the sending process, automatically check the integrity and format correctness of the file to ensure accurate data transmission; receive the irradiation dose of the high-dose small ball in the first-stage treatment and the irradiation dose of the high-dose small ball in the second-stage treatment calculated by the TPS system.

[0040] Preferably, it further includes a post-treatment module, which is used to generate a dose volume histogram and an evaluation report after the end of the second-stage treatment.

[0041] Preferably, the post-treatment module includes:

[0042] A registration unit, which is used to obtain the simulated CT image 1 used in the first-stage treatment plan and the simulated CT image 2 used in the second-stage treatment plan. After importing the two sets of images, perform deformation registration using the demons algorithm and the B-spline algorithm to obtain the transformation matrix;

[0043] An EQD2 dose acquisition unit, which uses pydicom to read the physical dose file exported by the TPS system and converts the physical dose into EQD2 dose using the LQ formula;

[0044] The dose distribution acquisition unit uses SimpleITK to calculate the DVF according to the transformation matrix obtained by the registration unit, extracts the image features of the simulated CT image one and the simulated CT image two through the CNN of PyTorch, optimizes the DVF calculation to obtain more accurate DVF data, and migrates the contours and dose data of the simulated CT image one to the simulated CT image two according to the calculated DVF. Numpy is used for the superposition of dose data to calculate the superimposed dose distribution;

[0045] The report generation unit generates a dose volume histogram using matplotlib based on the dose distribution output by the dose distribution acquisition unit, and generates an evaluation report containing dose conversion and superposition results.

[0046] The system disclosed by the present invention is used to guide target area delineation through multi-modal image deformation registration technology, supports the automatic / manual design of high-dose small ball target areas, and cooperates with the treatment planning system (TPS) to complete the phased treatment process. The system disclosed by the present invention finally evaluates the overall treatment effect through deformation registration and dose superposition, and is applicable to the precise SFRT treatment planning of equipment such as accelerators, TOMO, and CyberKnife.

[0047] Compared with the prior art solutions, the technical solution disclosed by the present invention has the following characteristics:

[0048] 1) Deformation registration: Spatial segmentation treatment needs to adapt to phased anatomical changes. The traditional registration method has insufficient accuracy. The B-spline deformation model combined with deep learning can significantly improve the registration robustness;

[0049] 2) Automatic design of high-dose small balls: Reduce manual intervention through geometric constraint algorithms to ensure that the distance and volume between small balls meet clinical requirements (such as the minimum distance > 5 mm);

[0050] 3) Cooperation with TPS: The standardized DICOM interface avoids data conversion losses, improves the efficiency of different TPS plan designs, makes full use of the professionalism of different accelerator TPSs in dose calculation, and realizes complementary advantages;

[0051] 4) Prognosis evaluation closed-loop: Quantify and evaluate the comprehensive effect of multiple treatments through dose superposition and EQD2 conversion, support clinical decision-making, and provide a scientific basis for evaluating treatment prognosis.

[0052] Due to the adoption of the above technical means, the technical solution disclosed by the present invention has the following beneficial effects compared with the prior art solutions:

[0053] 1) Improve the accuracy of target delineation in volumetric modulated arc therapy (VMAT). By combining multi-modal image fusion and deformable registration techniques, more comprehensive and accurate image information is provided for doctors to delineate the target area, assisting them to more accurately determine the location and scope of the target area, reducing the target delineation errors caused by inaccurate images, and improving the accuracy of treatment;

[0054] 2) Implement the function of automatic or manual delineation of high-dose small spherical target areas, greatly shortening the delineation time of doctors, improving the delineation efficiency and consistency, reducing the workload of doctors, and at the same time reducing the influence of subjective factors on the delineation results, improving the consistency of the delineation results;

[0055] 3) Establish cooperation with other professional treatment planning system (TPS) for dose calculation, accurately transmit the structures delineated by this software system to the TPS for dose calculation, make full use of the professional dose calculation ability of the TPS, improve the accuracy and reliability of dose calculation, and provide support for formulating more reasonable treatment plans;

[0056] 4) During and after the treatment process, accurately complete the conversion from physical dose to EQD2, deformable registration, dose summation, and related parameter calculations, providing scientific and accurate data basis for evaluating the treatment prognosis, thus helping to optimize subsequent treatment plans and improve the treatment effect and quality of life of patients.

[0057] 5) Adopt modular design and standard interfaces, which are convenient for integration with other systems and expansion of new functions. The system can be continuously optimized and upgraded according to clinical needs, and has strong system compatibility and expandability. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a module diagram of a volumetric modulated arc therapy (VMAT) assistance system based on multi-modal images. In the figure, different colors distinguish different modules;

[0059] Figure 2 It is a technical flow chart of a volumetric modulated arc therapy (VMAT) assistance system based on multi-modal images. In the figure, different colors distinguish different modules. DETAILED DESCRIPTION OF THE INVENTION

[0060] The following further elaborates the present invention in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of this application.

[0061] In combination with Figure 1 and Figure 2 , a volumetric modulated arc therapy (VMAT) assistance system based on multi-modal images disclosed in the embodiments of the present invention is implemented using the following tools:

[0062] 1. Core framework: Python integrates SimpleITK (registration), PyTorch (deep learning optimization), and pydicom (DICOM interaction).

[0063] 2. Visualization: Matplotlib / VTK is used to achieve interactive display of 3D images and target regions.

[0064] 3. Interface design: The standardized DICOM RT interface is seamlessly docked with the TPS system (such as CyberKnife).

[0065] The system implemented based on the above tools specifically includes the following modules:

[0066] (1) Deformable registration module for multimodal images

[0067] For multimodal images such as DICOM image files of PET / CT, MRI, and simulated CT before the start of treatment and after the first-stage treatment, the images are fused to achieve precise registration of multimodal images and obtain precise fused images.

[0068] In the embodiment of the present invention, the deformable registration module specifically includes the following units:

[0069] Image preprocessing unit, which is used to preprocess the imported multimodal images. In the embodiment of the present invention, the image preprocessing unit uses SimpleITK to standardize the imported images and eliminate the differences brought by different devices and imaging conditions. Taking a PET / CT image as an example, the image preprocessing unit uses SimpleITK to adjust the pixel value range of the image to a unified standard and eliminate the imaging differences of different devices.

[0070] Noise removal unit, which is used to perform noise removal and contrast enhancement operations on the images processed by the image preprocessing unit through OpenCV to improve the image quality. A preferred implementation is that the noise removal unit first filters the images processed by the image preprocessing unit through the fastNlMeansDenoisingColored function of OpenCV to remove the noise points in the images and make the images clearer. Then, the noise removal unit uses the CLAHE algorithm of OpenCV for contrast enhancement, converts the image from the BGR color space to the LAB space, performs contrast enhancement processing on the brightness channel and then converts it back to the BGR space to highlight the detail information in the images.

[0071] The feature extraction unit is used to extract features from the image processed by the noise removal unit using multiple feature extraction methods, providing rich image feature information for subsequent registration. A preferred implementation is that the feature extraction unit can use OpenCV for edge detection and texture feature extraction, or use algorithms such as SIFT, SURF, or CNN of PyTorch for key point detection and deep learning feature extraction. For example, in the embodiment of the present invention, the feature extraction unit uses the Canny algorithm of OpenCV for edge detection to extract the edge features of the image. At the same time, the feature extraction unit extracts texture features by using the Laplacian operator after converting the image to a grayscale image, and the feature extraction unit uses the SIFT algorithm to detect key points, obtaining key points and descriptors, providing feature point pairs for the registration unit.

[0072] The registration unit is used to fuse the multi-modal images processed by the noise removal unit, achieve precise registration of the multi-modal images, and obtain the precise fused image. A preferred implementation is that the registration unit adopts a combination of coarse registration and fine registration to achieve precise registration of the multi-modal images, and further includes:

[0073] The coarse registration sub-unit is used to initially align the multi-modal images processed by the noise removal unit using affine transformation, eliminate large-scale displacements and rotations, and then perform multi-scale registration to gradually refine the registration accuracy. In the embodiment of the present invention, the coarse registration sub-unit first uses the BFMatcher matcher to match the feature points according to the key points and descriptors detected by the feature extraction unit, filters out good matching point pairs, and calculates the affine transformation matrix. Subsequently, the coarse registration sub-unit uses the affine transformation matrix to perform affine transformation on the image to be registered to achieve initial alignment. Then, the coarse registration sub-unit performs multi-scale registration, scales the reference image and the initially aligned image at different scales, such as 0.5, 0.75, 1.0 times scaling, and repeats the processes of key point detection, matching, and calculation of the transformation matrix at different scales to gradually refine the registration accuracy.

[0074] A fine registration subunit is used to perform deformation registration by using methods such as B-spline and free-form deformation, and optimize the registration parameters using gradient descent, Powell algorithm, etc., to handle local non-rigid deformation and achieve precise registration of multi-modal images. In the embodiments of the present invention, the fine registration subunit uses the B-spline transformation of SimpleITK to perform deformation registration on the image processed by the coarse registration subunit. The fine registration subunit converts the reference image and the image to be registered into SimpleITK image objects, initializes the B-spline transformation, sets the relevant parameters of the image registration method, such as the metric criterion is the mean square error, the optimizer is the gradient descent method, sets parameters such as the learning rate and the number of iterations, executes the registration process, obtains the final transformation matrix, and resamples the image processed by the coarse registration subunit to achieve precise registration.

[0075] (2) Target volume delineation and high-dose small sphere target volume delineation module

[0076] It is used to complete the delineation of the outer contour target volume by manual delineation method on the fused precise image output by the deformation registration module before the treatment starts and after the first-stage treatment, and complete the delineation of the high-dose small sphere target volume by manual delineation method or automatic delineation method.

[0077] When performing the delineation of the outer contour target volume by manual delineation method, the doctor uses the delineation tool provided by the target volume delineation and high-dose small sphere target volume delineation module to delineate the target volume, and the target volume delineation and high-dose small sphere target volume delineation module provides auxiliary reference lines and marks according to the fused precise image to help the doctor more accurately determine the target volume boundary and complete the delineation of the outer contour target volume.

[0078] When performing the delineation of the high-dose small sphere target volume by manual delineation method:

[0079] The doctor uses the target volume delineation and high-dose small sphere target volume delineation module to set the small sphere radius interface and input the desired radius value (for example, 5 mm), and the target volume delineation and high-dose small sphere target volume delineation module automatically displays the corresponding setting input box on the interface.

[0080] The doctor uses the target volume delineation and high-dose small sphere target volume delineation module to set the small sphere coordinates and input the x, y, z coordinate values of the current small sphere. The target volume delineation and high-dose small sphere target volume delineation module generates a small sphere structure using SimpleITK according to the input radius value and coordinates. Assuming that the current small sphere is the first small sphere, the target volume delineation and high-dose small sphere target volume delineation module automatically names the generated small sphere structure as "small sphere 1", and so on, which will not be elaborated here.

[0081] The target region delineation and high-dose small sphere target region delineation module uses numpy to calculate the distance between the currently generated small sphere and other already generated small spheres, and determines whether it meets the pre-set geometric requirements (for example, the minimum distance is 10 mm, the maximum distance is 30 mm, etc.): If it does not meet the requirements, the doctor modifies the parameters such as the radius and coordinates of the current small sphere through the target region delineation and high-dose small sphere target region delineation module. The target region delineation and high-dose small sphere target region delineation module regenerates the small sphere structure based on the updated parameters and makes the judgment again until it meets the requirements.

[0082] The target region delineation and high-dose small sphere target region delineation module saves the small spheres that meet the requirements into the DICOM STRUCTURE file. For "Small Sphere 1", the save path is "D:\SFRT\structures\Small Sphere 1.dcm", and so on, which will not be elaborated here.

[0083] When using the automatic delineation method for high-dose small sphere target region delineation:

[0084] The doctor inputs the small sphere radius and the number of generated small spheres (for example, the small sphere radius is 4 mm and the number is 10) into the target region delineation and high-dose small sphere target region delineation module.

[0085] The target region delineation and high-dose small sphere target region delineation module uses numpy to randomly generate 10 coordinate points within the delineation region, and uses numpy to determine whether the geometric requirements (such as the minimum distance between coordinate points is 10 mm, the maximum distance is 30 mm, etc.) are met between these coordinate points: If not, the target region delineation and high-dose small sphere target region delineation module regenerates the coordinate points until the generated coordinate points meet the requirements.

[0086] For the coordinate points that meet the requirements, the target region delineation and high-dose small sphere target region delineation module uses SimpleITK to generate small sphere structures according to each coordinate point and the set radius, and names them "Small Sphere 1" to "Small Sphere 10" in sequence. Finally, these small spheres are saved into the DICOM STRUCTURE file, and the save path is under the folder "D:\SFRT\structures\Automatically delineated small spheres".

[0087] (3) Collaboration module

[0088] It is used to collaborate with different TPS systems before the treatment starts and after the first stage of treatment, and the TPS system calculates the doses for the first stage of treatment and the second stage of treatment. In the present invention, before the treatment starts and after the first stage of treatment, the professional dose calculation algorithm inside the existing TPS system is used to calculate the doses for the first stage of treatment of the irradiation of high-dose small spheres and the second stage of conventional radiotherapy in SFRT treatment, giving full play to the advantages of the existing TPS system in dose calculation.

[0089] A preferred embodiment is that the collaboration module further includes:

[0090] A data export unit, configured to export the outer contour target area obtained by the target area delineation and high-dose small ball target area delineation module, as well as the target area structure and CT images of the high-dose small ball target area in DICOM format together.

[0091] An interface unit, which is an interface with different TPS systems, and is used for: sending the data exported by the data export unit to the TPS system (such as the TPS supporting the CyberKnife system), and automatically checking the integrity and format correctness of the file during the sending process to ensure accurate data transmission; receiving the irradiation doses of the target area and high-dose small ball target area in the first-stage treatment and the irradiation doses of the target area and high-dose small ball target area in the second-stage conventional radiotherapy calculated by the TPS system.

[0092] (IV) Post-treatment processing module

[0093] It is used to generate a dose volume histogram and an evaluation report after the end of the second-stage treatment. Doctors can analyze the treatment effect based on the evaluation report, providing a basis for adjusting the subsequent further treatment plan.

[0094] A preferred embodiment is that the post-treatment processing module further includes:

[0095] A registration unit, configured to obtain the first simulation CT image used in the first-stage treatment plan and the second simulation CT image used in the second-stage treatment plan. After importing the two sets of images, the demons algorithm and B-spline algorithm are used for deformation registration to obtain a transformation matrix. In the implementation of the present invention, the registration unit first converts the images into SimpleITK image objects, then initializes the B-spline transformation, sets the relevant parameters of the registration method, and then executes the registration process to obtain the final transformation matrix.

[0096] An EQD2 dose acquisition unit, which reads the physical dose file exported by the TPS system using pydicom, converts the physical dose into EQD2 dose using the LQ formula, and stores the converted EQD2 dose in a new array.

[0097] A dose distribution acquisition unit, which calculates the DVF using SimpleITK according to the transformation matrix obtained by the registration unit, extracts the image features of the first simulation CT image and the second simulation CT image through the CNN of PyTorch, optimizes the DVF calculation, and obtains more accurate DVF data. According to the calculated DVF, the contours and dose data in the first simulation CT image are migrated to the second simulation CT image, and the dose data is superimposed using numpy to calculate the superimposed dose distribution.

[0098] The report generation unit generates a dose volume histogram (DVH graph) using matplotlib based on the dose distribution output by the dose distribution acquisition unit to visually display the dose distribution. Meanwhile, the report generation unit generates an evaluation report containing dose conversion and superposition results. In the embodiments of the present invention, the report content includes information such as total physical dose, total EQD2 dose, PVDR value, etc., and saves the report as a file named "D:\SFRT\reports\Treatment Evaluation Report.pdf" to provide data support for evaluating the treatment prognosis.

[0099] The implementation method of the above-mentioned multi-modal image-based spatial segmentation radiotherapy assistance system may include the following steps:

[0100] Step 1. Pre-treatment preparation stage: The doctor imports the DICOM files of multi-modal images such as PET / CT, MRI, and simulation CT into the above system. Select the multi-modal image fusion function, and the system automatically preprocesses the images, extracts features, performs rough registration and fine registration to generate the fused multi-modal image. On the fused image, the doctor can select manual, automatic, or deep learning methods to outline the target area, and save the target area structure after completion.

[0101] Step 2. Dose calculation and radiotherapy implementation in the first stage of treatment: Send the outlined target area structure to the corresponding TPS system through the interface between the system and the TPS system for dose calculation. The doctor waits for the TPS system to return the dose calculation result and views and confirms it in the system. According to the treatment plan generated by the TPS system, use the corresponding treatment equipment to perform SFRT treatment on the patient, and first complete the irradiation of the high-dose small ball.

[0102] Step 3. Dose calculation and radiotherapy implementation in the second stage of treatment: Rescan the patient's CT image, and repeat the above process to send the outlined target area structure to the corresponding TPS system through the interface between the software system and the TPS system for dose calculation. The doctor waits for the TPS system to return the dose calculation result and views and confirms it in the software system. According to the treatment plan generated by the TPS system, use the corresponding treatment equipment to perform conventional radiotherapy irradiation on the patient.

[0103] Step 4. Post-treatment evaluation stage: After the irradiation is completed, the system performs deformation registration, contour migration, dose superposition, conversion from physical dose to EQD2, and calculation of relevant parameters on the two simulation CT images to generate a DVH graph and an evaluation report. The doctor analyzes the treatment effect based on the evaluation report to provide a basis for adjusting the subsequent further treatment plan.

Claims

1. A spatial segmentation radiotherapy auxiliary system based on multimodal images, characterized in that: include: The multimodal image deformation registration module is used to import multimodal images before the start of treatment and after the first stage of treatment, fuse images of different modalities, achieve accurate registration of multimodal images, and obtain accurate fused images; The target area delineation and high-dose ball target area delineation modules are used to manually or automatically delineate the high-dose ball target area on the fused precise image output by the deformable registration module before the start of treatment and after the first stage of treatment, and to manually delineate the outer contour target area, including: When the outer contour target area is delineated manually, the doctor uses the delineation tools provided by the target area delineation and high-dose microsphere target area delineation modules to delineate the target area. The target area delineation and high-dose microsphere target area delineation modules provide auxiliary reference lines and marks based on the fused precise images to help the doctor determine the target area boundary more accurately and complete the outer contour target area delineation. When manual delineation is used to delineate the high-dose microsphere target area: Set the desired ball radius value; Set the ball coordinates; Target delineation and high-dose ball target delineation module generates ball structure based on the set radius value and ball coordinates; The target area delineation and high-dose ball target area delineation module calculates the distance between the currently generated ball and other generated balls, and determines whether it meets the preset geometric requirements. If it does not meet the requirements, the expected ball radius value and ball coordinates are reset, and the target area delineation and high-dose ball target area delineation module regenerates the ball structure based on the updated parameters and makes a judgment again until it meets the requirements; Save the ball structure that meets the requirements; When automatic delineation is used to delineate the high-dose spherical target area: Set the ball radius and the number of balls generated; The target area delineation and high-dose ball target area delineation module randomly generates 10 coordinate points in the delineated area and determines whether the coordinate points meet the pre-set geometric requirements. If not, the coordinate points are regenerated until the generated coordinate points meet the requirements. For coordinate points that meet the requirements, the target area delineation and high-dose spherical target area delineation modules generate spherical structures according to each coordinate point and the set radius, and save the generated spherical structures; The collaboration module is used to collaborate with different TPS systems before the start of treatment and after the first stage of treatment. The TPS system calculates the dose of the first stage of treatment and the conventional radiotherapy of the second stage target area for the irradiation of high-dose spheres in SFRT treatment based on the high-dose sphere target area and the outer contour target area obtained by the target area delineation and the high-dose sphere target area delineation module.

2. The spatial segmentation radiotherapy auxiliary system based on multimodal images as claimed in claim 1, characterized in that: The multimodal images include DICOM image files of PET / CT, MRI and simulated CT. Multimodal images such as DICOM image files of PET / CT, MRI and simulated CT are fused to achieve accurate registration of the multimodal images and obtain accurate fused images.

3. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 1, characterized in that: The deformation registration module includes An image preprocessing unit, used for preprocessing the imported multimodal images; A noise removal unit, used for performing noise removal and contrast enhancement operations on the image processed by the image preprocessing unit; A feature extraction unit, used to extract features of the image processed by the noise removal unit, and provide image feature information for subsequent registration; The registration unit is used to fuse the multimodal images processed by the noise removal unit based on the image feature information obtained by the feature extraction unit, so as to achieve accurate registration of the multimodal images and obtain an accurate fused image.

4. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 3, characterized in that: The preprocessing is to adjust the pixel value ranges of images of different modalities to a unified standard.

5. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 3, characterized in that: The noise removal unit: Use OpenCV's CLAHE algorithm to enhance the contrast of the image processed by the image preprocessing unit; After converting the image from BGR color space to LAB space, the brightness channel is contrast enhanced and then converted back to BGR space.

6. The spatial segmentation radiotherapy auxiliary system based on multimodal images as claimed in claim 3, characterized in that: The feature extraction unit: Use OpenCV for edge detection and texture feature extraction; Use SIFT, SURF algorithms or PyTorch's CNN for keypoint detection and deep learning feature extraction.

7. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 3, characterized in that: The registration unit comprises: The coarse registration subunit is used to initially align the multimodal images processed by the noise removal unit using affine transformation based on the image feature information obtained by the feature extraction unit, eliminate large-scale displacement and rotation, and then perform multi-scale registration to gradually refine the registration accuracy; The fine registration subunit is used to use SimpleITK's B-spline transformation to perform deformation registration on the image processed by the coarse registration subunit. It converts the reference image and the image to be registered into SimpleITK image objects, initializes the B-spline transformation, sets the relevant parameters of the image registration method, executes the registration process, obtains the final transformation matrix, and resamples the image processed by the coarse registration subunit to achieve accurate registration.

8. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 1, characterized in that: The collaboration module includes: A data export unit, used for exporting the target area delineation and the target area structure of the high-dose ball target area obtained by the high-dose ball target area delineation module together with the CT image; The interface unit is an interface with different TPS systems and is used to: The data exported by the data export unit is sent to the TPS system. During the sending process, the integrity and format correctness of the file are automatically checked to ensure accurate data transmission; the irradiation dose of the high-dose beads in the first stage of treatment and the irradiation dose of the high-dose beads in the second stage of treatment calculated by the TPS system are received.

9. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 1, characterized in that: It also includes a post-treatment processing module for generating a dose volume histogram and an evaluation report after the second stage of treatment.

10. The multimodal image-based spatial segmentation radiotherapy auxiliary system according to claim 1, characterized in that: The post-treatment processing module comprises: A registration unit is used to obtain a simulated CT image 1 used for the first-stage treatment plan and a simulated CT image 2 used for the second-stage treatment plan. After importing the two sets of images, the demons algorithm and the B-spline algorithm are used to perform deformation registration to obtain a transformation matrix. The EQD2 dose acquisition unit uses pydicom to read the physical dose file exported by the TPS system and converts the physical dose into EQD2 dose using the LQ formula; The dose distribution acquisition unit uses SimpleITK to calculate the DVF according to the transformation matrix obtained by the registration unit, and uses PyTorch's CNN to extract the image features of the simulated CT image 1 and the simulated CT image 2, optimizes the DVF calculation, and obtains more accurate DVF data. According to the calculated DVF, the contour and dose data of the simulated CT image 1 are transferred to the simulated CT image 2, and numpy is used to superimpose the dose data to calculate the dose distribution after superposition; The report generation unit generates a dose volume histogram using matplotlib based on the dose distribution output by the dose distribution acquisition unit, and generates an evaluation report containing dose conversion and overlay results.

Images