Radionuclide therapeutic dose calculation system

Abstract

The invention relates to the technical field of ray therapy, and discloses a radionuclide therapeutic dose calculation system, which is used for importing a quantitative PET / CT (positron emission tomography / computed tomography) image of a patient before radionuclide therapy and a quantitative SPECT / CT image acquired after the therapy. According to the quantitative SPECT image, the radionuclide type, the drug name, the image acquisition time, the height, the weight and the radionuclide distribution condition in the body of a patient are obtained, and the DICOM analysis module reads information contained in a quantitative SPECT image tag. According to the method, the individualization of the dose calculation phantom is realized; the individualized computer phantom of the patient can be quickly generated according to the CT image of the patient; and the anatomical characteristics of the patient are accurately described to the greatest extent. The method does not depend on any standard reference body model, solves the problem that the specificity of the body shape and the anatomical structure of the patient is difficult to consider by adopting the standard reference body model in the prior art, and improves the accuracy of dose calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of radiation therapy, in particular to a radionuclide therapy dose calculation system. Background Art

[0002] Radionuclide therapy is a long-standing and promising treatment for various cancers. Currently proposed targeted radionuclide therapy (TRT) is capable of selectively delivering high radiation doses to target sites while minimizing toxicity to normal tissues. Currently, the most established application of TRT is the palliative treatment of patients with unresectable metastatic neuroendocrine tumors (NETs), which has significantly improved overall and progression-free survival. In current clinical practice, the standard approach for TRT treatment is to use a one-size-fits-all empirical regimen, administering the same dose and number of cycles to all patients, regardless of height, weight, or tumor burden. However, this regimen may not be optimally tailored to individual differences in tumor burden, patient physiology, body size, and overall health. Therefore, personalized dosimetry and dosimetry-based treatment planning approaches are essential to ensure a balance between treatment efficacy and patient safety.

[0003] Calculating a patient's internal absorbed dose from radionuclides involves both the physical properties of the radionuclide and its distribution within the patient's body. This distribution is closely related to the patient's anatomy. First, the MIRD-based method for calculating the average organ absorbed dose is based on a standard human model. However, different patients vary in race, gender, age, body shape, and organ location, volume, and mass.

[0004] Second, this method only provides the average absorbed dose to an organ or tissue. However, due to the uneven distribution of radionuclides in the human body and the influence of organ anatomy, the dose is not uniformly distributed within the organ or tissue volume, making this method difficult to reflect the dose distribution within the organ. Furthermore, this method assumes that the radionuclides are uniformly distributed within the source organ, which is inconsistent with the actual situation and may lead to errors.

[0005] Then, the VSV method requires the S value of each radioactive isotope to be pre-indexed with the corresponding voxel size, and VSV is mostly based on a uniform water medium for dose calculation, which has large errors for tissues with large differences in density and material composition, such as lungs and bones.

[0006] Finally, the radioactivity distribution data of this method comes from planar SPECT imaging, and there is overlap between different tissues, making it difficult to accurately quantify the activity distribution.

[0007] To this end, the present invention provides a radionuclide therapy dose calculation system. Summary of the Invention

[0008] In view of the above-mentioned deficiencies in the prior art, the present invention provides a radionuclide therapy dose calculation system that can quickly and accurately calculate the individualized dose of radionuclide therapy for patients based on quantitative SPECT / CT images and the Monte Carlo method.

[0009] The present invention provides the following technical solution: a radionuclide therapy dose calculation system, comprising the following steps:

[0010] S1. Import the patient's quantitative PET / CT images taken before radionuclide therapy and the quantitative SPECT / CT images collected after therapy. The radionuclide type, drug name, image acquisition time, height, weight, and radionuclide distribution in the patient's body are obtained from the quantitative SPECT images. The DICOM parsing module reads the information contained in the quantitative SPECT image tags and determines the nuclide decay energy spectrum used in the subsequent MC simulation based on the type of radionuclide used.

[0011] S2. Use segmentation tools to segment the patient's outer contours, lesions, and organs at risk (OARs) on PET / CT and SPECT / CT images to obtain organ contour masks. Each image is input into the segmentation model tool. The tool provides preliminary segmentation results of lesions and OARs based on the deep network. The doctor reviews and modifies the results and generates organ contour masks based on the segmentation results.

[0012] S3. Determine the physiological pharmacokinetic model of the drug based on the obtained therapeutic drug name, determine the corresponding population physiological pharmacokinetic model based on the drug name and the patient's age and gender, and confirm the patient's body shape and tumor burden based on the patient's pre-treatment PET / CT and pre-cycle SPECT / CT images. Perform fitting and optimization based on the population physiological pharmacokinetic model to obtain the patient's individualized physiological pharmacokinetic model;

[0013] S4. Register the SPECT / CT images acquired at different time points after treatment to the reference image. Select an appropriate reference image for registration based on the metabolic characteristics of the drug used in the treatment.

[0014] CT-based SPECT image registration: resample the SPECT image to keep its resolution and size consistent with CT, select a registration algorithm, use global image information as a criterion to register the non-reference CT image to the reference image, and save the deformation field. Apply the obtained deformation field to the non-reference SPECT image to register the non-reference SPECT image to the reference SPECT image.

[0015] Contour-based registration: resample the SPECT image to match the resolution and size of the CT image; select a registration algorithm, using the lesion and OAR contour information obtained in step 2 as the criterion, register the non-reference CT image to the reference image and save the deformation field; apply the obtained deformation field to the non-reference SPECT image to register the non-reference SPECT image to the reference SPECT image;

[0016] SPECT image-based registration: Using the acquired lesion and OAR contour information as the measurement standard, the non-reference SPECT image is directly registered to the reference SPECT image, and the SPECT is resampled to keep its resolution and size consistent with CT;

[0017] S5. Extract the radioactivity information corresponding to each acquisition time for each voxel of the multiple sets of registered SPECT images. Fit the radionuclide variation curve of the voxel with time based on the acquisition time and activity information. Calculate the area under the curve, i.e., the voxel time-integrated activity (TIA), to obtain the distribution of radioactive decay in the patient. Extract the radioactivity information of each voxel in each organ based on the organ contour mask. Automatically select a fitting function model based on drug information, organ metabolism pattern, individualized physiological pharmacokinetic model, and voxel radioactivity variation. Automatically select the appropriate fitting parameter value range. Process voxels with poor fitting. Integrate the TIC curves of all voxels to obtain the voxel cumulative radioactive decay count. Store the TIC curves in Bq / ml in a time-integrated activity image (TIA) with a resolution and size consistent with SPECT.

[0018] S6. Generating a density distribution of each voxel of the patient's personalized computer phantom based on the reference CT image and its corresponding CT value-to-density conversion relationship; determining the material composition of each voxel of the patient's personalized computer phantom based on the organ segmentation result; converting the patient's CT image into a density map based on the CT value-to-density conversion relationship of the image acquisition device; determining the voxel material composition based on the organ contour mask pair; and generating the patient's personalized computer phantom;

[0019] S7. Perform GPU-accelerated Monte Carlo simulation of the absorbed dose calculation model based on the decay distribution and the patient's individualized computer phantom. Calculate the initial geometric position probability distribution of the decay particles in the Monte Carlo simulation using the acquired TIA images. Generate decay secondary particles in batches based on the type of radionuclide and decay spectrum sampling. Distribute the secondary particles to GPU threads for simultaneous simulation calculations.

[0020] S8. Calculate the absorbed dose parameters such as the dose-volume histogram, mean absorbed dose, and equivalent uniform dose (EUD) of each tissue and OAR based on the obtained absorbed dose results and the obtained patient organ contour mask, and calculate the BED according to the principles of radiobiology.

[0021] Preferably, the voxel time-integrated activity calculation process in S5 is as follows: the uptake and excretion of radioactive drugs by human tissues are usually described by exponential decay models, including but not limited to single exponential models and double exponential models. Taking the single exponential model as an example, the relationship between the radioactive activity A(t) and the scanning time t, the initial activity A0, and the effective decay constant λ of the radioactive drug in the voxel is described as follows: A(t) = A0 × e (-λt) ;

[0022] By fitting, we can obtain the fixed parameters such as A0 and λ, and calculate the integral of A(t) from t0 to the end time of the cycle tc (generally > 1000h), which is the cumulative radioactivity of the voxel. The formula is as follows:

[0023]

[0024] For the case of a specified reference image, the reference image activity distribution Ar and the fitted λ are used to represent A(t) and perform integral calculations.

[0025] Preferably, the absorbed dose calculation model in S7 transmits particles generated by the radionuclide decay module, records the interaction between the particles and the voxels during the transmission process, and records the deposited energy E and voxel index generated in the interaction. After the transmission is completed, the deposited energy in the voxel is the sum of the deposited energies generated by all interactions. The corresponding density of the voxel is determined according to the voxel index, and the voxel mass is determined according to the voxel size and density. The voxel absorbed dose is the ratio of the voxel deposited energy to the voxel mass. The absorbed doses of all voxels are summarized as the patient absorbed dose map.

[0026] Preferably, the BED dose calculation formula in S8 is as follows:

[0027]

[0028] Where D is the absorbed dose corresponding to the voxel, λ is the effective decay constant of the voxel, and α / β is a classic concept in radiobiology, derived from the LQ model. It reflects the intrinsic sensitivity of tumors or organs to dose fractionation (fractionated irradiation) or dose rate (continuous low-dose irradiation), which is determined by the biological characteristics of the tissue corresponding to the voxel.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] (1) Individualized dose calculation phantoms are achieved: The present invention can quickly generate a patient-specific computer phantom based on the patient's CT images, accurately describing the patient's anatomical characteristics to the greatest extent possible. This solves the problem of previous technical methods using standard reference phantoms that made it difficult to consider the specificity of the patient's body shape and anatomical structure, thereby improving the accuracy of dose calculations.

[0031] (2) Reduced reliance on pre-existing databases for radionuclide therapy dose calculations: The dose calculation of the present invention does not rely on pre-existing dose calculation data that is not based on patient data, except for standard data such as radionuclide spectra. It has flexible processing capabilities for various conditions such as patient race, gender, body shape, condition, and image resolution and size.

[0032] (3) Improving the accuracy of calculations in media with uneven density and material composition distribution: The present invention uses the MC method for dose calculation, which can truly simulate radiation transmission in uneven media, rather than performing analytical corrections based on pre-calculated results, thereby reducing calculation errors in uneven media and tissue boundaries;

[0033] (4) Improve the efficiency of MC dose calculation: As a statistical method, the MC method needs to calculate tens of millions or even more particles to reduce statistical uncertainty, so as to obtain accurate calculation results. The traditional MC method uses the CPU for one-way calculation, and the time cost required to complete the corresponding calculation task is huge. Even if a cluster CPU is used for calculation, it still takes many hours or even days to complete. The present invention adopts a fast Monte Carlo simulation dose calculation based on GPU, which can complete a patient dose calculation in about 1 minute, greatly reducing the calculation time of the method, improving the efficiency of dose calculation, and making it possible to apply the MC method to nuclear medicine clinical practice;

[0034] (5) Reducing the hardware cost of MC dose calculation: As mentioned above, the traditional MC method can improve computational efficiency by increasing the number of CPUs. However, the hardware procurement and maintenance costs of clustered CPUs are huge, making it unsuitable for promotion in hospitals. The GPU-based fast MC dose calculation used in the present invention uses GPU parallel computing, requiring only a personal host equipped with a GPU to complete the high-efficiency calculation, greatly reducing hardware costs and facilitating its promotion.

[0035] (6) Improve the accuracy of patient radioactivity distribution data: Using patient quantitative SPECT / CT for dose calculation reduces the activity quantification error caused by organ overlap in planar scan images and fully considers the uneven distribution of radioactivity in tissues;

[0036] (7) Reduce toxic and side effects: Provide accurate voxel-level dose calculation results for TRT to evaluate multiple dose parameters such as the average dose of OARs such as the kidney, and provide a basis for preventing and treating toxic and side effects during and after TRT treatment;

[0037] (8) Improve treatment effect: Provide accurate voxel-level dose calculation results for TRT to evaluate multiple dose parameters such as the average dose of the patient's lesion and other treatment areas, accurately measure the dose coverage of each treatment area, and conduct timely medical evaluation and intervention for lesions with insufficient dose, thereby improving patient treatment effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic diagram of the method of the present invention;

[0039] Figure 2 This is a schematic diagram of the 3DResU-Net network structure of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. In order to keep the following description of the embodiments of the present disclosure clear and concise, the present disclosure omits detailed descriptions of known functions and known components to avoid unnecessary confusion of the concepts of the present invention.

[0041] See also Figure 1 , a radionuclide therapy dose calculation system, comprising the following steps:

[0042] S1. Import the quantitative PET / CT images of the patient before radionuclide treatment and the quantitative SPECT / CT images collected after treatment. According to the quantitative SPECT images, obtain the type of radionuclide, drug name, image acquisition time, height, weight and distribution of radionuclide in the patient's body. The DICOM parsing module reads the information contained in the quantitative SPECT image tag and determines the nuclide decay energy spectrum used in the subsequent MC simulation based on the type of radionuclide used.

[0043] S2. Use segmentation tools to segment the patient's outer contours, lesions, and organs at risk (OARs) on PET / CT and SPECT / CT images to obtain organ contour masks. Input each image into the segmentation model tool. The tool gives preliminary segmentation results of lesions and OARs based on the deep network. The doctor reviews and modifies them and generates organ contour masks based on the segmentation results.

[0044] The segmentation model uses a pre-trained network structure including but not limited to 3DResU-Net, such as Figure 2, input the patient's SPECT image, and output the organ contour and the mask of the lesion. Taking 3DResU-Net as an example, the main differences compared with the original 3DU-Net[2] include: using a residual module in the encoder, downsampling the image twice in the axial direction, using deep supervision for supervised learning at multiple scales, replacing the pooling layer with a convolution layer in the downsampling, replacing batch normalization with instance normalization in the normalization layer, and replacing ReLU with LReLU in the activation layer.

[0045] The cross entropy loss function used is:

[0046] The Dice loss function formula is as follows:

[0047]

[0048] Total loss function: L = 0.5 × L ce +0.5×L dice ;

[0049] Among them, p i,k is the predicted probability that the i-th voxel in the sample belongs to the k-th class, y i,k is the true label of the i-th voxel (1 if it belongs to the k-th class, otherwise 0), N is the total number of voxels in the sample, K is the number of classes, that is, the number of output channels of the neural network, and ε is a minimum value to prevent the denominator from being 0, which is set to 1.

[0050] S3. Determine the physiological pharmacokinetic model of the drug based on the obtained therapeutic drug name, determine the corresponding population physiological pharmacokinetic model based on the drug name and the patient's age and gender, and confirm the patient's body shape and tumor load based on the patient's pre-treatment PET / CT and pre-cycle SPECT / CT images to obtain the patient's individualized physiological pharmacokinetic model through fitting and optimization based on the population physiological pharmacokinetic model.

[0051] S4. Register the SPECT / CT images acquired at different time points after treatment to the reference image. Select an appropriate reference image for registration based on the metabolic characteristics of the drug used in the treatment.

[0052] Two images are involved in the registration process. One image, the moving image IM(x), is deformed to fit the other image, the fixed image IF(x). Registration is the problem of finding a transformation T(x) = x + u(x) that spatially aligns IM(T(x)) with IF(x). This transformation is defined as a mapping from the fixed image to the moving image, T: The quality of the registration is defined by a distance or similarity metric S such as the sum of squared differences (SSD), correlation ratio, or mutual information (MI) metric.

[0053] The image registration of the present invention includes but is not limited to the following three methods.

[0054] CT-based SPECT image registration: The reference SPECT / CT image is used as the fixed image, and the other SPECT / CT images are used as the moving images. The moving image is transformed using a B-spline transformation model. The similarity between the transformed image and the fixed image is calculated using similarity metrics such as the sum of squared differences (SSD), correlation ratio, or mutual information (MI). The similarity metric function is optimized, and the B-spline transformation parameters are determined iteratively using methods such as gradient descent and Newton's method to optimize the similarity metric function. The corresponding B-spline deformation field is applied to the corresponding SPECT image of the moving CT image to obtain the SPECT image registration result.

[0055] Contour-based registration: The reference SPECT / CT image is used as the fixed image, and the other SPECT / CT images are used as the moving images. The moving images are transformed using a B-spline transformation model. In addition to similarity metrics such as the sum of squared differences (SSD), correlation ratio, or mutual information (MI), the similarity metric function incorporates information based on lesion and OAR contours. By optimizing the similarity metric function, the B-spline transformation parameters are determined using iterative optimization methods such as gradient descent and Newton's method to achieve the optimal similarity metric function. The corresponding B-spline deformation field is applied to the corresponding SPECT image of the moving CT image to obtain the SPECT image registration result.

[0056] SPECT image-based registration: The reference SPECT / CT image is used as the fixed image, and the other SPECT / CT images are used as the moving images. The moving images are transformed using a B-spline transformation model. In addition to similarity metrics such as the sum of squared differences (SSD), mean squared error (MSE) correlation ratio, or mutual information (MI), the similarity metric function is further integrated with lesion and OAR contour information. By optimizing the similarity metric function, the B-spline transformation parameters are determined using iterative optimization methods such as gradient descent and Newton's method to achieve the optimal similarity metric function. The corresponding B-spline deformation field is applied to the moving SPECT image to obtain the SPECT image registration result.

[0057] S5. For each voxel of the multiple sets of registered SPECT images, the radioactivity information corresponding to each acquisition time is extracted. The radionuclide change curve of the voxel is fitted according to the acquisition time and activity information, and the area under the curve, i.e., the voxel time-integrated activity (TIA), is calculated, thereby obtaining the distribution of radioactive decay in the patient's body. The radioactivity information of each voxel of each organ is extracted according to the organ contour mask. The fitting function model is automatically selected according to the drug information, organ metabolism pattern, individualized physiological pharmacokinetic model and voxel radioactivity change. The appropriate fitting parameter value range is automatically selected, and the voxels with poor fitting are processed. The TIC curves of all voxels are integrated to obtain the voxel cumulative radioactive decay number, which is stored in Bq / ml in the time-integrated activity image TIA with the same resolution and size as SPECT.

[0058] The calculation process of voxel time-integrated activity is as follows: The uptake and excretion of radioactive drugs by human tissues are usually described by exponential decay models, including but not limited to single exponential models and double exponential models. Taking the single exponential model as an example, the relationship between the radioactivity A(t) and the scanning time t, the initial activity A0, and the effective decay constant λ of the radioactive drug in the voxel is described as: A(t) = A0 × e (-λt) ;

[0059] By fitting, we can obtain the fixed parameters such as A0 and λ, and calculate the integral of A(t) from t0 to the end time of the cycle tc (generally > 1000h), which is the cumulative radioactivity of the voxel. The formula is as follows:

[0060]

[0061] For the case of a specified reference image, the reference image activity distribution Ar and the fitted λ are used to represent A(t) and perform integral calculations.

[0062] S6. Generate a density distribution for each voxel of the patient's personalized computer phantom based on the reference CT image and its corresponding CT value-to-density conversion relationship. Determine the material composition of each voxel of the patient's personalized computer phantom based on the organ segmentation results. Convert the patient's CT image into a density map based on the CT value-to-density conversion relationship of the image acquisition device. Determine the voxel material composition based on the organ contour mask pair to generate the patient's personalized computer phantom.

[0063] S7. Based on the decay distribution and the patient's individualized computer phantom, a GPU-accelerated Monte Carlo simulation of the absorbed dose calculation model is performed. The initial geometric position probability distribution of the decay particles in the Monte Carlo simulation is calculated through the acquired TIA images. Decay secondary particles are generated in batches based on the type of radionuclide and the decay spectrum sampling. The secondary particles are distributed to the GPU threads for simultaneous simulation calculations.

[0064] The absorbed dose calculation model transmits particles generated by the radionuclide decay module, records the interaction between the particles and the voxels during the transmission process, and records the deposited energy E and voxel index generated in the interaction. After the transmission is completed, the deposited energy in the voxel is the sum of the deposited energies generated by all interactions. The corresponding density of the voxel is determined according to the voxel index, and the voxel mass is determined according to the voxel size and density. The voxel absorbed dose is the ratio of the voxel deposited energy to the voxel mass. The summary of all voxel absorbed doses is the patient absorbed dose map.

[0065] Preferably, the BED dose calculation formula in S8 is as follows:

[0066]

[0067] Where D is the absorbed dose corresponding to the voxel, λ is the effective decay constant of the voxel, and α / β is a classic concept in radiobiology, derived from the LQ model. It reflects the intrinsic sensitivity of tumors or organs to dose fractionation (fractionated irradiation) or dose rate (continuous low-dose irradiation), which is determined by the biological characteristics of the tissue corresponding to the voxel.

[0068] S8. Calculate the absorbed dose parameters such as the dose-volume histogram, mean absorbed dose, and equivalent uniform dose (EUD) of each tissue and OAR based on the obtained absorbed dose results and the obtained patient organ contour mask, and calculate the BED according to the principles of radiobiology.

[0069] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the scope of the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the spirit and scope of protection of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present invention.

Claims

1. A radionuclide therapy dose calculation system, characterized in that: The following steps are involved: S1. Import the patient's quantitative PET / CT images taken before radionuclide therapy and the quantitative SPECT / CT images collected after therapy. The radionuclide type, drug name, image acquisition time, height, weight, and radionuclide distribution in the patient's body are obtained from the quantitative SPECT images. The DICOM parsing module reads the information contained in the quantitative SPECT image tags and determines the nuclide decay energy spectrum used in the subsequent MC simulation based on the type of radionuclide used. S2. Use segmentation tools to segment the patient's outer contours, lesions, and organs at risk (OARs) on PET / CT and SPECT / CT images to obtain organ contour masks. Each image is input into the segmentation model tool. The tool provides preliminary segmentation results of lesions and OARs based on the deep network. The doctor reviews and modifies the results and generates organ contour masks based on the segmentation results. S3. Determine the physiological pharmacokinetic model of the drug based on the obtained therapeutic drug name, determine the corresponding population physiological pharmacokinetic model based on the drug name and the patient's age and gender, and confirm the patient's body shape and tumor burden based on the patient's pre-treatment PET / CT and pre-cycle SPECT / CT images. Perform fitting and optimization based on the population physiological pharmacokinetic model to obtain the patient's individualized physiological pharmacokinetic model; S4. Register the SPECT / CT images acquired at different time points after treatment to the reference image. Select an appropriate reference image for registration based on the metabolic characteristics of the drug used in the treatment. CT-based SPECT image registration: resample the SPECT image to keep its resolution and size consistent with CT, select a registration algorithm, use global image information as a criterion to register the non-reference CT image to the reference image, and save the deformation field. Apply the obtained deformation field to the non-reference SPECT image to register the non-reference SPECT image to the reference SPECT image. Contour-based registration: resample the SPECT image to match the resolution and size of the CT image; select a registration algorithm, using the lesion and OAR contour information obtained in step 2 as the criterion, register the non-reference CT image to the reference image and save the deformation field; apply the obtained deformation field to the non-reference SPECT image to register the non-reference SPECT image to the reference SPECT image; SPECT image-based registration: Using the acquired lesion and OAR contour information as the measurement standard, the non-reference SPECT image is directly registered to the reference SPECT image, and the SPECT is resampled to keep its resolution and size consistent with CT; S5. Extract the radioactivity information corresponding to each acquisition time for each voxel of the multiple sets of registered SPECT images. Fit the radionuclide variation curve of the voxel with time based on the acquisition time and activity information. Calculate the area under the curve, i.e., the voxel time-integrated activity (TIA), to obtain the distribution of radioactive decay in the patient. Extract the radioactivity information of each voxel in each organ based on the organ contour mask. Automatically select a fitting function model based on drug information, organ metabolism pattern, individualized physiological pharmacokinetic model, and voxel radioactivity variation. Automatically select the appropriate fitting parameter value range. Process voxels with poor fitting. Integrate the TIC curves of all voxels to obtain the voxel cumulative radioactive decay count. Store the TIC curves in Bq / ml in a time-integrated activity image (TIA) with a resolution and size consistent with SPECT. S6. Generating a density distribution of each voxel of the patient's personalized computer phantom based on the reference CT image and its corresponding CT value-to-density conversion relationship; determining the material composition of each voxel of the patient's personalized computer phantom based on the organ segmentation result; converting the patient's CT image into a density map based on the CT value-to-density conversion relationship of the image acquisition device; determining the voxel material composition based on the organ contour mask pair; and generating the patient's personalized computer phantom; S7. Perform GPU-accelerated Monte Carlo simulation of the absorbed dose calculation model based on the decay distribution and the patient's individualized computer phantom. Calculate the initial geometric position probability distribution of the decay particles in the Monte Carlo simulation using the acquired TIA images. Generate decay secondary particles in batches based on the type of radionuclide and decay spectrum sampling. Distribute the secondary particles to GPU threads for simultaneous simulation calculations. S8. Calculate the absorbed dose parameters such as the dose-volume histogram, mean absorbed dose, and equivalent uniform dose (EUD) of each tissue and OAR based on the obtained absorbed dose results and the obtained patient organ contour mask, and calculate the BED according to the principles of radiobiology.

2. A radionuclide therapy dose calculation system according to claim 1, characterized in that: The calculation process of voxel time-integrated activity in S5 is as follows: The uptake and excretion of radioactive drugs by human tissues are usually described by exponential decay models, including but not limited to single exponential models and double exponential models. Taking the single exponential model as an example, the relationship between the radioactive activity A(t) and the scanning time t, the initial activity A0, and the effective decay constant λ of the radioactive drug in the voxel is described as: A(t) = A0 × e (-λt) ; By fitting, we can obtain the fixed parameters such as A0 and λ, and calculate the integral of A(t) from t0 to the end time of the cycle tc (generally > 1000h), which is the cumulative radioactivity of the voxel. The formula is as follows: For the case of a specified reference image, the reference image activity distribution Ar and the fitted λ are used to represent A(t) and perform integral calculations.

3. The radionuclide therapy dose calculation system according to claim 1, characterized in that: The absorbed dose calculation model in S7 transmits particles generated by the radionuclide decay module, records the interaction between the particles and the voxels during the transmission process, and records the deposited energy E and voxel index generated in the interaction. After the transmission is completed, the deposited energy in the voxel is the sum of the deposited energies generated by all interactions. The corresponding density of the voxel is determined according to the voxel index, and the voxel mass is determined according to the voxel size and density. The voxel absorbed dose is the ratio of the voxel deposited energy to the voxel mass. The summary of all voxel absorbed doses is the patient absorbed dose map.

4. The radionuclide therapy dose calculation system according to claim 1, characterized in that: The formula for calculating the BED dose in S8 is as follows: Where D is the absorbed dose corresponding to the voxel, λ is the effective decay constant of the voxel, and α / β is a classic concept in radiobiology, derived from the LQ model. It reflects the intrinsic sensitivity of tumors or organs to dose fractionation (fractionated irradiation) or dose rate (continuous low-dose irradiation), which is determined by the biological characteristics of the tissue corresponding to the voxel.

Images