Dosimetry parameter proliferation method

Abstract

The invention provides a dosimetry parameter proliferation method, which is characterized in that dose exposure which endangers an organ is reflected by assuming that a positive / negative case is represented as a high / low DVH parameter, positive and negative case data are expanded through interpolation according to the DVH parameter corresponding to the assumed positive / negative case, a richer and more balanced data set is constructed, and the method is applied to the proliferation of the dosimetry parameter. The method aims at improving the predictive ability and reliability of the model, clinical medical knowledge and experience are introduced into a data generation process, and particularly, a solution with more clinical significance is provided for the situation that positive samples are scarce. According to the method, the authenticity of the value-added data and the practicability of the prediction model can be remarkably improved, so that more powerful technical support is provided for the development of precise radiotherapy.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data proliferation, and in particular to a dosimetric parameter proliferation method. Background Art

[0002] Radiotherapy is a core method widely used in the treatment of malignant tumors. It uses high-energy rays to locally irradiate tumors to destroy tumor cells while protecting normal tissues from damage as much as possible. However, the effectiveness and safety of radiotherapy depend to a large extent on the accurate evaluation and prediction of dose distribution and biological effects. Dose-Volume Histogram (DVH), as an important quantitative tool, can be used to evaluate the dose distribution of treatment plans and predict treatment-related adverse reactions (such as radiation-induced lung injury, radiation-induced heart disease, etc.). Especially in recent years, DVH analysis has been widely used in studies to correlate dose distribution with specific treatment effects or complication rates.

[0003] Although the predictive role of DVH analysis in radiotherapy has been widely recognized, its clinical application still faces some significant challenges. The most prominent problems are the lack of positive case data and the imbalance of sample distribution. Taking radiation-induced lung injury (RILI) as an example, although this complication has a high potential incidence in clinical practice, the actual collection of severe cases (such as grade III or above lung injury) is extremely difficult. This is because the incidence of severe cases is relatively low, while mild or asymptomatic cases account for the vast majority. The scarcity of positive case samples not only affects model training and prediction accuracy, but also limits the promotion and application of DVH in complex clinical settings.

[0004] To address this problem, existing technologies attempt to introduce universal data augmentation methods to generate new data by transforming, interpolating or simulating existing data to increase the sample size, improve data diversity and improve model generalization. Such methods include random perturbations of the DVH curve, numerical scaling, linear interpolation or mixing of different samples to generate new samples; Monte Carlo simulation technology can also be used based on statistical distribution characteristics to generate "pseudo samples" that conform to the rules. In addition, in the field of medical imaging, geometric and spatial transformations (such as image rotation, scaling, translation, etc.) and generative adversarial networks (GANs) are also used to generate more new data that conform to distribution characteristics.

[0005] However, these methods are mostly based on mathematical algorithms or statistical modeling, which can only expand the amount of data to a certain extent, lacking relevance to clinical practice and integration with medical background. They often ignore the complex biological effects and individual differences of patients in radiotherapy, and the generated data may be far from the actual clinical situation, and may even amplify data deviations, resulting in poor performance of the prediction model in practical applications. Summary of the invention

[0006] The technical problem that the technical solution of the present invention needs to solve is that clinical data in radiotherapy is scarce, especially the number of positive cases (serious radiation side effects) is insufficient. The dosimetric parameter (DVH) of organs at risk in radiotherapy is an important basis for assessing the risk of side effects, but due to the low incidence of serious side effects, the number of positive samples obtained clinically is very limited. This data imbalance has a significant impact on the training and prediction performance of machine learning models, making it difficult to meet the needs of actual applications.

[0007] The technical solution of the present invention provides a dosimetric parameter proliferation method, comprising the following steps:

[0008] Obtain the patient's dosimetric parameters and the corresponding organ disease conditions after radiotherapy, and establish a dosimetric parameter relationship model for organ disease positive and organ disease negative according to the organ disease conditions after radiotherapy. Organ disease is used to describe the disease caused by radiation from radiotherapy equipment;

[0009] Establish that positive cases are manifested by higher patient dosimetric parameters, reflecting greater dose exposure to organs at risk, and negative cases are manifested by lower patient dosimetric parameters, reflecting less dose exposure to organs at risk;

[0010] For the current values ​​of patient dosimetric parameters corresponding to the positive cases in the dosimetric parameter relationship model, the positive medical weight is set according to medical experience to determine the interpolation interval, and interpolation is performed toward 100% to obtain the interpolated patient dosimetric parameters of the positive cases, so as to give priority to simulating high-dose distributions that are closer to the actual situation;

[0011] For the current values ​​of patient dosimetric parameters corresponding to negative cases in the dosimetric parameter relationship model, the negative medical weight is set according to medical experience to determine the interpolation interval, and interpolation is performed toward 0% to obtain the interpolated patient dosimetric parameters of the negative cases, so as to give priority to simulating a low-dose distribution that is closer to the actual situation;

[0012] The interpolated patient dosimetric parameters of positive cases and negative cases were combined with the dosimetric parameter relationship model to form an extended dosimetric parameter dataset for organ diseases, which was then used in machine learning algorithms to train organ disease prediction models and improve the accuracy of organ disease prediction.

[0013] Preferably, the positive case interpolation formula is as follows:

[0014] Vi 阳 _new=Vi 阳 _old+α 阳 ×(100%-Vi 阳 _old)

[0015] Among them, Vi 阳 _new is the patient dosimetric parameter after interpolation of positive cases, Vi 阳 _old is the patient dosimetric parameter before interpolation of positive cases, α 阳 Positive medical weights are set according to medical experience.

[0016] Preferably, the negative case interpolation formula is as follows:

[0017] Vi 阴 _new=Vi 阴 _old×(1-α 阴 )

[0018] Among them, Vi 阴 _new is the patient dosimetric parameter after interpolation of negative cases, Vi 阴 _old is the patient dosimetric parameter before interpolation of negative cases, α 阴 Negative medical weights are set according to medical experience.

[0019] The technical solution of the present invention proposes a dosimetric parameter proliferation method, which assumes that positive / negative cases show higher / lower DVH parameters, reflecting larger / smaller dose exposures of endangered organs, and expands the positive and negative case data by combining piecewise linear interpolation and medical experience weight interpolation according to the corresponding DVH parameters of the assumed positive / negative cases, so as to construct a richer and more balanced data set, aiming to improve the predictive ability and reliability of the model, introduce clinical medical knowledge and experience into the data generation process, and provide a more clinically meaningful solution for the situation where positive samples are scarce. This method can significantly improve the authenticity of value-added data and the practicality of prediction models, thereby providing more powerful technical support for the development of precision radiotherapy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a relationship model of dosimetric parameters of radiation-induced lung injury. DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0022] The first goal of the embodiment of the present invention is to expand the positive case data through data proliferation technology, so that the model can learn the characteristics of severe radiation side effects more comprehensively and improve its predictive ability for high-risk patients. This is particularly important for the prediction of side effects of radiation-threatening organs such as radiation pneumonia and radiation heart disease. By generating new data that conforms to medical experience through interpolation methods, the present invention effectively solves the problem of scarcity of positive case samples, making the training data more comprehensive and diversified, thereby enhancing the generalization performance of the model.

[0023] The second goal of the embodiment of the present invention is to further balance the data set by multiplying the data of negative cases. In clinical practice, a large amount of negative case (no serious side effects) data also contains important information. By multiplying the data of negative cases by interpolation, the relationship between low-dose parameters and low risk can be better captured, so that the model can accurately predict low-risk patients, thereby avoiding overtreatment and unnecessary adjustments, and optimizing the treatment plan.

[0024] The ultimate goal of the embodiments of the present invention is to provide an efficient and reliable risk assessment tool for clinical decision-making in radiotherapy. By improving data quality and model performance, the present invention can help doctors more accurately predict the risk of severe radiation side effects in patients, thereby optimizing radiotherapy plans, reducing the radiation dose to endangered organs, and reducing the risk of side effects in patients. This can not only improve the safety and effectiveness of radiotherapy, but also provide data support for the rational allocation of medical resources. In addition, the present invention can also provide a general technical method for the study of radiation damage to other endangered organs, and promote further development in the field of medical data modeling and analysis.

[0025] The embodiment of the present invention is specifically used to solve the difficulty of machine learning modeling caused by the scarcity of positive and negative cases in radiotherapy. By interpolating the dosimetric parameters (DVH-related data) of existing cases, clinically significant proliferation data is generated, thereby optimizing the performance of the prediction model and improving the prediction accuracy of radiotherapy side effects. The embodiment of the present invention provides a dosimetric parameter proliferation method, which specifically includes the following steps:

[0026] Step 1: Obtain the patient's dosimetric parameters and the corresponding organ disease conditions after radiotherapy, and establish a dosimetric parameter relationship model for organ disease positive and organ disease negative based on the organ disease conditions after radiotherapy. Organ disease is used to describe diseases caused by radiation from radiotherapy equipment, including radiation-induced lung injury and radiation-induced heart disease.

[0027] Step 2: Establish that positive cases are represented by higher patient dosimetric parameters, reflecting greater dose exposure to organs at risk, and negative cases are represented by lower patient dosimetric parameters, reflecting less dose exposure to organs at risk.

[0028] Step 3: For the current values ​​of patient dosimetric parameters corresponding to the positive cases in the dosimetric parameter relationship model, set the positive medical weight according to medical experience to determine the interpolation interval, interpolate to 100%, and obtain the interpolated patient dosimetric parameters of the positive cases, so as to give priority to simulating the high dose distribution that is closer to the actual situation. The interpolation formula for positive cases is as follows:

[0029] Vi 阳 _new=Vi 阳 _old+α 阳 ×(100%-Vi 阳 _old)

[0030] Among them, Vi 阳 _new is the patient dosimetric parameter after interpolation of positive cases, Vi 阳 _old is the patient dosimetric parameter before interpolation of positive cases, α 阳 is the positive medical weight set according to medical experience, α 阳 Usually it is 0.1 or 0.2. For example: Vi 阳 _old is V20 = 55%, α 阳 is 0.2, Vi 阳 _new is V20=55%+0.2×(100%-55%)=64%.

[0031] Step 4: For the current values ​​of patient dosimetric parameters corresponding to negative cases in the dosimetric parameter relationship model, set the negative medical weight according to medical experience to determine the interpolation interval, interpolate to 0%, and obtain the patient dosimetric parameters after interpolation of negative cases, so as to give priority to simulating low-dose distribution that is closer to the actual situation. The negative case interpolation formula is as follows:

[0032] Vi 阴 _new=Vi 阴 _old×(1-α 阴 )

[0033] Among them, Vi 阴 _new is the patient dosimetric parameter after interpolation of negative cases, Vi 阴_old is the patient dosimetric parameter before interpolation of negative cases, α 阴 is the negative medical weight set according to medical experience, α 阴 Usually [0,1]\alpha\in[0,1]. For example: Vi 阴 _old is V20 = 20%, α 阴 is 0.2, Vi 阴 _new is V20=20%×(1-0.2)=16%.

[0034] Step 5: Combine the interpolated patient dosimetric parameters of positive cases and negative cases with the dosimetric parameter relationship model to form an extended dosimetric parameter dataset for organ diseases. Using this dataset, apply machine learning algorithms (such as random forest or support vector machine) to train the organ disease prediction model to improve the accuracy of organ disease prediction.

[0035] The model is applied to clinical prediction:

[0036] Input new patient data and obtain the DVH parameters of the new patient, such as V20=40%, V10=60%, etc.

[0037] Model prediction: The patient’s DVH parameters are input into the trained machine learning model. The model outputs the prediction results (e.g., SPR=1 indicates high risk, SPR=0 indicates low risk) based on the rules learned from the proliferation dataset.

[0038] Clinical decision support, guiding doctors to optimize radiotherapy plans based on the predicted results. For example:

[0039] If the model predicts SPR = 1, the doctor can appropriately reduce the dose to the organ at risk and adjust the plan to reduce the risk of side effects.

[0040] If the model predicts SPR = 0, the current plan can be maintained to ensure the treatment effect.

[0041] The lung is one of the common organs at risk in chest radiotherapy, especially when the treatment target is close to the lung tissue, the dose distribution may cause radiation-induced lung injury (RILI). In order to evaluate and predict the risk of lung injury, DVH parameters such as V20, V30, etc. are of great significance to clinical radiotherapy planning. This embodiment hopes to expand the sample size of positive cases (severe lung injury) through the proliferation method, thereby optimizing the performance of the machine learning model in the prediction of radiotherapy side effects. Taking the organ disease as radiation-induced lung injury (RILI) as an example, the steps are as follows:

[0042] First, the dosimetric parameters (DVH data) of a group of radiotherapy patients were collected with the following radiotherapy OAR limits.

[0043] When V20 ≥ 30%, the probability of lung radiation damage increases significantly;

[0044] o When V30 ≥ 20%, the risk of radiation pneumonitis increases significantly.

[0045] Dosimetric parameters (DVH parameters) such as V3, V5, V8, V10, V13, V15, V20, V25, V30, V35, V40, V45, V50 (Gy), and whether each patient developed severe radiation lung injury (grade 3 and above, SPR) after radiotherapy. The results of SPR are expressed as a binary variable: positive radiation lung injury (SPR = 1, severe radiation lung injury) or negative radiation lung injury (SPR = 0, no or mild injury). Through these basic data, a radiation lung injury dosimetric parameter relationship model between dosimetric parameters and clinical response (positive radiation lung injury and negative radiation lung injury) can be established (such as Figure 1 shown).

[0046] For all positive cases (SPR = 1), based on the assumption of medical experience, a higher dose distribution is strongly positively correlated with a higher risk of adverse reactions to lung injury, and positive cases are set up to show higher patient dosimetric parameters, reflecting a greater dose exposure to organs at risk. For all negative cases (SPR = 0), it is assumed that a lower dose distribution is associated with a lower risk of adverse reactions to lung injury, and negative cases are set up to show lower patient dosimetric parameters, reflecting a smaller dose exposure to organs at risk.

[0047] For each dosimetric parameter of a positive case in the radiation lung injury dosimetric parameter relationship model, interpolation processing is performed from the current dose value to the maximum dose value (100%). Specifically:

[0048] o For each positive case dosimetric parameter (such as V20), start from the existing value (e.g. 52.2%), interpolate according to the proportion determined by the positive medical weight to generate a new dosimetric parameter. For example, interpolate once to get 76.1%. For higher dose DVH parameters such as V20 and V30, interpolate from the current value to 100%. For example, if V20 = 55%, V20 can be interpolated to 70% by the proliferation method.

[0049] oOther dosimetric parameters (such as V10, V30, etc.) are interpolated in the same way to generate a new set of parameter values.

[0050] o The proliferation data of this group were regarded as new cases (dosimetric parameters of patients with radiation-induced lung injury after interpolation of positive cases), and their SPR values ​​remained positive (SPR = 1).

[0051] Through the above method, the data of all positive cases are multiplied, and the sample size of positive cases can be expanded while maintaining the clinical significance of the original data.

[0052] For each dosimetric parameter of a negative case (SPR=0) in the dosimetric parameter relationship model of radiation lung injury, an interpolation process is performed from the current dose value to the minimum dose value (0%).

[0053] For all negative cases, interpolation is performed for each dosimetric parameter, starting from 0% and extending to the current dose value range. Specifically:

[0054] o For each negative case dosimetric parameter (such as V20), interpolate from 0% to the current value (such as 38.5%), and generate new dosimetric parameters according to the proportion determined by the negative medical weight, for example, interpolate once to get 19.3%, and interpolate from 0% to the current value for lower dose DVH parameters such as V20 and V30. For example, if V20 = 20%, the interpolation value is 15%.

[0055] o Other dosimetric parameters (such as V10, V30, etc.) are also interpolated in the same way to generate a new set of parameter values.

[0056] o The proliferation data of this group were regarded as new cases (dosimetric parameters of patients with radiation-induced lung injury after interpolation of negative cases), and their SPR values ​​remained negative (SPR = 0).

[0057] By using the above method, the data of all negative cases are multiplied, which can expand the sample size of negative cases and simulate more cases that conform to clinical laws.

[0058] The dosimetric parameters of patients with radiation-induced lung injury after interpolation of positive cases and negative cases were combined with the relationship model of radiation-induced lung injury dosimetric parameters to form an extended dosimetric parameter dataset for radiation-induced lung injury. Using this dataset, machine learning algorithms (such as random forest or support vector machine) were applied to train the radiation-induced lung injury prediction model to improve the accuracy of radiation-induced lung injury prediction.

[0059] The model trained using the extended dosimetry parameter dataset for radiation-induced lung injury was validated in the clinic to ensure its predictive ability in practical applications. The model output is used to guide doctors to adjust radiotherapy plans to reduce the risk of lung injury.

[0060] The heart is an important organ at risk in chest radiotherapy. Excessive radiation doses can lead to cardiac lesions, such as coronary artery disease, cardiac fibrosis, etc. In order to assess the risk of cardiac damage, this embodiment uses a data proliferation method to generate more proliferation data, combined with reasonable OAR dose limits to improve the prediction accuracy of cardiac lesions. Taking radiation-induced cardiac lesions (RICO) as an example, the steps are as follows:

[0061] First, the dosimetric parameters (DVH data) of a group of patients receiving thoracic radiotherapy were collected with the following radiotherapy OAR limits.

[0062] o When V30 ≥ 20%, the risk of cardiac injury increases significantly;

[0063] o When V40 ≥ 10%, the risk of long-term heart problems (such as heart failure, coronary heart disease, etc.) increases. Dosimetric parameters (DVH parameters) such as V5, V10, V20, V30, V40 (Gy), etc., and whether each patient has severe radiation-induced heart disease (grade 3 and above, SPR) after radiotherapy are recorded. The results of SPR are expressed as a binary variable: positive radiation-induced heart disease (SPR = 1, severe radiation-induced heart disease) or negative radiation-induced heart disease (SPR = 0, no heart damage). Through these basic data, a relationship model of radiation-induced heart disease dosimetric parameters and clinical responses (positive radiation-induced heart disease and negative radiation-induced heart disease) can be established.

[0064] For all positive cases (SPR = 1), based on the assumption of medical experience, a higher dose distribution is strongly positively correlated with a higher risk of adverse reactions to cardiac injury, and positive cases are set up with higher patient dosimetric parameters, reflecting a greater dose exposure to organs at risk. For all negative cases (SPR = 0), it is assumed that a lower dose distribution is associated with a lower risk of adverse reactions to cardiac injury, and negative cases are set up with lower patient dosimetric parameters, reflecting a smaller dose exposure to organs at risk.

[0065] For each dosimetric parameter of a positive case in the dosimetric parameter relationship model of radiation heart disease, an interpolation process is performed from the current dose value to the maximum dose value (100%). Specifically:

[0066] o For higher dose DVH parameters such as V30 and V40, interpolate from the current value to a higher dose range, for example, interpolate V30=30% to 40%.

[0067] oThe interpolation interval was set to 0.2 to simulate the occurrence of cardiac damage under high-dose conditions.

[0068] o The proliferation data of this group are regarded as new cases (dosimetric parameters of patients with radiation-induced heart disease after interpolation of positive cases), and their SPR values ​​remain positive (SPR = 1).

[0069] Through the above method, the data of all positive cases are multiplied, and the sample size of positive cases can be expanded while maintaining the clinical significance of the original data.

[0070] For each dosimetric parameter of a negative case (SPR=0) in the dosimetric parameter relationship model of radiation heart disease, an interpolation process is performed in the range extending from the current dose value to the minimum dose value (0%).

[0071] For all negative cases, interpolation is performed for each dosimetric parameter, starting from 0% and extending to the current dose value range. Specifically:

[0072] o For lower dose DVH parameters such as V30, V40, etc., interpolate from 0% to the current value, for example, interpolate V30=10% to 15%.

[0073] The interpolation interval is set to 0.1 to simulate the situation where the heart is not damaged under low dose distribution.

[0074] o The proliferation data of this group were regarded as new cases (dosimetric parameters of patients with radiation-induced heart disease after interpolation of negative cases), and their SPR values ​​remained negative (SPR = 0).

[0075] By using the above method, the data of all negative cases are multiplied, which can expand the sample size of negative cases and simulate more cases that conform to clinical laws.

[0076] The dosimetric parameters of patients with radiation-induced heart disease after interpolation of positive cases and the dosimetric parameters of patients with radiation-induced heart disease after interpolation of negative cases are combined with the relationship model of dosimetric parameters of radiation-induced heart disease to form an extended dosimetric parameter dataset of radiation-induced heart disease. Using this dataset, machine learning algorithms (such as random forest or support vector machine) are applied to train the prediction model of radiation-induced heart disease to improve the accuracy of radiation-induced heart disease prediction.

[0077] The model trained using the extended dosimetry parameter dataset of radiation-induced heart disease was validated in the clinic to ensure its predictive ability in practical applications. The model output is used to guide doctors to adjust radiotherapy plans to reduce the risk of radiation-induced heart disease.

[0078] The embodiments of the present invention are used for data processing and modeling in radiotherapy, and are particularly suitable for the evaluation and prediction of the dose to organs at risk (DVH) and the risk of adverse reactions to radiotherapy. The data proliferation method is used to solve the problem of scarcity of clinical positive cases and improve the prediction performance of machine learning models. The embodiments of the present invention can be widely used for risk prediction of various organ-at-risk injuries such as radiation pneumonia and radiation heart disease, helping doctors optimize radiotherapy plans, reduce the risk of adverse reactions, and improve treatment effects.

[0079] The advantages of the embodiments of the present invention are as follows:

[0080] 1. Combine medical experience to generate data with stronger clinical significance

[0081] Based on the medical experience of dose and radiotherapy side effects, the present invention designs a differentiated proliferation method for positive and negative cases, and the generated data is more in line with the actual clinical situation. Compared with the traditional proliferation method of simple mathematical interpolation or statistical simulation, the generated data of the present invention is more credible in biological significance and clinical usability, providing high-quality input for the construction of radiotherapy-related models.

[0082] 2. Effectively solve the problem of scarcity of positive cases

[0083] It is difficult to collect positive cases (such as severe radiation pneumonia cases) during radiotherapy, but this type of data is crucial to the construction of prediction models. The present invention effectively expands the positive case data and enhances the model's learning ability for positive features through a high-dose interpolation strategy based on positive cases, thereby significantly improving the prediction accuracy of radiotherapy side effects.

[0084] 3. Improve data diversity and optimize machine learning models

[0085] By interpolating and multiplying positive and negative cases respectively, the present invention significantly expands the coverage of the data set and increases the diversity of the data. This design not only alleviates the problem of sample imbalance, but also improves the generalization ability of the machine learning model, especially in the prediction of extreme cases or borderline cases in clinical practice.

[0086] 4. The interpolation strategy is consistent with the dosimetry law to ensure the reliability of the generated data

[0087] According to the medical law that positive cases are more likely to produce side effects with high doses and negative cases are less likely to produce side effects with low doses, the present invention designs a reasonable interpolation range and direction, so that the generated proliferation data is logically self-consistent and has clear rules, ensuring that the data will not introduce erroneous deviations in model training.

[0088] 5. Reduce data collection costs and improve research efficiency

[0089] By utilizing existing case data for interpolation and proliferation, the present invention reduces the need for collecting a large number of cases in clinical practice. In particular, when there are few positive cases or severe cases are difficult to obtain, high-quality data simulating real situations can be quickly generated, thereby reducing research costs and significantly improving research efficiency.

[0090] 6. Applicable to the prediction of adverse reactions to multiple organs at risk

[0091] The method of the present invention is not only applicable to the proliferation of radiation damage data of the lung, but can also be extended to the prediction of adverse reactions of other organs at risk (such as heart, kidney, etc.). By adjusting relevant dosimetric parameters and interpolation strategies, the present invention has strong versatility and high adaptability, and can provide support for different types of radiotherapy adverse reaction research.

[0092] 7. Improve the ability of prediction models to guide clinical decision making

[0093] The present invention optimizes the performance of the prediction model by generating more high-quality proliferation data, so that it can more accurately predict the risk of radiotherapy side effects in practical clinical applications, provide a more reliable scientific basis for the formulation and adjustment of radiotherapy plans, and further improve the safety and effectiveness of radiotherapy.

[0094] 8. Break through the limitations of traditional data proliferation methods

[0095] Traditional data proliferation methods are mostly based on simple random perturbations or geometric transformations, ignoring the medical background and individual differences of patients. This invention combines clinical experience with proliferation strategies, truly realizing data expansion based on medical laws, and the generated data is more targeted and practical, breaking through the limitations of existing technologies.

[0096] In summary, the present invention has significant advantages in terms of the clinical significance of data generation, performance improvement of prediction models, wide application scope and optimization of research costs, and provides an innovative and effective solution for the research and practice of radiotherapy-related side effects.

Claims

1. A dosimetric parameter multiplication method, characterized in that: The following steps are involved: Obtain the patient's dosimetric parameters and the corresponding organ disease conditions after radiotherapy, and establish a dosimetric parameter relationship model for organ disease positive and organ disease negative according to the organ disease conditions after radiotherapy. Organ disease is used to describe the disease caused by radiation from radiotherapy equipment; Establish that positive cases manifest as higher patient dosimetric parameters, reflecting greater dose exposure to organs at risk, and negative cases manifest as lower patient dosimetric parameters, reflecting less dose exposure to organs at risk; For the current values ​​of patient dosimetric parameters corresponding to the positive cases in the dosimetric parameter relationship model, the positive medical weights are set according to medical experience to determine the interpolation interval, and interpolation is performed toward 100% to obtain the interpolated patient dosimetric parameters of the positive cases, so as to give priority to simulating high-dose distributions that are closer to the actual situation; For the current values ​​of patient dosimetric parameters corresponding to negative cases in the dosimetric parameter relationship model, the negative medical weight is set according to medical experience to determine the interpolation interval, and interpolation is performed toward 0% to obtain the interpolated patient dosimetric parameters of the negative cases, so as to give priority to simulating low-dose distributions that are closer to the actual situation; The interpolated patient dosimetric parameters of positive cases and negative cases were combined with the dosimetric parameter relationship model to form an extended dosimetric parameter dataset for organ diseases, which was then used in machine learning algorithms to train organ disease prediction models and improve the accuracy of organ disease prediction.

2. A dosimetric parameter multiplication method as claimed in claim 1, characterized in that: The positive case interpolation formula is as follows: We 阳 _new=We 阳 _old+α 阳 ×(100%-We 阳 _old) Among them, Vi 阳 _new is the patient dosimetric parameter after interpolation of positive cases, Vi 阳 _old is the patient dosimetric parameter before interpolation of positive cases, α 阳 Positive medical weights are set according to medical experience.

3. A dosimetric parameter multiplication method as claimed in claim 1, characterized in that: The interpolation formula for negative cases is as follows: You 阴 _new=You 阴 _old×(1-α 阴 ) Among them, Vi 阴 _new is the patient dosimetric parameter after interpolation of negative cases, Vi 阴 _old is the patient dosimetric parameter before interpolation of negative cases, α 阴 Negative medical weights are set according to medical experience.

Images