A method for evaluating the therapeutic effect of brain tumor disease

Abstract

The application discloses a brain tumor disease treatment effect evaluation method, and belongs to the technical field of radiotherapy plan evaluation, and specifically comprises the following steps: acquiring patient brain multi-modal image data and extracting three-dimensional tumor regions and corresponding perfusion parameters; constructing a biological target region hierarchical structure containing tumor overall regions and hypoxic subregions; obtaining a relative hypoxia degree parameter by calculating the cerebral blood flow ratio of the hypoxic subregion to other regions of the tumor; deconstructing historical treatment case data to establish a dose-therapeutic effect correlation parameter set; training a dose adjustment strategy model for the hypoxic subregion; and generating a final radiotherapy dose plan scheme according to the dose adjustment value output by the model combined with a clinical guideline basic dose. The application overcomes the limitations of traditional uniform dose irradiation through a data-driven dose decision mechanism, thereby providing an objective and reliable evaluation basis for clinical selection and optimization of personalized treatment schemes.

Description

Technical Field

[0001] This invention relates to the field of radiotherapy planning and evaluation technology, specifically to a method for evaluating the treatment effect of brain tumors. Background Technology

[0002] Comprehensive evaluation of brain tumor treatment outcomes is crucial for developing individualized treatment plans and improving patient prognosis. Currently, clinical practice primarily relies on post-treatment imaging follow-up (such as changes in tumor volume) and clinical symptom monitoring, a method inherently lagging in its approach. Although modern radiotherapy techniques, such as image-guided radiotherapy and intensity-modulated radiotherapy (IMRT), have achieved sub-millimeter-level precise dose delivery at the physical level, providing a more accurate dosimetric basis for efficacy assessment, existing evaluation systems still have significant limitations.

[0003] Specifically, current assessment methods are mostly based on changes in anatomical imaging after treatment, failing to fully integrate pre-treatment information on the biological heterogeneity within the tumor. Due to abnormal vascular distribution within the tumor microenvironment, varying degrees of hypoxia can form. Cells in these areas exhibit significantly enhanced radiation resistance, a crucial factor influencing treatment efficacy and leading to local recurrence. However, traditional assessment models struggle to quantify these biological characteristics before or early in treatment and correlate them with specific treatment parameters (such as dose distribution), thus hindering accurate prediction of treatment outcomes.

[0004] Therefore, despite continuous advancements in radiotherapy techniques, methods for early, quantitative prediction and assessment of treatment outcomes remain insufficient. The existing system lacks a comprehensive analytical framework that integrates multimodal imaging biological characteristics, individualized treatment parameters, and long-term clinical outcomes, which limits the prospective optimization of treatment plans and the accurate assessment of patient prognosis. Summary of the Invention

[0005] The purpose of this invention is to provide a method for evaluating the therapeutic effect of brain tumors, and to solve the following technical problems:

[0006] Current methods for evaluating the effectiveness of brain tumor treatment rely on post-treatment imaging, and cannot predict early efficacy based on tumor biological heterogeneity and treatment parameters before treatment.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for evaluating the treatment efficacy of brain tumors includes the following steps:

[0009] S1. Acquire multimodal imaging data of the patient's brain, extract three-dimensional tumor regions and their corresponding perfusion parameters, and establish a tumor biomarker database; the multimodal imaging dataset includes structural magnetic resonance imaging data and perfusion-weighted imaging data;

[0010] S2, perform voxel-level cerebral blood flow calculation and hypoxia region identification on the tumor biomarker database, and construct a hierarchical structure of biological target area containing the whole tumor region and hypoxia subregion based on the identification results;

[0011] S3, in the biological target area hierarchy, screen for target cases with hypoxic sub-regions, extract the blood perfusion parameters of the hypoxic sub-regions and the corresponding parameters of the whole tumor area, and calculate the set of relative hypoxia degree parameters;

[0012] S4, deconstruct historical treatment case data, identify differences between target cases and historical cases in terms of relative hypoxia degree parameters and initial treatment dose, and establish a dose-efficacy correlation parameter set by combining the long-term follow-up results of corresponding cases;

[0013] S5, construct the attribution mapping relationship between the dose-efficacy correlation parameter set and the clinically successful treatment dose, and train and output a dose adjustment strategy model for the hypoxic subregion;

[0014] S6. Obtain the dose-efficacy correlation parameter set of the current patient's tumor and input it into the strategy model. Based on the dose adjustment value output by the model and combined with the baseline dose of the clinical guidelines, generate the final radiotherapy dose plan.

[0015] As a further aspect of the present invention: the specific process for determining the three-dimensional tumor region in step S1 is as follows:

[0016] The structural magnetic resonance imaging data is preprocessed, including noise filtering and intensity normalization. Based on the preprocessed structural magnetic resonance imaging data, a three-dimensional spatial coordinate system of the brain is established using a voxel mesh reconstruction method to generate a three-dimensional brain model.

[0017] The three-dimensional brain model is divided into multiple regularly arranged three-dimensional mesh units, and the average gray value of all voxels in each three-dimensional mesh unit is calculated. The average gray value of each three-dimensional mesh unit is compared with a preset gray value threshold range, and all three-dimensional mesh units whose average gray value exceeds the gray value threshold range are marked as abnormal mesh units. Spatially adjacent abnormal mesh units are merged to form a continuous three-dimensional abnormal region. The boundary optimization processing of the three-dimensional abnormal region is performed to obtain the three-dimensional tumor region.

[0018] As a further aspect of the present invention: in step S1, the specific process for obtaining the perfusion parameters is as follows:

[0019] The perfusion-weighted imaging data is spatially registered with the three-dimensional brain model, and the perfusion-weighted imaging data corresponding to the three-dimensional tumor region is extracted based on the registration result. The perfusion-weighted imaging data includes image sequences at multiple time points continuously acquired after contrast agent injection.

[0020] The signal intensity of each voxel at different time points is obtained from the continuously acquired image sequence at multiple time points, forming the time-signal intensity curve of each voxel, and the time-signal intensity curve of each voxel is converted into the contrast agent concentration-time curve.

[0021] The anterior cerebral artery region was selected as the reference region, and the average concentration-time curve of this region was extracted as the arterial input function. A single-compartment model based on tracer kinetics was adopted, and the deconvolution equation between the concentration-time curve of each voxel and the arterial input function was solved by singular value decomposition to obtain the quantified value of cerebral blood flow of each voxel.

[0022] As a further aspect of the present invention, the specific process of spatial registration is as follows:

[0023] Feature point sets are extracted from the structural magnetic resonance imaging data and perfusion-weighted imaging data, respectively. The feature point sets include feature points of the midline structure of the brain, the outline of the ventricles, and the main sulci and gyri of the brain.

[0024] The rigid registration algorithm based on mutual information performs preliminary alignment of two sets of feature points, and the free deformation registration algorithm based on B-splines performs nonlinear optimization of local details to establish a spatial correspondence between perfusion-weighted imaging data and the three-dimensional brain model.

[0025] As a further aspect of the present invention: the specific construction process of the biological target region hierarchy in S2 is as follows:

[0026] Based on the quantified cerebral blood flow values ​​of each voxel within the three-dimensional tumor region, a set of continuous voxels with cerebral blood flow values ​​below a preset threshold is marked as a hypoxic subregion. The three-dimensional tumor region is used as the first-level target region, and the hypoxic subregion is used as the second-level target region contained within the first-level target region, thus establishing a hierarchical structure of biological target regions with spatial inclusion relationships.

[0027] As a further aspect of the present invention: the specific process for obtaining the relative hypoxia degree parameter set in S3 is as follows:

[0028] The average cerebral blood flow value of all voxels within the hypoxic subregion is calculated as a first parameter; the average cerebral blood flow value of the remaining voxels within the three-dimensional tumor region excluding the hypoxic subregion is calculated as a second parameter; the ratio of the first parameter to the second parameter is calibrated as the relative hypoxia degree parameter of the hypoxic subregion; the relative hypoxia degree parameters of all target cases are collected to form the relative hypoxia degree parameter set.

[0029] As a further aspect of the present invention: the specific process of establishing the dose-efficacy correlation parameter set in S4 is as follows:

[0030] Radiotherapy planning data and long-term follow-up data of historical brain tumor cases were collected. The radiotherapy planning data included the initial dose allocated to the three-dimensional tumor region and the relative hypoxia degree parameter of the hypoxic subregion for each case. The long-term follow-up data indicated whether local recurrence occurred in each case. The tumor volume of each historical case was calculated based on the three-dimensional tumor region. For each historical case marked as having local recurrence, the difference value of the relative hypoxia degree parameter and the tumor volume ratio between it and each historical case without recurrence were calculated. A reference case set was selected from the historical cases without recurrence based on the parameter difference value and the volume ratio. The statistical characteristic value of the initial dose of the reference case set was calculated as the reference dose value, and the dose adjustment value of the recurrence historical case was calculated. The dose adjustment value was the reference dose value minus the initial dose of the recurrence historical case. The dose-efficacy correlation parameter set was constructed by using the relative hypoxia degree parameter and tumor volume of all recurrence historical cases as input features and the corresponding dose adjustment value as output labels.

[0031] As a further aspect of the present invention: the specific construction process of the dose adjustment strategy model in S5 is as follows:

[0032] The dose-efficacy correlation parameter set is divided into a training set and a validation set. The relative hypoxia level and tumor volume of cases in the training set are used as input features, and the corresponding dose adjustment values ​​are used as prediction targets. A gradient boosting regression algorithm is employed to establish a nonlinear mapping relationship between the input features and the prediction targets through iterative optimization of decision tree combinations. During training, the model parameters are optimized by minimizing the mean squared error between the predicted dose adjustment value and the actual calculated dose adjustment value. The validation set is used to evaluate the generalization performance of the trained model, and cross-validation is used to calculate the model's average prediction accuracy. Finally, the model exhibiting the highest prediction accuracy on the validation set is selected as the trained dose adjustment strategy model.

[0033] As a further aspect of the present invention: the specific process of selecting a reference case set from historical cases that have never relapsed based on the parameter difference value and volume ratio is as follows:

[0034] Calculate the absolute value d of the difference in relative hypoxia degree parameter between each non-relapsed case and the current relapsed case; calculate the tumor volume ratio r between each non-relapsed case and the current relapsed case.

[0035] According to the calculation formula Calculate the candidate score S, where w1 and w2 are preset unit coefficients, and form a reference case set by including all non-relapsed cases whose candidate scores are higher than the preset score threshold.

[0036] As a further aspect of the present invention: S2 further includes generating a final radiotherapy dose plan based on the standard dose value determined by clinical treatment guidelines that match the characteristics of the patient's three-dimensional tumor region if there is no hypoxic subregion within the three-dimensional tumor region.

[0037] The beneficial effects of this invention are:

[0038] 1) This invention establishes a comparative analysis mechanism between relapsed and non-relapsed cases by collecting radiotherapy planning data and long-term follow-up results of historical brain tumor cases. It uses gradient boosting regression algorithm to construct a precise mapping relationship from relative hypoxia degree parameters to dose adjustment values, realizing scientific decision-making on dose adjustment values ​​for hypoxic areas. It provides personalized dose adjustment schemes validated by large sample data for tumor areas with different hypoxia degrees, significantly improving the scientificity and reliability of dose decision-making.

[0039] 2) Addressing the bottleneck of existing radiotherapy techniques' insufficient consideration of differences in radiosensitivity within tumors, this invention extracts and quantifies the key "relative hypoxia degree parameter," transforming the differences in radioresistance among different regions within the tumor into a calculable basis for dose adjustment. This achieves a fundamental shift from traditional "uniform dose coverage" to "biologically guided dose sculpting." The effect is that radiotherapy planning can redistribute dose based on the inherent radiosensitivity of different regions within the tumor, thereby directly responding to the biological challenge of radioresistance caused by hypoxia at the physical dose level, providing a precise dosimetric approach to overcoming local recurrence.

[0040] 3) Based on the baseline dose determined according to clinical treatment guidelines, this invention forms a comprehensive dose distribution that conforms to clinical norms and addresses biological heterogeneity by superimposing model-optimized dose adjustment values ​​at the spatial location of hypoxic subregions. This preserves the safety and standardization of standard treatment protocols while effectively overcoming the radiation resistance of hypoxic areas through local dose enhancement. As a result, the final radiotherapy plan achieves a better balance between tumor control rate and normal tissue protection, providing technical support for improving patient prognosis. Attached Figure Description

[0041] The invention will now be further described with reference to the accompanying drawings.

[0042] Figure 1 This is a schematic diagram of a method for evaluating the therapeutic effect of brain tumor disease according to the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 As shown, this invention provides a method for evaluating the therapeutic effect of brain tumor disease, comprising the following steps:

[0045] S1. Acquire multimodal imaging data of the patient's brain, extract three-dimensional tumor regions and their corresponding perfusion parameters, and establish a tumor biomarker database; the multimodal imaging dataset includes structural magnetic resonance imaging data and perfusion-weighted imaging data;

[0046] The patient's brain is scanned using a magnetic resonance imaging (MRI) system to acquire structural MRI data. This data, based on the resonance and relaxation properties of hydrogen nuclei in a strong magnetic field, can generate high-resolution images of the brain's anatomical structure, clearly distinguishing gray matter, white matter, cerebrospinal fluid, and abnormal lesions. Simultaneously, perfusion-weighted imaging data is acquired. This involves rapidly and continuously acquiring multi-temporal image sequences after intravenous injection of a paramagnetic contrast agent. Based on the signal intensity-time curve caused by the local magnetic field change when the contrast agent first passes through the tissue, parameters reflecting the microcirculation and blood flow state of the tissue are calculated using the tracer dilution principle. Based on the structural MRI data, image segmentation and 3D reconstruction techniques are used to determine the spatial extent and contour of the three-dimensional tumor. Furthermore, the perfusion-weighted imaging data is spatially correlated with the three-dimensional tumor region, and raw perfusion parameter data precisely corresponding to the spatial location of this region is extracted. The three-dimensional tumor region information and its corresponding perfusion parameter data are integrated to form a structured tumor biomarker.

[0047] S2, perform voxel-level cerebral blood flow calculation and hypoxia region identification on the tumor biomarker database, and construct a hierarchical structure of biological target area containing the whole tumor region and hypoxia subregion based on the identification results;

[0048] Based on perfusion-weighted imaging data from the tumor biomarker database, a tracer kinetic model is used to calculate the cerebral blood flow value of each voxel within the three-dimensional tumor region. This calculation involves analyzing the time-signal intensity curve of each voxel, converting it into a contrast agent concentration-time curve, and selecting the time curve of the large artery region as the arterial input function. The quantitative cerebral blood flow is obtained by deconvolution operation of the model. The calculated cerebral blood flow values ​​of all voxels are compared with a threshold obtained statistically based on the blood flow distribution of normal brain tissue. A continuous set of spatial voxels with cerebral blood flow values ​​consistently below the threshold is identified and marked as hypoxic subregions. The three-dimensional tumor region is used as the primary target region, and one or more of the identified hypoxic subregions are used as secondary target regions to construct a hierarchical structure of biological target regions with clear spatial affiliation.

[0049] S3, in the biological target area hierarchy, screen for target cases with hypoxic sub-regions, extract the blood perfusion parameters of the hypoxic sub-regions and the corresponding parameters of the whole tumor area, and calculate the set of relative hypoxia degree parameters;

[0050] For target cases with hypoxic subregions in the hierarchical structure of the biological target area, the average cerebral blood flow value of all voxels in the hypoxic subregion and the average cerebral blood flow value of the remaining voxels in the three-dimensional tumor region excluding the hypoxic subregion are calculated respectively. The average cerebral blood flow value of the hypoxic subregion is divided by the average cerebral blood flow value of other regions within the tumor to obtain the relative hypoxia degree parameter of the hypoxic subregion. This ratio eliminates the differences in baseline blood flow levels between individual patients, thereby more objectively quantifying the hypoxia severity of a specific region relative to its own tumor background. The relative hypoxia degree parameters corresponding to each hypoxic subregion in all target cases are collected to form a set of relative hypoxia degree parameters for subsequent analysis.

[0051] S4, deconstruct historical treatment case data, identify differences between target cases and historical cases in terms of relative hypoxia degree parameters and initial treatment dose, and establish a dose-efficacy correlation parameter set by combining the long-term follow-up results of corresponding cases;

[0052] Treatment records containing historical brain tumor cases were collected. These records included the initial prescribed dose for the three-dimensional tumor region as recorded in the radiotherapy plan for each case, the relative hypoxia degree parameter of the hypoxic subregion calculated based on its imaging data, and the long-term clinical follow-up results recording whether local recurrence occurred. Based on the characteristics of the current target case, a subset of cases with similar ranges of relative hypoxia degree parameters were selected from the historical cases. The distribution differences in initial treatment doses between recurrent and non-recurrent cases in this subset were analyzed. For recurrent cases, by comparing the dose differences with non-recurrent cases and combining their relative hypoxia degree parameters, the dose adjustment amount that may be needed to achieve the therapeutic effect was deduced. The relative hypoxia degree parameter, initial / derived dose, and corresponding treatment result (recurrent / non-recurrent) of each historical case (or deduced case) were correlated and coded to construct a parameter set for revealing the correlation between dose and efficacy.

[0053] S5, construct the attribution mapping relationship between the dose-efficacy correlation parameter set and the clinically successful treatment dose, and train and output a dose adjustment strategy model for the hypoxic subregion;

[0054] Using the relative hypoxia level parameter from the dose-efficacy correlation parameter set as input features, and the recommended dose value or dose adjustment amount derived from analysis and associated with treatment success (no relapse) as the target output; a regression model is trained on the parameter set using the gradient boosting regression algorithm in machine learning; this training process iteratively constructs a series of decision trees, continuously correcting the prediction error of the preceding trees, thereby learning a high-precision mapping relationship between complex, nonlinear input features and continuous output values; during the training process, cross-validation technology is used to evaluate the generalization ability of the model, and the final model parameters are determined by optimizing the loss function (such as mean squared error), thereby obtaining a strategy model that can predict personalized dose adjustment values ​​based on the input relative hypoxia level parameter.

[0055] S6. Obtain the dose-efficacy correlation parameter set of the current patient's tumor and input it into the strategy model. Based on the dose adjustment value output by the model and combined with the baseline dose of the clinical guidelines, generate the final radiotherapy dose plan.

[0056] For new patients requiring treatment planning, steps S1 to S3 are first executed to obtain the relative hypoxia level parameters of their tumors and add them to the parameter set. These parameters are then input into a pre-trained dose adjustment strategy model, which automatically outputs a suggested dose adjustment value for the specific hypoxic subregion of the patient. Simultaneously, based on the clinical characteristics of the patient's three-dimensional tumor region (such as type, location, and volume), a uniform baseline dose is determined with reference to authoritative clinical treatment guidelines. Finally, in the radiotherapy planning system, the patient's hypoxic subregion is precisely located in three-dimensional space against the background of the baseline dose distribution, and the dose adjustment value output by the model is superimposed on this region. Through optimization algorithms, the physical feasibility and gradient smoothness of the dose distribution are ensured, thereby generating a final radiotherapy dose plan that integrates standard treatment and personalized enhancement strategies.

[0057] In another preferred embodiment of the present invention, the specific process of determining the three-dimensional tumor region in S1 is as follows:

[0058] The structural magnetic resonance imaging data is preprocessed, including noise filtering and intensity normalization. Based on the preprocessed structural magnetic resonance imaging data, a three-dimensional spatial coordinate system of the brain is established using a voxel mesh reconstruction method to generate a three-dimensional brain model.

[0059] The three-dimensional brain model is divided into multiple regularly arranged three-dimensional mesh units, and the average gray value of all voxels in each three-dimensional mesh unit is calculated. The average gray value of each three-dimensional mesh unit is compared with a preset gray value threshold range, and all three-dimensional mesh units whose average gray value exceeds the gray value threshold range are marked as abnormal mesh units. Spatially adjacent abnormal mesh units are merged to form a continuous three-dimensional abnormal region. The boundary optimization processing of the three-dimensional abnormal region is performed to obtain the three-dimensional tumor region.

[0060] First, the structural magnetic resonance imaging (MRI) data is preprocessed. A specific filtering algorithm is used to eliminate random noise in the images, as the original images are subject to noise during acquisition due to equipment precision and human body micro-movements, interfering with subsequent analysis. Simultaneously, intensity normalization is performed to adjust the image signals obtained from different scanning sequences to a unified standard. This is because different scanning parameters can cause the same tissue to exhibit different signal intensities, and normalization eliminates this technical difference. Based on the preprocessed data, a voxel grid reconstruction method is used to establish a three-dimensional spatial coordinate system for the brain. This involves rearranging consecutive two-dimensional slices in three dimensions according to their spatial relationships to form a complete three-dimensional brain model. Next, the three-dimensional model is divided into multiple regularly arranged cubic grid cells, and the average gray value of all voxels within each cell is calculated. The principle behind this is that tumor tissue and normal brain tissue typically have different signal characteristics in MRI images; calculating the average value of local regions can better reflect tissue characteristics. The average grayscale value of each grid cell is then compared with a preset grayscale threshold range, which is statistically derived from analyzing a large amount of normal brain tissue image data. All grid cells exceeding this range are marked as anomalous cells, because tumor tissue often exhibits different signal intensities than normal tissue in magnetic resonance images. Spatially adjacent anomalous grid cells are then merged to form a continuous three-dimensional anomalous region, based on the characteristic that tumors typically grow in clusters. Finally, the boundaries of the three-dimensional anomalous region are optimized using morphological manipulation methods to smooth the contours and fill internal voids, resulting in a precise three-dimensional tumor region that accurately reflects the actual shape and extent of the tumor.

[0061] In another preferred embodiment of the present invention, the specific process for obtaining the perfusion parameters in step S1 is as follows:

[0062] The perfusion-weighted imaging data is spatially registered with the three-dimensional brain model, and the perfusion-weighted imaging data corresponding to the three-dimensional tumor region is extracted based on the registration result. The perfusion-weighted imaging data includes image sequences at multiple time points continuously acquired after contrast agent injection.

[0063] The signal intensity of each voxel at different time points is obtained from the continuously acquired image sequence at multiple time points, forming the time-signal intensity curve of each voxel, and the time-signal intensity curve of each voxel is converted into the contrast agent concentration-time curve.

[0064] The anterior cerebral artery region was selected as the reference region, and the average concentration-time curve of this region was extracted as the arterial input function. A single-compartment model based on tracer kinetics was adopted, and the deconvolution equation between the concentration-time curve of each voxel and the arterial input function was solved by singular value decomposition to obtain the quantified value of cerebral blood flow of each voxel.

[0065] First, based on the established spatial correspondence, image data perfectly matching the spatial location of the three-dimensional tumor region is extracted from perfusion-weighted imaging data. This data includes complete image sequences acquired continuously at multiple time points after contrast agent injection. The principle behind this is that only by ensuring precise spatial correspondence can the subsequently calculated blood flow parameters accurately reflect the true blood flow status of each region of the tumor. Next, the signal intensity values ​​of each voxel at different time points are obtained from these continuous time-point image sequences. These values ​​are then connected in chronological order to form the time-signal intensity curve for each voxel. This is because the contrast agent changes the local magnetic resonance signal as it flows through the tissue, and tracking the signal changes over time reflects the dynamic distribution process of the contrast agent. Finally, the time-signal intensity curve is converted into a contrast agent concentration- The time-conversion curve is based on a specific mathematical relationship between contrast agent concentration and signal intensity. This conversion quantifies signal changes into specific concentration changes. Subsequently, the anterior cerebral artery region is selected as a reference area, and the average concentration-time curve of this region is extracted as the arterial input function. This region is chosen because the anterior cerebral artery has a relatively straight course and clear visualization, providing reliable arterial blood samples. Finally, a single-compartment model based on tracer kinetics is used to solve the deconvolution equation between the concentration-time curve of each voxel and the arterial input function using mathematical methods. The principle of this calculation process is to decompose the contrast agent concentration change in the tissue into a comprehensive result of arterial input and local blood flow effects. By performing inverse calculations, the quantified value of cerebral blood flow for each voxel can be obtained.

[0066] It is understandable that accurate calculation of cerebral blood flow requires complete kinetic information and reliable reference benchmarks; images at a single time point cannot reflect the dynamic characteristics of blood flow. Its advantage lies in the fact that through complete kinetic model analysis, it can eliminate the interference of intravascular contrast agents on tissue perfusion assessment, obtaining purer tissue blood flow parameters; obtaining precise quantitative values ​​of cerebral blood flow for each voxel, which is a key basis for identifying hypoxic areas; from the perspective of the overall plan, the precise implementation of this step provides a reliable quantitative data foundation for subsequent identification of hypoxic subregions. Only based on accurate cerebral blood flow values ​​can regions with different blood flow conditions within the tumor be correctly distinguished, thereby providing a scientific basis for dose adjustment and ensuring that subsequent dose increases can accurately target hypoxic areas that truly require enhanced irradiation, thus improving the targeting and effectiveness of radiotherapy.

[0067] In another preferred embodiment of the present invention, the specific process of spatial registration is as follows:

[0068] Feature point sets are extracted from the structural magnetic resonance imaging data and perfusion-weighted imaging data, respectively. The feature point sets include feature points of the midline structure of the brain, the outline of the ventricles, and the main sulci and gyri of the brain.

[0069] The rigid registration algorithm based on mutual information performs preliminary alignment of two sets of feature points, and the free deformation registration algorithm based on B-splines performs nonlinear optimization of local details to establish a spatial correspondence between perfusion-weighted imaging data and the three-dimensional brain model.

[0070] First, feature point sets are extracted from structural magnetic resonance imaging (MRI) data and perfusion-weighted imaging data. These feature points are selected from stable anatomical structures such as the midline structure of the brain, the outline of the ventricles, and the major sulci and gyri. This is because these structures can maintain stable morphological features under different imaging modes and can serve as reliable registration reference points. In practice, the system automatically identifies the boundaries of these anatomical structures through edge detection and contour tracking algorithms, and then selects feature points on these boundaries at fixed intervals to ensure that the feature points are evenly distributed and representative. Next, a rigid registration algorithm based on mutual information is used to initially align the two sets of feature points. This method analyzes the statistical patterns of grayscale distribution in corresponding regions of the two images to find spatial transformation parameters that maximize the correlation between the two images, effectively solving the problem of overall displacement and rotation caused by different patient positions. After the initial alignment is completed, a spline curve-based free deformation registration algorithm is used to perform nonlinear optimization of local details. This algorithm sets a control grid on the image and adjusts the displacement of the grid points to achieve local image deformation, which can accurately compensate for local geometric distortions caused by organ deformation or pulse sequence differences. Finally, a precise spatial correspondence is established between the perfusion-weighted imaging data and the three-dimensional model of the brain. This is achieved by applying the calculated spatial transformation parameters to the entire image dataset, ensuring that each voxel can find an accurate corresponding spatial position in both images.

[0071] It is understandable that patient positioning and scanning parameters differ when data is acquired by different imaging devices, necessitating precise registration to achieve spatial consistency of multimodal data. The advantage lies in employing a global-to-local registration strategy, ensuring both registration stability and precise matching of details, overcoming the challenge of single registration methods simultaneously achieving both global and local accuracy. The direct purpose of this approach is to establish an accurate spatial basis for subsequent analysis, ensuring that functional data obtained from perfusion imaging precisely corresponds to anatomical structures. From an overall perspective, the precise implementation of this step ensures that all subsequent analyses are conducted within a unified spatial reference system, allowing tumor regions identified on anatomical images to be accurately mapped onto functional images. This, in turn, guarantees that calculated cerebral blood flow values, identified hypoxic subregions, and final dose adjustments are precisely applied to the target location, providing a reliable spatial guarantee for truly precise personalized radiotherapy.

[0072] In another preferred embodiment of the present invention, the specific construction process of the biological target region hierarchy in S2 is as follows:

[0073] Based on the quantified cerebral blood flow values ​​of each voxel within the three-dimensional tumor region, a set of continuous voxels with cerebral blood flow values ​​below a preset threshold is marked as a hypoxic subregion. The three-dimensional tumor region is used as the first-level target region, and the hypoxic subregion is used as the second-level target region contained within the first-level target region, thus establishing a hierarchical structure of biological target regions with spatial inclusion relationships.

[0074] In another preferred embodiment of the present invention, the specific process for obtaining the relative hypoxia degree parameter set in step S3 is as follows:

[0075] The average cerebral blood flow value of all voxels within the hypoxic subregion is calculated as a first parameter; the average cerebral blood flow value of the remaining voxels within the three-dimensional tumor region excluding the hypoxic subregion is calculated as a second parameter; the ratio of the first parameter to the second parameter is calibrated as the relative hypoxia degree parameter of the hypoxic subregion; the relative hypoxia degree parameters of all target cases are collected to form the relative hypoxia degree parameter set.

[0076] In another preferred embodiment of the present invention, the specific process of establishing the dose-efficacy correlation parameter set in step S4 is as follows:

[0077] Radiotherapy planning data and long-term follow-up data of historical brain tumor cases were collected. The radiotherapy planning data included the initial dose allocated to the three-dimensional tumor region and the relative hypoxia degree parameter of the hypoxic subregion for each case. The long-term follow-up data indicated whether local recurrence occurred in each case. The tumor volume of each historical case was calculated based on the three-dimensional tumor region. For each historical case marked as having local recurrence, the difference value of the relative hypoxia degree parameter and the tumor volume ratio between it and each historical case without recurrence were calculated. A reference case set was selected from the historical cases without recurrence based on the parameter difference value and the volume ratio. The statistical characteristic value of the initial dose of the reference case set was calculated as the reference dose value, and the dose adjustment value of the recurrence historical case was calculated. The dose adjustment value was the reference dose value minus the initial dose of the recurrence historical case. The dose-efficacy correlation parameter set was constructed by using the relative hypoxia degree parameter and tumor volume of all recurrence historical cases as input features and the corresponding dose adjustment value as output labels.

[0078] First, complete treatment data for historical brain tumor cases needs to be collected, including radiotherapy planning records and long-term follow-up results for each case. These treatment planning records contain initial dose data for the three-dimensional tumor region and relative hypoxia parameters of hypoxic subregions obtained through image analysis. The follow-up results clearly record whether the patient experienced local recurrence. The principle behind this is that only data based on real clinical treatment results can establish a reliable dose prediction relationship. Next, the tumor volume for each case is calculated based on the spatial data of the three-dimensional tumor region. This is obtained by counting the total number of voxels constituting the tumor region and multiplying it by the volume of a single voxel, as tumor volume is a crucial factor affecting treatment efficacy. For each recurrent case, the difference in relative hypoxia parameters between it and all non-recurrent cases needs to be calculated. This difference is obtained by directly calculating the absolute difference in relative hypoxia parameters between the two cases, and simultaneously calculating the tumor volume ratio between the two cases, obtained by dividing the tumor volumes of the two cases. The principle behind this is that by comparing these key parameters of recurrent and non-recurrent cases, key factors affecting treatment success can be identified. Then, a reference case set is selected from the non-recurring cases based on the calculated parameter differences and tumor volume ratios. This is done by setting an appropriate threshold range to select cases with small parameter differences and similar tumor volumes, as these cases have better comparability. Next, the statistical characteristics of the initial dose for all cases in the reference case set are calculated as reference dose values, typically using the mean or median, to obtain a baseline value representing a successful treatment dose. Then, the dose adjustment value for recurring cases is calculated by subtracting the initial dose actually used for the recurring case from the reference dose value. This difference reflects the amount of dose adjustment required to achieve a successful treatment effect. Using the relative hypoxia level parameter of the recurring cases as input features and the calculated dose adjustment value as output labels, a dose-efficacy correlation parameter set is constructed. This established correspondence allows the model to learn the mapping pattern from hypoxia level to the required dose adjustment. Finally, a gradient boosting regression model was trained on the training sample set. The model parameters were continuously adjusted to minimize the difference between the predicted value and the actual dose adjustment value. The prediction accuracy of the model was evaluated by cross-validation, which involves dividing the data into multiple subsets and using different subsets for training and validation in turn, thereby selecting the model with the highest prediction accuracy as the final pre-trained dose prediction model.

[0079] It is understandable that medical decision-making is essentially a reasoning process based on experiential knowledge, and historical case data serves as a systematic carrier of this experience. By analyzing treatment data from a large number of relapsed and non-relapsed cases, the potential patterns between relative hypoxia and optimal dose adjustment can be revealed. This data-driven approach is more systematic and reproducible than relying solely on the physician's personal experience. The effectiveness of radiotherapy is directly reflected in long-term follow-up results. Relapsed cases indicate inadequacies in the original dosage regimen, while non-relapsed cases represent relatively successful treatment experiences. Comparative analysis can establish a reliable mapping from biological characteristics to clinical decisions. By systematically analyzing the successes and failures of historical cases, the inherent patterns between relative hypoxia and required dose adjustment can be discovered, avoiding the repetition of past erroneous treatment plans. An intelligent system can be established that automatically recommends appropriate dose adjustments based on the patient's specific hypoxia parameters. By providing precise dose adjustment suggestions based on the individualized hypoxia characteristics of different patients, effective treatment of hypoxic areas is ensured while avoiding the risks of overtreatment, ultimately improving the success rate and safety of treatment protocols.

[0080] In another preferred embodiment of the present invention, the specific construction process of the dose adjustment strategy model in step S5 is as follows:

[0081] The dose-efficacy correlation parameter set is divided into a training set and a validation set. The relative hypoxia level and tumor volume of cases in the training set are used as input features, and the corresponding dose adjustment values ​​are used as prediction targets. A gradient boosting regression algorithm is employed to establish a nonlinear mapping relationship between the input features and the prediction targets through iterative optimization of decision tree combinations. During training, the model parameters are optimized by minimizing the mean squared error between the predicted dose adjustment value and the actual calculated dose adjustment value. The validation set is used to evaluate the generalization performance of the trained model, and cross-validation is used to calculate the model's average prediction accuracy. Finally, the model exhibiting the highest prediction accuracy on the validation set is selected as the trained dose adjustment strategy model.

[0082] In another preferred embodiment of the present invention, the specific process of selecting a reference case set from the historical cases that have never relapsed based on the parameter difference value and volume ratio is as follows:

[0083] Calculate the absolute value d of the difference in relative hypoxia degree parameter between each non-relapsed case and the current relapsed case; calculate the tumor volume ratio r between each non-relapsed case and the current relapsed case.

[0084] According to the calculation formula Calculate the candidate score S, where w1 and w2 are preset unit coefficients, and form a reference case set by including all non-relapsed cases whose candidate scores are higher than the preset score threshold.

[0085] First, the absolute value of the difference in relative hypoxia severity between each non-relapsed case and the current relapsed case is calculated. This is obtained by subtracting the relative hypoxia severity parameters of the two cases and taking the absolute value. This calculation method is used because absolute values ​​can eliminate directional influences and accurately reflect the magnitude of the difference in hypoxia severity between the two cases. Simultaneously, the tumor volume ratio between each non-relapsed case and the current relapsed case is calculated. This is obtained by dividing the tumor volumes of the two cases. Using a ratio rather than an absolute value difference is because the ratio better reflects the relative difference in volume and avoids affecting matching accuracy due to large absolute volume differences. Then, the above two parameters are combined into a candidate score according to a calculation formula that includes preset weighting coefficients. These weighting coefficients are obtained by analyzing historical data. The parameters are determined based on their influence on the treatment outcome. In the calculation formula, the absolute value of the difference in relative hypoxia is processed in reciprocal form. This is because the smaller the difference, the closer the hypoxia levels of the two cases are, and the higher the matching score should be. The tumor volume ratio is processed as the absolute difference from 1. This is because when the tumor volumes of the two cases are exactly the same, the ratio is 1, and the matching degree is the highest. As the ratio deviates from 1, the matching degree decreases accordingly. Finally, the calculated candidate scores are compared with a preset score threshold. This threshold is determined by statistical analysis to ensure that the selected reference cases are both numerous and maintain good matching quality. All non-recurring cases with candidate scores higher than the threshold are included in the reference case set.

[0086] The key is to select highly comparable cases from a large number of non-relapsed cases as reference benchmarks. Its advantage lies in its ability to systematically assess the similarity between cases through multi-parameter comprehensive evaluation and quantitative scoring mechanisms, avoiding biases that may result from single-parameter matching. The direct purpose of this is to establish a reliable set of reference cases, providing an accurate basis for subsequent dose adjustment calculations. From an overall perspective, the precise implementation of this screening process ensures the quality of the reference case set, enabling the reference dose values ​​calculated based on this set to accurately reflect the dose levels required for successful treatment under similar disease conditions. This provides high-quality training samples for the dose prediction model, ensuring that the dose adjustment values ​​output by the model are both consistent with clinical reality and individualized, ultimately improving the reliability and effectiveness of the entire radiotherapy planning method.

[0087] In another preferred embodiment of the present invention, step S2 further includes generating a final radiotherapy dose plan based on a standard dose value determined by clinical treatment guidelines that match the characteristics of the patient's three-dimensional tumor region if there is no hypoxic subregion within the three-dimensional tumor region.

[0088] It is understandable that radiotherapy needs to follow scientifically validated clinical standards, and clinical treatment guidelines are the concrete embodiment of this standardized treatment. Their advantage lies in systematically matching the tumor characteristics of individual patients with the guidelines, ensuring both the standardization of treatment plans and taking into account individual differences. The direct purpose of this is to establish a safe and effective dose base for the entire radiotherapy plan. From the perspective of the overall plan, the reasonable determination of the base dose provides a reliable benchmark for subsequent dose adjustments, enabling dose increases for hypoxic areas to be carried out within a safe range. This ensures effective treatment of drug-resistant areas and prevents the risk of complications caused by overtreatment. Ultimately, the goal of individualized and precise radiotherapy is achieved while adhering to the principles of standardized treatment.

[0089] It is understandable that when there is no hypoxic area inside the tumor, its radiosensitivity is comparable to that of normal tissue, and the standard dosage regimen can achieve the therapeutic effect. This avoids unnecessary calculation and analysis steps, improves the efficiency of the entire treatment plan generation process, and reduces the risks that may be caused by overtreatment. It optimizes the workflow while ensuring the therapeutic effect and improves the utilization efficiency of medical resources. It ensures the targeted nature of the treatment and avoids the waste of resources, ultimately achieving a balance between personalized treatment and efficient operation.

[0090] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for evaluating the therapeutic effect of brain tumor disease, characterized in that, Includes the following steps: S1. Acquire multimodal imaging data of the patient's brain, extract three-dimensional tumor regions and their corresponding perfusion parameters, and establish a tumor biomarker database; the multimodal imaging dataset includes structural magnetic resonance imaging data and perfusion-weighted imaging data; S2, perform voxel-level cerebral blood flow calculation and hypoxia region identification on the tumor biomarker database, and construct a hierarchical structure of biological target area containing the whole tumor region and hypoxia subregion based on the identification results; S3, in the biological target area hierarchy, screen for target cases with hypoxic sub-regions, extract the blood perfusion parameters of the hypoxic sub-regions and the corresponding parameters of the whole tumor area, and calculate the set of relative hypoxia degree parameters; S4, deconstruct historical treatment case data, identify differences between target cases and historical cases in terms of relative hypoxia degree parameters and initial treatment dose, and establish a dose-efficacy correlation parameter set by combining the long-term follow-up results of corresponding cases; The specific process for establishing a dose-efficacy correlation parameter set is as follows: Radiotherapy planning data and long-term follow-up data of historical brain tumor cases were collected. The radiotherapy planning data included the initial dose allocated to the three-dimensional tumor region and the relative degree of hypoxia in the hypoxic subregion for each case. The long-term follow-up data indicated whether local recurrence occurred in each case. The tumor volume of each historical case is calculated based on the three-dimensional tumor region. For each historical case marked as locally recurrent, the difference in relative hypoxia parameters and the tumor volume ratio between it and each non-recurrent historical case are calculated. A reference case set is selected from the non-recurrent historical cases based on the parameter difference and volume ratio. The statistical characteristic value of the initial dose of the reference case set is calculated as the reference dose value, and the dose adjustment value of the recurrent historical case is calculated. The dose adjustment value is the reference dose value minus the initial dose of the recurrent historical case. Using the relative hypoxia level and tumor volume of all relapsed cases as input features and the corresponding dose adjustment value as output label, the dose-efficacy correlation parameter set is constructed. The specific process for selecting a reference case set from historical cases that have never relapsed based on the parameter differences and volume ratios is as follows: Calculate the absolute value d of the difference in relative hypoxia degree parameter between each non-relapsed case and the current relapsed case; Calculate the tumor volume ratio r between each non-recurring case and the current recurring case; According to the calculation formula Calculate the candidate score S, where w1 and w2 are preset unit coefficients, and form a reference case set by including all non-relapsed cases whose candidate scores are higher than the preset score threshold. S5, construct the attribution mapping relationship between the dose-efficacy correlation parameter set and the clinically successful treatment dose, and train and output a dose adjustment strategy model for the hypoxic subregion; S6. Obtain the dose-efficacy correlation parameter set of the current patient's tumor and input it into the strategy model. Based on the dose adjustment value output by the model and combined with the baseline dose of the clinical guidelines, generate the final radiotherapy dose plan.

2. The method for evaluating the therapeutic effect of brain tumor disease according to claim 1, characterized in that, In S1, the specific process for determining the three-dimensional tumor region is as follows: The structural magnetic resonance imaging data is preprocessed, including noise filtering and intensity normalization. Based on the preprocessed structural magnetic resonance imaging data, a three-dimensional spatial coordinate system of the brain is established using a voxel mesh reconstruction method to generate a three-dimensional brain model. The three-dimensional brain model is divided into multiple regularly arranged three-dimensional grid units, and the average gray value of all voxels in each three-dimensional grid unit is calculated. The average gray value of each three-dimensional grid unit is compared with a preset gray value threshold range, and all three-dimensional grid units with average gray values ​​exceeding the gray value threshold range are marked as abnormal grid units. Spatially adjacent abnormal mesh cells are merged to form a continuous three-dimensional abnormal region. The boundary of the three-dimensional abnormal region is then optimized to obtain a three-dimensional tumor region.

3. The method for evaluating the therapeutic effect of brain tumor disease according to claim 2, characterized in that, In step S1, the specific process for obtaining the infusion parameters is as follows: The perfusion-weighted imaging data is spatially registered with the three-dimensional brain model, and the perfusion-weighted imaging data corresponding to the three-dimensional tumor region is extracted based on the registration result. The perfusion-weighted imaging data includes image sequences at multiple time points continuously acquired after contrast agent injection. The signal intensity of each voxel at different time points is obtained from the continuously acquired image sequence at multiple time points, forming the time-signal intensity curve of each voxel, and the time-signal intensity curve of each voxel is converted into the contrast agent concentration-time curve. The anterior cerebral artery region was selected as the reference region, and the average concentration-time curve of this region was extracted as the arterial input function. A single-compartment model based on tracer kinetics was adopted, and the deconvolution equation between the concentration-time curve of each voxel and the arterial input function was solved by singular value decomposition to obtain the quantified value of cerebral blood flow of each voxel.

4. The method for evaluating the therapeutic effect of brain tumor disease according to claim 3, characterized in that, The specific process of spatial registration is as follows: Feature point sets are extracted from the structural magnetic resonance imaging data and perfusion-weighted imaging data, respectively. The feature point sets include feature points of the midline structure of the brain, the outline of the ventricles, and the main sulci and gyri of the brain. The rigid registration algorithm based on mutual information performs preliminary alignment of two sets of feature points, and the free deformation registration algorithm based on B-splines performs nonlinear optimization of local details to establish a spatial correspondence between perfusion-weighted imaging data and the three-dimensional brain model.

5. The method for evaluating the therapeutic effect of brain tumor disease according to claim 1, characterized in that, In S2, the specific construction process of the biological target region hierarchy is as follows: Based on the quantified cerebral blood flow values ​​of each voxel within the three-dimensional tumor region, a set of continuous voxels with cerebral blood flow values ​​below a preset threshold is marked as a hypoxic subregion. The three-dimensional tumor region is used as the first-level target region, and the hypoxic subregion is used as the second-level target region contained within the first-level target region, thus establishing a hierarchical structure of biological target regions with spatial inclusion relationships.

6. The method for evaluating the therapeutic effect of brain tumor disease according to claim 1, characterized in that, In S3, the specific process for obtaining the relative hypoxia degree parameter set is as follows: The average cerebral blood flow value of all voxels within the hypoxic subregion is calculated as a first parameter; the average cerebral blood flow value of the remaining voxels within the three-dimensional tumor region excluding the hypoxic subregion is calculated as a second parameter; the ratio of the first parameter to the second parameter is calibrated as the relative hypoxia degree parameter of the hypoxic subregion; the relative hypoxia degree parameters of all target cases are collected to form the relative hypoxia degree parameter set.

7. The method for evaluating the therapeutic effect of brain tumor disease according to claim 1, characterized in that, In S5, the specific construction process of the dose adjustment strategy model is as follows: The dose-efficacy correlation parameter set is divided into a training set and a validation set. The relative hypoxia level and tumor volume of cases in the training set are used as input features, and the corresponding dose adjustment values ​​are used as prediction targets. A gradient boosting regression algorithm is employed to establish a nonlinear mapping relationship between the input features and the prediction targets through iterative optimization of decision tree combinations. During training, the model parameters are optimized by minimizing the mean squared error between the predicted dose adjustment value and the actual calculated dose adjustment value. The validation set is used to evaluate the generalization performance of the trained model, and cross-validation is used to calculate the model's average prediction accuracy. Finally, the model exhibiting the highest prediction accuracy on the validation set is selected as the trained dose adjustment strategy model.

8. The method for evaluating the therapeutic effect of brain tumor disease according to claim 1, characterized in that, S2 further includes generating a final radiotherapy dose plan based on the standard dose value determined by clinical treatment guidelines that match the characteristics of the patient's three-dimensional tumor region if there is no hypoxic subregion within the three-dimensional tumor region.

Images