Deformation registration method and device for close-range radiotherapy accumulated dose assessment

Abstract

The invention discloses a deformation registration method and device for close-range radiotherapy accumulated dose evaluation. The method comprises the following steps: S1, importing an image, a sketch, a treatment plan and a dose of preorder treatment; s2, automatically segmenting the pelvis, and dividing the pelvis into an inner pelvic cavity area and an outer pelvic cavity area by taking the pelvis as a boundary; s3, performing reconstruction segmentation on the source application channel in combination with the image and the treatment plan of the previous treatment and the current treatment; s4, performing substructure segmentation on the endangered organs in the pelvic cavity in combination with the dose distribution of the previous treatment and the current treatment; s5, performing rigid registration on the two groups of images of the preorder treatment and the current treatment based on the pelvis information; s6, performing deformation registration of the two groups of images on the basis of a multi-index evaluation model by combining internal and external partitions of the pelvic cavity and each organ at risk and substructure of the organ at risk in the pelvic cavity on the basis of a rigid registration result; and S7, based on a deformation registration result, mapping the preorder image dose to the current image so as to carry out dose superposition evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of brachytherapy, and particularly relates to a deformation registration method and device for brachytherapy cumulative dose evaluation. BACKGROUND

[0002] Cervical cancer, prostate cancer and other pelvic tumors are diseases that have a significant impact worldwide, and radiotherapy plays an important role in the comprehensive treatment of these tumors. Brachytherapy has irreplaceable value in the treatment of cervical cancer, prostate cancer and other pelvic malignant tumors, and its fundamental advantage lies in the ability to accurately deliver high-dose radiation to the inside of the tumor or the adjacent area, while protecting the surrounding key normal organs (such as the rectum, bladder, small intestine, urethra, etc.) to the greatest extent. This "inside-out" treatment method follows the physical law that the radiation dose decreases rapidly with the square of the distance, so that the tumor core can receive a devastating curative dose, while the adjacent sensitive tissues only bear a lower dose of irradiation, thereby significantly improving the local tumor control rate and patient survival rate while effectively reducing the risk of serious complications such as radiation cystitis and proctitis, greatly ensuring the long-term quality of life of patients. The application of modern high-dose rate technology makes the treatment process more convenient and efficient. Therefore, with its unparalleled precise focusing ability and excellent normal tissue protection effect, brachytherapy has become an indispensable means in the comprehensive treatment strategy for pelvic tumors.

[0003] Cervical cancer is a major public health problem worldwide and one of the most common malignancies that seriously threatens women's health. In 2020, there were an estimated 604,000 new cases of cervical cancer worldwide, with 342,000 deaths, 85% of which occurred in developing countries. Radiotherapy can be used for cervical cancer at all clinical stages, especially for advanced cervical cancer. External beam radiotherapy (EBRT) combined with brachytherapy (BT) has become an important mode of radiotherapy for cervical cancer. In the radiotherapy of cervical cancer, due to the high dose rate characteristics of brachytherapy, in order to protect the surrounding normal tissues as much as possible, BT mainly irradiates the primary lesion of cervical cancer, while EBRT mainly irradiates the parametrial region and lymphatic drainage area, both of which play their own advantages, perfectly combine, control the tumor while fully considering the safety of the surrounding normal tissues, reduce the side effects and improve the quality of life of patients. Brachytherapy, also known as internal irradiation, is the implantation of encapsulated radioactive sources into the tumor site through applicators or source catheters for irradiation. The intensity of the radioactive source used for brachytherapy is small, and the effective treatment distance of the rays is short, most of its energy is absorbed by the tumor tissue. The dose distribution around the radioactive source decreases with the square of the distance from the radioactive source - inverse square law. Brachytherapy is rarely used alone and is generally used as an adjunct to external irradiation. It can give a higher dose to a specific site and thus improve the local control rate of the tumor and the quality of life of patients. For the radiotherapy of cervical cancer, only by accurately allocating the dose of internal and external radiotherapy and fully considering the dose limit of organs at risk (OARs), can we improve the local control rate of the tumor while reducing the incidence of radiotherapy complications as much as possible. Image deformation registration technology tracks the deformation and displacement of the tumor target and OARs, realizes the dynamic accumulation of internal and external irradiation doses, and evaluates the cumulative irradiation dose received by the target and OARS during the entire radiotherapy process. However, in actual radiotherapy, there are many challenges in accurately evaluating and optimizing the superposition effect of internal and external irradiation doses. Cervical cancer is one of the most common malignancies in women worldwide, and radiotherapy plays a crucial role in its comprehensive treatment. Currently, the clinical radiotherapy method commonly used is the combination of internal and external irradiation, aiming to maximize the irradiation dose of the tumor target area while minimizing the damage to the surrounding normal tissues.

[0004] As a common female malignancy, the radiotherapy regimen of cervical cancer usually combines external beam radiotherapy (EBRT) and high dose rate brachytherapy (HDR-BT). Although three-dimensional image-guided technology has significantly improved treatment accuracy, there are still key bottlenecks in clinical practice: when traditional BT fraction plans are independently optimized, the cumulative dose effects of previous EBRT and BT fractions cannot be considered comprehensively, which may lead to overlapping of dose hotspots of OARs such as bladder and rectum, increasing the risk of 3-5 grade urinary / gastrointestinal toxicity (5-year incidence rate of 6.8%-8.5%). More difficultly, the complex deformation of pelvic anatomy due to tumor regression and organ filling changes makes dose accumulation a technical difficulty.

[0005] However, in actual radiotherapy, there are many challenges in accurately evaluating and optimizing the superposition effect of internal and external irradiation doses. Due to patient position shift between fractions, dynamic deformation of pelvic organs (such as differences in bladder filling state), and interference of brachytherapy source metal artifacts, there is significant spatial heterogeneity in internal and external irradiation CT images. Traditional registration algorithms (such as deformation models based on B-spline or optical flow) have poor adaptability to nonlinear deformation and are difficult to analyze complex anatomical structure changes, which results in low registration similarity index, large registration error in the rectal region, and direct distortion of the spatial mapping of dose superposition. In response to this, experts and scholars have different solutions, such as: an internal and external irradiation dose superposition method, system, terminal, and storage medium (Patent No. CN 118628544B), which proposes an innovative method to solve the interference problem of brachytherapy sources in internal and external irradiation dose superposition: a Swin Transformer network is used to segment the brachytherapy source / implant needle region to generate a binary mask, and the target region is filled with the minimum value of the surrounding tissue pixels to generate an auxiliary image; then a generative adversarial network (GAN) is used to fuse the original slice, auxiliary image, and segmentation mask to synthesize a high-fidelity no-source positioning image; finally, based on the synthesized image, deformation registration is performed to achieve accurate superposition of multiple internal and external irradiation doses in a unified coordinate system and calculation of equivalent biological dose (EQD2), significantly reducing the registration error caused by metal artifacts. Patent CN 119701233 A discloses a cervical cancer internal and external irradiation dose superposition system based on deep learning, which realizes high-precision spatial alignment of internal and external irradiation CT images through a deformation registration algorithm that combines V-Net and Transformer; uses a convolutional neural network (CNN) to directly learn the dose distribution mapping relationship from the registered image, replacing the traditional physical dose calculation model; finally, the comprehensive dose and volume histogram (Dose and Volume Histogram, DVH) is generated through weighted superposition, and the statistical analysis and visualization tools are combined to evaluate the irradiation dose of the target region and normal tissues, significantly improving the efficiency and accuracy of dose superposition and providing an AI-driven solution for precise radiotherapy throughout the entire process.

[0006] Solving the problem of precise dose superposition in the combined EBRT and BT radiotherapy of pelvic tumors such as cervical cancer and prostate cancer is the key to improving the efficacy of radiotherapy and reducing toxic side effects.

[0007] In the radiotherapy of pelvic tumors such as cervical cancer and prostate cancer, brachytherapy has become a key means to improve efficacy due to its precise high-dose delivery and excellent normal tissue protection capabilities. However, there are serious technical challenges in clinical practice: the dynamic deformation of pelvic organs (such as bladder filling differences) and the metal artifacts of the applicator lead to distortion in image registration between multiple fractions, which seriously hinders the precise superposition of external beam and brachytherapy doses; traditional algorithms cannot analyze the multi-organ displacement of bladder-rectum-tumor, resulting in prediction bias in cumulative dose, which directly affects tumor control and complication risk. Achieving precise deformation registration between fractions in pelvic brachytherapy, and then achieving precise deformation cumulative dose evaluation of organs at risk, is a core problem that needs to be solved for individualized precision radiotherapy. The existing technology mainly has the following shortcomings:

[0008] 1. Traditional image registration methods are not suitable for pelvic anatomy, and existing mainstream technologies (such as the global affine transformation described in patent CN114902836B) assume that the pelvic cavity is a rigid structure, ignoring the elastic deformation of organs such as the bladder and rectum between fractions. Clinical data shows that changes in the filling state of the bladder in cervical cancer patients can cause the anterior wall of the rectum to displace 12-18 mm, resulting in errors in dose mapping after rigid registration.

[0009] 2. The homogeneous weight model leads to dose evaluation bias, and the general deformation registration based on the Demons algorithm applies uniform deformation constraints to all voxels, without considering the spatial heterogeneity of different regions of organs at risk and radiation dose-sensitive regions. Actual measurements show that the dose gradient of the anterior wall of the rectum is as high as 5-8 Gy / mm, but the registration error ratio of traditional methods for this region and non-sensitive areas is as high as 2.3:1 (actual measurement results in Medical Physics in 2021), which seriously affects the evaluation accuracy of key dose parameters such as D2cc.

[0010] 3. The organ-level registration strategy blurs the dose gradient details, and when using B-spline free-form deformation (such as patent CN115620282A) for organ-level registration, the rectum / bladder is modeled as a single continuum, which cannot analyze the micro-scale deformation of the inner thin layer and outer wall. The American Association of Physicists in Medicine (AAPM) TG-43 report indicates that such methods can produce dose calculation errors of up to 12-18% at 3 mm from the surface of the applicator, directly leading to prediction bias in complications such as rectal perforation.

[0011] 4. Dynamic tracking and biological effect modeling are missing, commercial radiotherapy systems rely on rigid image fusion between fractions, lacking probabilistic modeling of multi-organ linked displacement of bladder-rectum-target region. RTOG 0417 clinical trial data shows that the existing technology has a cumulative calculation error of 21.7% for rectal V45 volume, and does not integrate the radiosensitivity difference of substructures such as mucosa and muscle layer (ICRU 89 report suggests that the radiation tolerance dose of the two is 7.3 times different), resulting in insufficient prediction specificity of radiation enteritis of less than 65%.

[0012] 5. The deformation registration deviation of the high-dose region near the target region near the pelvic OAR will seriously affect the accuracy of the final dose accumulation evaluation; while the part of the OAR far from the target region drops rapidly, the deformation registration deviation of this region has little effect on the evaluation of the final cumulative dose. The dose evaluation of the pelvic OAR such as the bladder and rectum is mainly to evaluate the maximum dose evaluation parameters such as D2cc and D1cc, so it is more necessary to accurately evaluate the dose accumulation of the high-dose region. SUMMARY

[0013] To solve the clinical problems of anatomic deformation and dose accumulation in fractionated radiotherapy of pelvic tumors, it is necessary to break through the rigid assumption of traditional registration framework and the limitation of homogeneous deformation modeling. The present application proposes a deformation registration dose accumulation method based on a multi-index evaluation system: by establishing a pelvic organ partition elastic registration model (to solve the anatomic mismatch caused by bladder / rectum elastic deformation), a substructure radiation dose evaluation sensitivity weighting algorithm (to overcome the evaluation deviation of homogeneous deformation constrained dose mapping), an innovative solution for precise evaluation of pelvic tumor brachytherapy dose accumulation is formed. The solution of partitioning the pelvic region, automatically segmenting the substructures of pelvic OAR, multi-index evaluation of deformation registration, and dose accumulation evaluation.

[0014] In order to achieve the above object, the technical scheme of the present application provides a deformation registration method for cumulative dose evaluation of brachytherapy, which comprises the following steps: S1: introducing three-dimensional images of previous treatment, target region and critical organ contouring, treatment plan, and corresponding dose distribution; S2: automatically segmenting the pelvis based on the images of each fraction treatment, and segmenting the images as a whole into an intra-pelvic region and an extra-pelvic region based on the segmented pelvis; S3: combining the images and treatment plans of the previous and current treatments, and reconstructing and segmenting the source channel, and setting the pixel value in the segmented source channel region as the pixel value of the adjacent region outside the region; S4: combining the dose distributions of the previous and current treatments, and performing substructure segmentation on the critical organs in the intra-pelvic region, wherein the substructure needs to contain a region with a specified dose value in each fraction treatment, and the specified dose value is not less than 1 / M of the critical organ limit value, and M is the number of BT treatment fractions; S5: performing rigid registration on the two groups of images of the previous and current treatments based on the pelvis information; S6: based on the rigid registration result, combining the intra-pelvic and extra-pelvic partitions, and the critical organs and critical organ substructures in the intra-pelvic region, and performing deformation registration on the two groups of images based on a multi-index evaluation model; and S7: based on the deformation registration result, mapping the dose of the previous image to the current image to perform dose superposition evaluation.

[0015] Further, step S4 specifically comprises: S41: obtaining the minimum substructure region of each critical organ in each fraction image according to steps S411-S414 as follows: S411: calculating the centroids of the target region and the critical organ, connecting the two centroids, and the connecting line intersects the critical organ contour at point P1; S412: calculating the intersection of the specified dose value region and the critical organ to obtain intersection region A1; S413: uniformly expanding from point P1 to the inside of the critical organ until the expanded region A2 completely contains region A1; S414: calculating the volume V1 of A2 and the volume Vs of the critical organ, and calculating the volume ratio D1 of region A2 in all critical organs; S42: obtaining the minimum substructure region of each critical organ based on the two groups of images, and further expanding the substructure volume ratio of each critical organ to the same as the corresponding critical organ substructure volume ratio in the other group of images; S43: for the critical organ of the wall structure, the substructure is inwardly retracted from the outer wall to the inner wall to form a content region and remove the content region from the substructure, thereby forming the final critical organ substructure.

[0016] Further, in step S6, the deformation registration adopts two types of similarity measures of mutual information MI and kappa statistic KS to ensure that the overall structure and contour matching are considered in the registration process, and the multi-index evaluation model adopts the following objective function:

[0017]

[0018] wherein μ is the transformation parameter, MI1, MI2 are the mutual information of the intra-pelvic region R1 and the extra-pelvic region R2, respectively, KS oari , KS soar i , KS GTV are the KS metrics of each organ at risk, organ at risk substructure, and target volume, respectively, E bend is the bending energy penalty, ω M1 , ω M2 are the evaluation index weights of MI1 and MI2, respectively, ω KSoari , ω KSsoari , ω KSgtv are the evaluation index weights of the i-th organ at risk, i-th organ at risk substructure, and target volume, respectively, ω bend is the evaluation index weight of the bending energy penalty, N is the number of organs at risk, and ω s is the special priority weight of the total KS metric of all adjustable organ at risk substructures.

[0019] Further, in step S6, the weights of each evaluation index are determined by a grid search method, and in terms of image transformation, a general three-step strategy is used: first, rigid transformation, then affine transformation, and finally B-Spline transformation, wherein the B-Spline transformation adopts a six-level resolution registration strategy, at each resolution level, a Gaussian pyramid is used for smoothing and down-sampling the image; and an advanced stochastic gradient descent method is used for optimization at each level, and the maximum number of iterations is set; at the optimal resolution level of MI, bladder KS, rectum KS, and target volume KS, the grid spacing of the B-Spline transformation is specified by different voxels.

[0020] Further, in step S6, the mutual information MI is used to evaluate the degree of alignment between two images, which utilizes image voxel intensity for registration, and the formula is as follows:

[0021]

[0022] wherein I F is the fixed image, I M is the moving image, μ is the transformation parameter, LF, LM are the discrete intensity sets of the fixed image and the moving image, p(f, m) is the joint probability density function, representing the probability of the simultaneous occurrence of intensity f and intensity m, pF(f), pM(m) are the marginal probability density functions of the fixed image I F and the moving image I M , respectively, wherein in image registration, maximizing mutual information means that the two images are optimally aligned after transformation.

[0023] Further, in step S6, KS is used to measure the maximum overlap between the pair of image corresponding contours, and the formula is:

[0024]

[0025] where P0 is the actual overlap ratio of the pair of image corresponding contours, P e is the expected overlap ratio, represents the overlap probability in the random case, and 1 is an indicator function for judging whether the current point belongs to the foreground.

[0026] Further, the transform bending energy penalty is used to regularize the non-rigid transformation, preventing sharp bias in the deformation transformation, and the formula is as follows:

[0027]

[0028] where P is the number of points x, is the transformed point i and point j.

[0029] Further, in step S2, the segmentation of the pelvis is based on the CT value information in each image and the consistency of the pelvis structure in different images, and the threshold segmentation is first performed respectively, and then the two sets of segmentation results are aligned and cross-completed until the two sets of segmentation results are completely consistent.

[0030] Further, in step S3, the source channel is reconstructed based on the source channel residence point coordinates in the treatment plan information, and the channel diameter is adjusted in combination with the image information, so that the outer diameter of each source channel is completely coincided with the high-density information in the image.

[0031] The technical solution of the present invention also provides a deformable registration device for cumulative dose assessment of brachytherapy, which includes the following modules: an information import module: used to import three-dimensional images of previous treatments, outlines of target areas and organs at risk, treatment plans, and corresponding dose distributions; a pelvic region segmentation module: used to automatically segment the pelvis based on images of each treatment fraction, and to segment the entire image into an intrapelvic region and an extrapelvic region based on the segmented pelvis; a source channel reconstruction module: used to combine images and treatment plans of previous and current treatments to reconstruct and segment the source channel, and after segmentation, the pixel values ​​in the source channel region are set to the values ​​of adjacent pixels outside the region; a substructure segmentation module: used to combine images of previous and current treatments and treatment plans to reconstruct and segment the source channel. The dose distribution of the current treatment is used to perform substructure segmentation of the organs at risk in the pelvis, where the substructure needs to include the specified dose value area in each fractionated treatment, and the specified dose value is not less than 1 / M of the organ at risk limit, where M is the BT treatment fraction; the rigid registration module is used to rigidly register the two sets of images of the previous treatment and the current treatment based on pelvic information; the deformable registration module is used to perform deformable registration of the two sets of images based on the rigid registration results, combined with the internal and external pelvic divisions, as well as the organs at risk and their substructures in the pelvis, based on a multi-index evaluation model; the dose superposition evaluation module is used to map the dose of the previous image to the current image based on the deformable registration results for dose superposition evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 4 is a flow chart of the method for evaluating the cumulative dose of brachytherapy according to the present invention. DETAILED DESCRIPTION

[0034] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0035] The key points of the present invention include the following aspects:

[0036] (1) The inside and outside of the pelvic cavity are divided into sub-regions, and a progressive strategy combining rigid and elastic registration is adopted, which significantly optimizes the registration efficiency and accuracy. For the rigid structures such as bones and fat outside the pelvic cavity, 6-DOF rigid body registration is adopted. For the deformed organs such as bladder and rectum, three-level B-spline elastic registration is implemented, and curvature constraint is combined to improve the registration accuracy of internal organs and avoid the transmission of external errors to internal organs. This hierarchical mechanism improves the registration calculation efficiency while greatly reducing the error caused by the overall calculation.

[0037] (2) The sub-structure dose weight partition of the organ at risk is used to solve the error sensitivity problem of the high dose region. Based on the planned dose distribution, the high dose region of the organ at risk is divided into sub-structures, and a weighted term is introduced into the elastic registration objective function to preferentially ensure the deformation registration accuracy of the high dose region, while allowing errors in the low dose region.

[0038] (3) An automatic segmentation method for sub-structures of the organ at risk is proposed. In brachytherapy, the high dose region of the organ at risk is mainly the local region near the target. Through the proposed innovative method, the consistency of the position, shape and volume ratio of the sub-structure of the organ at risk in different treatment fractions is ensured, and the high dose rate region of each fraction is fully covered.

[0039] (4) Additional priority weight is added to the KS metric value of the sub-structure of the organ at risk to ensure the local precise registration of the high-risk region of the organ at risk outside the target, and precise dose control of the high-risk region is achieved.

[0040] Referring to Figure 1 The main steps of the technical solution adopted by the application include:

[0041] 1. Import the three-dimensional image of the previous treatment, the target and the organ at risk contouring, the treatment plan, and the corresponding dose distribution. The previous treatment can be EBRT treatment or the previous fraction of BT treatment.

[0042] 2. Based on the image of each fraction treatment, the pelvis is automatically segmented to divide the image into the pelvic cavity region R1 and the pelvic cavity region R2. The segmentation of the pelvis is mainly based on the CT value information in each fraction image and the consistency of the pelvis structure in different fraction images. First, threshold segmentation is performed respectively, and then the two sets of segmentation results are aligned and cross-completed until the two sets of segmentation results are completely consistent.

[0043] 3. Combine the image and treatment plan of the previous and current treatment to reconstruct and segment the source channel. The source channel is mainly reconstructed based on the source channel residence point coordinates in the treatment plan information, and the channel diameter is adjusted combined with the image information to make the outer diameter of each source channel completely coincide with the high-density information in the image.

[0044] 4. Substructure segmentation is performed on the pelvic cavity and the organs at risk, such as the bladder, rectum, urethra, etc., in combination with the dose distribution of the previous and current treatment; during the substructure segmentation, the local area of the organ at risk close to the tumor is mainly segmented, and the substructure needs to contain the specified dose value region in each fraction treatment, and the specified dose value is not less than 1 / M (M is the BT treatment fraction) of the organ at risk limit value. The substructure of the wall-shaped organ such as the bladder and rectum only includes the wall-shaped part, and the contents such as urine and rectal gas feces need to be removed; the volume of the substructure of the organ at risk in each fraction image is basically consistent.

[0045] 5. Rigid registration is performed based on the pelvic structure in the fraction image;

[0046] 6. For the source channel region in each fraction image, the image pixel value of the region adjacent to the outside of the source channel is replaced to avoid the deviation of mutual information calculation in the deformation registration;

[0047] 7. Based on the rigid registration result, in combination with the internal and external partitions of the pelvic cavity, as well as the organs at risk and the substructure of the organs at risk in the pelvic cavity, a multi-index evaluation model is used for deformation registration;

[0048] 8. Based on the final deformation registration result, the dose of the previous image is mapped to the current image for dose superposition evaluation.

[0049] In the above steps, the substructure segmentation of the organ at risk can be manually segmented or achieved by the following method:

[0050] 1. For each organ at risk in each fraction image, the minimum substructure region is obtained by the following steps:

[0051] (1) The centroids of the target region and the organ at risk are calculated, the two centroids are connected, and the intersection point is the point P1 on the outline of the organ at risk;

[0052] (2) The intersection region A1 is obtained by calculating the intersection of the specified dose value region and the organ at risk;

[0053] (3) The region A2 is uniformly expanded towards the inside of the organ at risk with the point P1 as the center until the region A2 completely contains the region A1;

[0054] (4) The volume V1 of the region A2 and the volume Vs of the organ at risk are calculated, and the volume ratio D1 of the region A2 in the entire organ at risk is calculated.

[0055] 2. The minimum substructure region of each organ at risk is obtained based on two groups of images, and then the substructure volume ratio of each organ at risk in the group with smaller volume ratio is further expanded to the same volume ratio as the corresponding organ at risk in the other group of images.

[0056] 3. The organ at risk is segmented from the outer wall of the organ to the inner wall to form a content region and removed from the substructure to form a final substructure of the organ at risk.

[0057] The deformation registration scheme in the technical solution of the present application is implemented as follows:

[0058] 1. Set the pre-treatment image as a motion image and fuse it into the positioning image of the current BT treatment fraction. Two types of similarity measures, mutual information (MI) and kappa statistic (KS), are used. Mutual information (MI) is a similarity measure based on intensity, which is used to evaluate the alignment between two images. It uses image voxel intensity for registration, and the formula is as follows:

[0059]

[0060] where I F is the fixed image, I M is the moving image, μ is the transformation parameter, and LF, LM are the discrete intensity sets of the fixed image and the moving image. p(f,m) is the joint probability density function, which represents the probability of the simultaneous occurrence of intensity f and intensity m. pF(f) and pM(m) are the marginal probability density functions of the fixed image I F and the moving image I M . This formula measures the entropy of the joint histogram of the two images, representing the amount of common information in the overlapping region of the images. In image registration, maximizing mutual information means that the two images are optimally aligned after transformation. KS is used to measure the maximum overlap between the corresponding contours of the paired images. The formula is as follows:

[0061]

[0062] where P0 is the actual overlap ratio of the corresponding contours of the paired images, P e is the expected overlap ratio, which represents the overlap probability in a random case, and 1 is an indicator function that determines whether the current point belongs to the foreground. The transformation bending energy penalty is used to regularize the non-rigid transformation to prevent sharp deviations in the deformation transformation, and the formula is as follows:

[0063]

[0064] where P is the number of points x, and point j after transformation. This formula represents the sum of the squared distances between all points, which penalizes sharp changes in the transformation and maintains the smoothness of the transformation.

[0065] The objective function that integrates multiple evaluation indicators is represented as:

[0066]

[0067] where MI1, MI2 are the mutual information of regions R1, R2, respectively, KS oari , KS soar i , KS GTV are the KS metrics of each organ at risk, organ at risk substructure, target region, respectively, E bend is the bending energy penalty. ω M1 , ω M2 are the evaluation index weights of MI1, MI2, respectively, ω KSoari , ω KSsoari , ω KSgtv are the KS evaluation index weights of the i-th organ at risk, i-th organ at risk substructure, target region, respectively, N is the number of organs at risk. ω s is the special priority weight of the total KS metric of all organ at risk substructures that can be adjusted, which is generally set to ω s = 1.0.

[0068] 2. The weights of each evaluation index are determined by the grid search method. In terms of image transformation, a general three-step strategy will be used: first, rigid transformation, then affine transformation, and finally B-Spline transformation. The B-Spline transformation adopts a six-level resolution registration strategy. At each resolution level, a Gaussian pyramid is used for smoothing and downsampling the image. An advanced stochastic gradient descent method is used for optimization at each level, and the maximum number of iterations is set. The grid spacing of the B-Spline transformation at the optimal resolution level of MI, bladder KS, rectum KS, and target KS is specified by different voxels.

[0069] The beneficial technical effects of the technical scheme of the present application

[0070] The present application innovatively constructs an efficient and accurate dose accumulation evaluation system for pelvic brachytherapy at different treatment fractions through pelvic partition progressive registration, dose distribution-based automatic segmentation of organ at risk substructures, priority weight of organ at risk substructure, and evaluation index weight optimization technology. It adopts a hierarchical registration strategy inside and outside the pelvis to improve computational efficiency while effectively controlling anatomical deformation error; establishes a hierarchical weight model for sensitive regions to significantly reduce deformation deviation in high-dose areas of organs at risk; and through adaptive fusion of local high-order deformation field and global field, realizes accurate control of dose accumulation evaluation of organs at risk in pelvic tumor brachytherapy. This technical scheme can seamlessly integrate with mainstream radiotherapy systems, greatly improving the accuracy and reliability of dose accumulation evaluation for pelvic tumor radiotherapy.

[0071] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A deformable registration method for brachytherapy cumulative dose assessment, characterized in that: The steps include: S1: Importing 3D images of previous treatments, delineation of target volumes and organs at risk, treatment planning, and corresponding dose distribution; S2: Automatically segment the pelvis based on the images of each treatment session, and then segment the entire image into the intrapelvic region and the extrapelvic region based on the segmented pelvis; S3: Reconstruct and segment the source channel based on the previous and current treatment images and treatment plan. After segmentation, the pixel values ​​within the source channel are set to the values ​​of the adjacent pixels outside the region. S4: Based on the dose distribution of the previous and current treatments, the organs at risk in the pelvis are segmented into substructures. The substructures need to include the area with the specified dose value in each treatment fraction. The specified dose value should be no less than 1 / M of the organ at risk limit, where M is the BT treatment fraction. S5: Rigid registration of the two sets of images of the previous treatment and the current treatment is performed based on pelvic information; S6: Based on the rigid registration results, deformable registration of the two image sets is performed based on the multi-index evaluation model, combining the internal and external pelvic divisions, as well as the organs at risk and their substructures within the pelvis. S7: Based on the deformable registration results, the dose of the previous image is mapped to the current image for dose superposition evaluation.

2. The method according to claim 1, characterized in that Step S4 specifically includes: S41: Obtain the minimum substructure region of each organ at risk in each fractionated image according to the following steps S411-S414: S411: Calculate the centroids of the target area and the organ at risk, connect the two centroids, and intersect the contour of the organ at risk at point P1; S412: Calculate the intersection of the specified dose value area and the organ at risk to obtain the intersection area A1; S413: With point P1 as the center, uniformly expand toward the inside of the organ at risk until the expanded area A2 completely contains area A1; S414: Calculate the volume V1 of A2 and the volume Vs of the organs at risk, and calculate the volume ratio D1 of region A2 in all organs at risk; S42: Based on the two sets of images, the minimum substructure area of ​​each organ at risk is obtained. Then, the group with the smaller substructure volume ratio of each organ at risk is further expanded until it has the same substructure volume ratio as the corresponding organ at risk in the other set of images; S43: For an organ at risk having a wall-like structure, its substructure is retracted from the outer wall to the inner wall of the organ to form a content region and the content region is removed from the substructure to form a final organ at risk substructure.

3. The method according to claim 1, characterized in that In step S6, deformation registration uses two types of similarity metrics, mutual information (MI) and kappa statistic (KS), to ensure that both overall structure and contour matching are taken into account during the registration process. The multi-index evaluation model uses the following objective function: Among them, μ is the transformation parameter, MI1 and MI2 are the mutual information of the pelvic region R1 and the pelvic region R2 respectively, KS oari , KS soari , KS GTV They are the KS metrics of each organ at risk, organ at risk substructure, and target area, respectively. bend is the transformation bending energy penalty; ω M1 、ω M2 are the evaluation index weights of MI1 and MI2 respectively, ω KSoari 、ω KSsoari 、ω KSgtv are the KS evaluation index weights of the i-th organ at risk, the i-th substructure of the organ at risk, and the target area, ω bend is the evaluation index weight of the bending energy penalty, N is the number of organs at risk, ω s A specific priority weighting of the total KS metric for all organ-at-risk substructures is adjustable.

4. The method according to claim 3, characterized in that In step S6, the weight of each evaluation index is determined by a grid search method. In terms of image transformation, a general three-step strategy is used: first, a rigid transformation is performed, then an affine transformation, and finally a B-Spline transformation. Among them, the B-Spline transformation adopts a six-level resolution registration strategy. At each resolution level, a Gaussian pyramid is used to smooth and downsample the image; and an advanced stochastic gradient descent method is used for optimization at each layer, and the maximum number of iterations is set; at the optimal resolution level of MI, bladder KS, rectal KS, and target area KS, the grid spacing of the B-Spline transformation is specified with different voxels.

5. The method according to claim 4, characterized in that In step S6, the mutual information MI is used to evaluate the degree of alignment between the two images. It uses the image voxel intensity for registration, and the formula is as follows: Among them, I F is a fixed image, I M is the moving image, μ is the transformation parameter, LF, LM are the discrete intensity sets of the fixed image and the moving image, p(f,m) is the joint probability density function, which represents the probability of intensity f and intensity m appearing at the same time, pF(f), pM(m) are the fixed image I F and moving image I M The marginal probability density function of , where, in image registration, maximizing mutual information means that the two images achieve optimal alignment after transformation.

6. The method according to claim 5, characterized in that In step S6, KS is used to measure the maximum overlap between corresponding contours of paired images, and the formula is: Among them, P0 is the actual overlap ratio of the corresponding contours of the paired images, P e is the expected overlap ratio, which indicates the overlap probability under random circumstances, and 1 is the indicator function used to determine whether the current point belongs to the foreground.

7. The method according to claim 6, characterized in that Transform Bend Energy Penalty E bend It is used to regularize non-rigid transformations and prevent sharp deviations in deformation transformations. The formula is as follows: Where P is the number of points x, are the transformed points i and j.

8. The method according to claim 1, characterized in that In step S2, the pelvis is segmented based on the CT value information in each fractionated image and the consistency of the pelvic structure in different fractionated images. Threshold segmentation is first performed separately, and then the two sets of segmentation results are aligned and cross-complemented until the two sets of segmentation results are completely consistent.

9. The method according to claim 1, characterized in that In step S3, the source channel is reconstructed based on the coordinates of the source channel residence point in the treatment plan information, and the channel diameter is adjusted in combination with the image information so that the outer diameter of each source channel completely coincides with the high-density information in the image.

10. A deformable registration device for brachytherapy cumulative dose assessment, characterized in that: Includes the following modules: Information import module: used to import the three-dimensional images of previous treatments, the outline of the target area and organs at risk, the treatment plan, and the corresponding dose distribution; Pelvic region segmentation module: used to automatically segment the pelvis based on the images of each treatment session, and divide the entire image into the pelvic region and the extrapelvic region based on the segmented pelvis; Source channel reconstruction module: used to reconstruct and segment the source channel by combining the previous and current treatment images and treatment plan. After segmentation, the pixel values ​​in the source channel area are set to the values ​​of the adjacent pixels outside the area. Substructure segmentation module: It is used to perform substructure segmentation of organs at risk in the pelvic cavity based on the dose distribution of the previous and current treatments. The substructure needs to include the area with the specified dose value in each treatment fraction. The specified dose value should not be less than 1 / M of the organ at risk limit, where M is the BT treatment fraction. Rigid registration module: used to rigidly register the two sets of images of the previous treatment and the current treatment based on pelvic information; Deformable Registration Module: This module is used to perform deformable registration of the two image sets based on the rigid registration results, combined with the internal and external pelvic partitions, as well as the organs at risk and their substructures within the pelvis, and a multi-index evaluation model. Dose superposition evaluation module: used to map the dose of the previous image to the current image based on the deformation registration results for dose superposition evaluation.

Images