A method and system for symptom management and prognosis assessment of tumor radiotherapy patients

Abstract

The present invention discloses a method and system for symptom management and prognosis evaluation of tumor radiotherapy patients, which relates to the technical field of tumor radiotherapy patient management, and includes: obtaining the basic information of tumor radiotherapy patients, performing prognosis evaluation according to the basic information, generating a first result, and judging the risk category according to the first result; obtaining the symptom information of tumor radiotherapy patients, evaluating the severity of symptoms according to the symptom information, and generating a second result; performing symptom management according to the second result and the risk category, and simultaneously generating prognosis data, and updating the first result through the prognosis data. By integrating a dual contrast learning network and a dual-path decision tree integration model, the present invention realizes the dynamic correlation analysis and adaptive feature enhancement of multi-dimensional medical data of tumor radiotherapy patients, thereby providing a precise prognosis evaluation and personalized symptom management plan for clinical practice.

Description

Technical Field

[0001] The present invention relates to the technical field of tumor radiotherapy patient management, and specifically provides a method and system for symptom management and prognosis evaluation of tumor radiotherapy patients. Background Art

[0002] With the continuous development of tumor radiotherapy technology, the wide application of new treatment methods such as precise radiotherapy and intensity-modulated radiotherapy has greatly improved the treatment effect of radiotherapy. However, as a local treatment method, radiotherapy inevitably causes damage to normal tissues while killing tumor cells, leading to various radiotherapy-related adverse reactions. Currently, the clinical symptom management of radiotherapy patients mainly relies on doctors' experience judgment, and the treatment plan is adjusted by observing the symptoms of patients. This traditional symptom management method has obvious limitations: firstly, the symptom assessment lacks standardized and quantitative indicators, making it difficult to accurately judge the severity of symptoms; secondly, relying solely on experience is difficult to effectively predict the prognosis risk of patients and cannot achieve timely and precise adjustment of the treatment plan; thirdly, existing prognosis assessment models often process information such as patients' basic characteristics, symptom manifestations, and treatment responses separately, failing to fully explore the correlation between multi-dimensional medical data, resulting in limited reliability and accuracy of the prediction results.

[0003] In addition, in recent years, artificial intelligence technology has been deeply applied in the medical field, and prognosis assessment methods based on machine learning have gradually become a research hotspot. Existing technologies usually use shallow neural networks or traditional tree models for prognosis prediction. These methods have the following problems when dealing with heterogeneous medical data: first, feature engineering overly relies on manual experience and is difficult to adaptively discover and enhance feature combinations valuable for prognosis prediction; second, the model structure is relatively simple and cannot effectively capture complex temporal dependence relationships and feature interaction patterns in medical data; third, the prediction results lack interpretability and are difficult to provide a reliable quantitative basis for clinical decision-making; fourth, the model update mechanism is not flexible enough to dynamically adjust the prediction strategy according to the symptom changes during the patient's treatment process. Therefore, how to design a technical solution that can comprehensively utilize patients' basic characteristics, symptom information, and treatment responses to achieve precise prognosis assessment and intelligent symptom management has become an urgent technical problem to be solved. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for symptom management and prognosis evaluation of tumor radiotherapy patients, which can solve the problems mentioned in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: A method for symptom management and prognosis evaluation of tumor radiotherapy patients, comprising: obtaining the basic information of tumor radiotherapy patients, performing prognosis evaluation based on the basic information to generate a first result, and judging the risk category according to the first result; obtaining the symptom information of tumor radiotherapy patients, evaluating the severity of symptoms according to the symptom information to generate a second result; performing symptom management according to the second result and the risk category, and simultaneously generating prognosis data, and updating the first result through the prognosis data. Among them, prognosis evaluation is performed based on the basic information to generate a first result. The following steps are included: constructing a patient feature matrix, respectively performing standardization processing and encoding conversion on the numerical features and categorical features in the basic information to generate a standardized feature matrix; constructing a dual contrast learning network to enhance the features of the standardized feature matrix to generate an enhanced feature matrix; based on the enhanced feature matrix, performing prognosis evaluation modeling through an ensemble tree model and outputting a first result.

[0007] As a preferred embodiment of the method for symptom management and prognosis evaluation of tumor radiotherapy patients according to the present invention, wherein: the basic information includes the demographic characteristics, disease-related information, treatment-related information, past medical history, and laboratory test results of tumor radiotherapy patients; the dual contrast learning network includes a global contrast learning module and a local contrast learning module.

[0008] As a preferred embodiment of the method for symptom management and prognosis evaluation of tumor radiotherapy patients according to the present invention, wherein: constructing a patient feature matrix, respectively performing standardization processing and encoding conversion on the numerical features and categorical features in the basic information to generate a standardized feature matrix, includes the following steps: constructing an original feature matrix, and dividing the features in the original feature matrix into numerical features and categorical features; performing standardization processing based on time decay on the numerical features, and performing weighted target encoding conversion on the categorical features; recombining the standardized numerical feature values and the encoded categorical feature values according to the corresponding positions of the original features to generate a standardized feature matrix.

[0009] Construct a dual contrast learning network to enhance the features of the standardized feature matrix and generate an enhanced feature matrix, including the following steps: Calculate the Pearson correlation coefficient between every two features in the standardized feature matrix to generate a feature correlation coefficient matrix; Traverse all the correlation coefficients in the feature correlation coefficient matrix, and divide the features into the first type of features and the second type of features according to the absolute value of the correlation coefficient; Among them, the condition for satisfying the first type of features is that in the feature correlation coefficient matrix, there is at least one feature with an absolute value of the correlation coefficient greater than the first preset threshold; The condition for satisfying the second type of features is that in the feature correlation coefficient matrix, the absolute values of all correlation coefficients are less than or equal to the first preset threshold; Input the first type of features into the global contrast learning module to generate a global feature representation matrix; Input the second type of features into the local contrast learning module to generate a local feature representation matrix; Take the ratio of the number of the first type of features to the total number of features as the global feature weight; Take the ratio of the number of the second type of features to the total number of features as the local feature weight; According to the global feature representation matrix, the local feature representation matrix, the global feature weight, and the local feature weight, generate an enhanced feature matrix through feature concatenation operation.

[0010] Based on the enhanced feature matrix, perform prognostic evaluation modeling through an ensemble tree model and output a first result, including the following steps: For each enhanced feature in the enhanced feature matrix, construct a dual-path decision tree ensemble model according to the corresponding values of each enhanced feature in the global feature representation matrix and the local feature representation matrix, and the global feature weight and the local feature weight: Calculate the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix, and the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix; Calculate the total variance, within-group variance, and between-group variance of each enhanced feature in the enhanced feature matrix; Compare the ratio of the within-group variance to the total variance with a second preset threshold to determine whether each enhanced feature value is in the feature stability interval; Perform a classification judgment on each enhanced feature in the enhanced feature matrix: If the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix is greater than the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix, and the enhanced feature value is within the feature stability interval, then use a depth-first decision tree for prediction; If the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix is greater than the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix, or the enhanced feature value exceeds the feature stability interval, then use a breadth-first decision tree for prediction.

[0011] Calculate the weight coefficients of the depth - first decision tree and the breadth - first decision tree, multiply the prediction result of the depth - first decision tree by the weight coefficient of the depth - first decision tree, and add the product of the prediction result of the breadth - first decision tree and the weight coefficient of the breadth - first decision tree to obtain the predicted value of the survival probability, the predicted value of the local recurrence risk, the predicted value of the distant metastasis risk, and the predicted value of the complication occurrence risk respectively, which form the first result.

[0012] As a preferred embodiment of the method for symptom management and prognosis assessment of tumor radiotherapy patients according to the present invention, wherein: for each predicted value in the first result, the credibility calculation rule is as follows: calculate the confidence interval stability score of each predicted value; calculate the similarity score between the predicted distribution and the true distribution of each predicted value; according to the comparison result of the confidence interval stability score and the third preset threshold, and the comparison results of the similarity score with the fourth preset threshold and the fifth preset threshold, divide the credibility into four levels: when the confidence interval stability score is higher than the third preset threshold and the similarity score is higher than the fifth preset threshold, the credibility is level 3; when the confidence interval stability score is higher than the third preset threshold and the similarity score is higher than the fourth preset threshold but not higher than the fifth preset threshold, the credibility is level 2; when the confidence interval stability score is higher than the third preset threshold, the credibility is level 1; in other cases, the credibility is level 0.

[0013] As a preferred embodiment of the method for symptom management and prognosis assessment of tumor radiotherapy patients according to the present invention, wherein: judging the risk category according to the first result includes the following steps: calculate the comprehensive score of the predicted value of the survival probability; calculate the comprehensive score of the risk events based on the predicted value of the local recurrence risk, the predicted value of the distant metastasis risk, and the predicted value of the complication occurrence risk; perform a weighted combination of the comprehensive score of the predicted value of the survival probability and the comprehensive score of the risk events, and combine with the credibility to calculate the final risk classification score, and classify the tumor radiotherapy patients into three categories: low - risk, medium - risk, and high - risk: when the final risk classification score is greater than or equal to the sixth preset threshold and the credibility is greater than or equal to level 2, it is determined as low - risk; when the final risk classification score is greater than or equal to the seventh preset threshold and less than the sixth preset threshold, or the credibility is level 1, it is determined as medium - risk; when the final risk classification score is less than the seventh preset threshold, or the credibility is 0, it is determined as high - risk.

[0014] As a preferred embodiment of the method for symptom management and prognosis assessment of tumor radiotherapy patients according to the present invention, wherein: the symptom information includes the symptom name, the symptom occurrence time, the symptom duration, the symptom location, and the symptom manifestation; according to the symptom information, evaluate the symptom severity with reference to the radiotherapy acute reaction grading standard to generate a second result including the symptom severity level.

[0015] As a preferred solution of the method for symptom management and prognosis evaluation of tumor radiotherapy patients according to the present invention, wherein: symptom management is performed according to the second result and the risk category, and prognosis data is generated at the same time. The first result is updated through the prognosis data, including the following steps: according to the symptom severity level in the second result and the risk category, extract the corresponding radiotherapy adjustment plan and nursing intervention measures from the supporting symptom management database; obtain each treatment record collected in the radiotherapy treatment room and the symptom severity level, and generate prognosis data including the symptom change trend; the treatment record includes the actual irradiation dose, the cumulative irradiation dose, the symptom duration, and the complication record; perform standardization processing on the prognosis data, and if there are missing data in the prognosis data, fill in the missing values; input the standardized prognosis data into the dual contrast learning network. If the similarity between the obtained new global feature representation matrix and the new local feature representation matrix is lower than the similarity of the corresponding matrix in the enhanced feature matrix, recalculate the feature weights and perform feature contrast learning, and input the optimized enhanced feature matrix into the dual-path decision tree integration model to update the survival probability prediction value, local recurrence risk prediction value, distant metastasis risk prediction value, and complication occurrence risk prediction value in the first result.

[0016] To further solve the above technical problems, the present invention provides the following technical solution: A symptom management and prognosis evaluation system for tumor radiotherapy patients, including: a prognosis evaluation module, configured to obtain the basic information of tumor radiotherapy patients, perform prognosis evaluation according to the basic information, generate a first result, and judge the risk category according to the first result; a symptom evaluation module, configured to obtain the symptom information of tumor radiotherapy patients, evaluate the symptom severity according to the symptom information, and generate a second result; a dynamic management module, configured to perform symptom management according to the second result and the risk category, and generate prognosis data at the same time, and update the first result through the prognosis data.

[0017] A computer device, including a memory and a processor, the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method for symptom management and prognosis evaluation of tumor radiotherapy patients as described above are implemented.

[0018] A computer-readable storage medium, on which a computer program is stored, and is characterized in that when the computer program is executed by a processor, the steps of the method for symptom management and prognosis evaluation of tumor radiotherapy patients as described above are implemented.

[0019] Advantages of the present invention: By constructing an innovative architecture integrating a dual contrast learning network and a dual-path decision tree ensemble model, the present invention realizes the dynamic correlation analysis and adaptive feature enhancement of multi-dimensional medical data of tumor radiotherapy patients. The present invention can not only accurately quantify the prognosis risk of patients, but also adjust personalized treatment plans based on symptom severity grading, and has the following significant advantages: First, by adopting time-decay-based standardization processing and weighted target coding conversion, the feature expression ability of heterogeneous medical data is improved; Second, through the synergistic effect of global and local contrast learning, the semantic correlation between features is enhanced, and the generalization performance of the model is improved; Third, based on the ensemble strategy of the dual-path decision tree and the multi-level credibility evaluation mechanism, the accuracy and reliability of the prediction results are significantly improved; Finally, the dynamic prognosis data update mechanism enables the model to adaptively adjust feature weights, realizes the continuous optimization of symptom management, and provides more scientific and accurate decision-making support for clinical practice. Brief Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 It is the overall flowchart of a method for symptom management and prognosis evaluation of tumor radiotherapy patients proposed by the present invention;

[0022] Figure 2 It is the computer equipment diagram of a method for symptom management and prognosis evaluation of tumor radiotherapy patients proposed by the present invention. Detailed Embodiments

[0023] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0025] Example 1, refer to Figure 1, which is an embodiment of the present invention, provides a method for symptom management and prognosis assessment of tumor radiotherapy patients.

[0026] In the relevant technologies, clinical symptom management of radiotherapy patients mainly relies on doctors' experience and judgment. This traditional method has limitations such as lack of standardization of symptom assessment, inaccurate prediction of prognosis risk, and insufficient mining of multi-dimensional medical data correlation. Although artificial intelligence technology has been widely used in the medical field in recent years, the existing machine learning methods still face problems such as feature engineering relying on manual experience, simple model structure, poor interpretability of prediction results, and inflexible update mechanism when processing heterogeneous medical data. Therefore, developing a technical solution that can comprehensively utilize multi-dimensional patient information to achieve accurate prognosis assessment and intelligent symptom management has become a clinical problem that needs to be solved urgently.

[0027] The present application provides an effective solution to the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to implement the symptom management and prognosis assessment method for tumor radiotherapy patients.

[0028] Figure 1 The overall flow chart of a method for symptom management and prognosis assessment of tumor radiotherapy patients is shown, including:

[0029] S1: obtaining basic information of a patient undergoing tumor radiotherapy, performing a prognosis assessment based on the basic information, generating a first result, and determining a risk category based on the first result;

[0030] S2: Acquire symptom information of patients undergoing tumor radiotherapy, evaluate the severity of the symptoms according to the symptom information, and generate a second result;

[0031] S3: Perform symptom management according to the second result and the risk category, generate prognostic data, and update the first result through the prognostic data.

[0032] Next, this embodiment will elaborate on S1 to S3 one by one:

[0033] S1: Obtain basic information of patients undergoing tumor radiotherapy, conduct prognosis assessment based on the basic information, generate a first result, and determine the risk category based on the first result.

[0034] S1.1 Obtain basic information of patients undergoing tumor radiotherapy.

[0035] Specifically, in one embodiment, the basic information of tumor radiotherapy patients may include: the patient's demographic characteristics (such as age, gender, height, weight, etc.), disease-related information (such as tumor type, stage, pathological classification, tumor size and location, etc.), treatment-related information (such as radiotherapy plan, cumulative dose, fractionated dose, target area range, etc.), past medical history (such as underlying diseases, surgical history, medication history, etc.), lifestyle (such as smoking and drinking history, eating habits, exercise habits, etc.), and laboratory test results (such as blood routine, biochemical indicators, tumor markers, etc.). These basic information can be collected through various channels and methods: Firstly, utilize the existing hospital information systems. For example, the Hospital Information System (HIS) can provide the patient's medical records, prescription information, etc., and the Electronic Medical Record System (EMR) contains detailed records of the diagnosis and treatment process. Secondly, through specially designed information collection forms, which should have standardized data item settings, support structured information entry, and be able to interface with the existing hospital systems for data. In addition, information can also be collected through the information collection interface of the mobile terminal APP or the web page, facilitating medical staff to supplement and update information at any time. All the collected information needs to undergo standardized processing and quality control to ensure the accuracy and integrity of the data. These processed basic data will provide comprehensive and reliable data support for subsequent prognosis assessment and risk classification.

[0036] S1.2: Conduct prognosis assessment based on the basic information to generate the first result.

[0037] S1.2.1: Construct a patient feature matrix, standardize the numerical features and perform encoding conversion on the categorical features in the patient's basic information respectively to generate a standardized feature matrix A.

[0038] Specifically, first construct the original feature matrix X of the patient's basic information, which includes a demographic feature vector, a disease-related feature vector, a treatment-related feature vector, a past medical history feature vector, a lifestyle feature vector, and a laboratory test feature vector:

[0039] ;

[0040] where m is the number of patients to be evaluated currently; n is the number of features; x ij is the j-th feature value of the i-th patient to be evaluated.

[0041] The features in the original feature matrix X are divided into numerical features and categorical features.

[0042] It should be noted that numerical features include: age, height, and weight in the demographic feature vector, tumor size and tumor location coordinates in the disease-related feature vector, cumulative radiation dose, fractionated dose, and target area range values in the treatment-related feature vector, as well as blood routine test values, biochemical index values, and tumor marker test values in the laboratory test feature vector. Categorical features include: gender in the demographic feature vector, tumor type, stage, and pathological classification in the disease-related feature vector, underlying disease type, surgical history type, and medication history type in the past medical history feature vector, and smoking and drinking type, eating habit type, and exercise habit type in the lifestyle feature vector.

[0043] Then, perform Z-score standardization on the numerical features in the original feature matrix X based on time decay:

[0044] ;

[0045] where, is the standardized value of the j-th feature; is the original value of the j-th feature of the patient to be evaluated; is the mean value of the j-th feature of the patient to be evaluated; is the standard deviation of the j-th feature of the historical patients; is the current time point; is the data collection time point of the historical patients; is the time decay coefficient, and its value range is [0.1, 0.5].

[0046] Perform weighted target encoding conversion on the categorical features in the original feature matrix X:

[0047] ;

[0048] where, is the encoded value of category c, is the target variable value of the r-th historical patient in category c, is the number of historical patients in category c, is the time weight coefficient of the r-th historical patient, which is calculated by the following formula:

[0049] ;

[0050] where, is the time weight decay coefficient, and its value range is [0.2, 0.6]; is the data collection time point of the r-th historical patient.

[0051] Finally, recombine the standardized numerical feature values and the encoded categorical feature values after the encoding conversion according to the corresponding positions of the original features to generate a standardized feature matrix A:

[0052] ;

[0053] wherein, is the j-th standardized feature value of the i-th patient to be evaluated. If the j-th feature is a numerical feature, then ; if the j-th feature is a categorical feature, then .

[0054] It should be noted that in the practical application of prognostic evaluation of tumor radiotherapy patients, first, the existing technology usually uses simple standardization or normalization methods to process numerical features, ignoring the timeliness characteristics of medical data, resulting in the same influence weight of new and old data on the model and being unable to reflect the dynamic change characteristics of the patient's state; second, for categorical features, traditional one-hot encoding or label encoding methods cannot effectively utilize the inherent association information between categories. Especially when dealing with categorical features with medical professional knowledge connotations such as disease classification and pathological grading, information loss often occurs; third, existing feature processing methods often process different types of features (such as demographic features, disease-related features, treatment-related features, etc.) separately and fail to fully consider the interaction relationship between features. To address these problems, the present invention introduces a time-weighted Z-score standardization method, which not only maintains the numerical distribution characteristics of the features but also assigns higher weights to recent data through an exponential decay function to more accurately reflect the current state of the patient; at the same time, the weighted target encoding conversion method provided by the present invention encodes categorical features by using the prognostic information of historical patients, which not only retains the medical semantic information of the categories but also introduces time weights to balance the influence of new and old cases. This improvement brings three significant effects: First, it improves the sensitivity of the model to changes in the patient's state, and the prediction accuracy is improved compared with traditional methods; second, through the time-weighting mechanism, the prediction results are more in line with the empirical cognition that "newer cases have higher reference value" in clinical practice, improving the reliability of the model in practical applications; third, the weighted target encoding method enables the prognostic information carried by categorical features to be fully utilized.

[0055] S1.2.2: Construct a dual contrast learning network to enhance the features of the standardized feature matrix A and generate an enhanced feature matrix B.

[0056] The dual contrast learning network includes a global contrast learning module and a local contrast learning module.

[0057] Among them, the loss function of the global contrast learning module is:

[0058] ;

[0059] Among them, is the feature representation of the i-th patient to be evaluated; is the feature representation corresponding to the positive-matched historical patient of the i-th patient to be evaluated; is the feature representation of the v-th negative-matched historical patient; is the cosine similarity function; is the global temperature parameter, and its value range is [0.1, 1.0]; N is the number of negative-matched historical patients.

[0060] Among them, the loss function of the local contrast learning module is:

[0061] ;

[0062] Among them, is the local feature representation of the patient to be evaluated; is the local feature representation of the w-th neighboring historical patient; is the local feature representation of the z-th non-neighboring historical patient; K is the number of neighboring historical patients; M is the number of non-neighboring historical patients; is the local temperature parameter, and its value range is [0.1, 1.0]; the value range of the preset threshold T1 is [0.3, 0.7].

[0063] Specifically, calculate the Pearson correlation coefficient between every two features in the standardized feature matrix A to generate a feature correlation coefficient matrix.

[0064] Traverse all the correlation coefficients in the feature correlation coefficient matrix, and classify the features into the first type of features and the second type of features according to the absolute value of the correlation coefficient. Among them, the condition for satisfying the first type of features is that in the feature correlation coefficient matrix, there is at least one feature whose absolute value of the correlation coefficient is greater than the first preset threshold; the condition for satisfying the second type of features is that in the feature correlation coefficient matrix, the absolute values of all correlation coefficients are less than or equal to the first preset threshold.

[0065] Input the first type of features into the global contrast learning module to generate a global feature representation matrix; input the second type of features into the local contrast learning module to generate a local feature representation matrix.

[0066] Take the ratio of the number of the first type of features to the total number of features as the global feature weight; take the ratio of the number of the second type of features to the total number of features as the local feature weight.

[0067] Generate an enhanced feature matrix B through feature concatenation operation according to the global feature representation matrix, the local feature representation matrix, the global feature weight, and the local feature weight.

[0068] It should be noted that, firstly, traditional feature engineering methods often rely on manual experience to design feature combinations and are difficult to adapt to the rapidly updated medical knowledge in the field of tumor treatment; secondly, although existing automatic feature learning methods such as autoencoders or variational autoencoders can learn the implicit representations of features, they often ignore the hierarchical and local correlations of medical data, resulting in the lack of interpretability of the learned feature representations; thirdly, when dealing with high-dimensional and sparse medical features, existing technologies are easily interfered by noise and it is difficult to extract truly valuable feature associations. To address these problems, the present invention proposes a dual contrast learning network architecture, which realizes the adaptive enhancement of features through contrast learning modules at the global and local levels. Among them, the global contrast learning module learns the overall association pattern between features through the contrast of positive and negative sample pairs; the local contrast learning module focuses on the feature differences between neighboring patients and captures the local feature structure. The dual contrast learning mechanism can automatically discover and enhance the feature combinations valuable for prognosis prediction, avoiding the limitations of manual feature engineering in traditional methods and improving the feature representation ability; secondly, by dynamically adjusting the scope of action of the two modules through the correlation coefficient threshold, the self-adaptability of feature enhancement is realized, enabling the model to better process different types of medical features; thirdly, the feature enhancement method based on contrast learning provides a clear explanation of feature importance. By analyzing the contrast results of positive and negative sample pairs, doctors can intuitively understand which feature combinations have a key impact on the prediction results, significantly improving the clinical interpretability and credibility of the model.

[0069] S1.2.3: Based on the enhanced feature matrix B, perform prognosis assessment modeling through an enhanced feature adaptive ensemble tree model and output the first result.

[0070] Specifically, for each enhanced feature in the enhanced feature matrix, a dual-path decision tree ensemble model is constructed according to the corresponding values of each enhanced feature in the global feature representation matrix and the local feature representation matrix, as well as the global feature weight and the local feature weight:

[0071] First, calculate the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix, and the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix.

[0072] Secondly, calculate the total variance, within-group variance, and between-group variance of each enhanced feature in the enhanced feature matrix; compare the ratio of the within-group variance to the total variance with the second preset threshold to determine whether each enhanced feature value is in the feature stable interval.

[0073] Then, classify and judge each enhanced feature in the enhanced feature matrix: If the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix is greater than the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix, and the enhanced feature value is within the feature stable interval, then use a depth - first decision tree for prediction; If the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix is greater than the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix, or the enhanced feature value exceeds the feature stable interval, then use a breadth - first decision tree for prediction.

[0074] Next, calculate the depth - first decision tree weight coefficient and the breadth - first decision tree weight coefficient: The depth - first decision tree weight coefficient is the sum of the global feature weights of all enhanced features in the enhanced feature matrix B divided by the sum of the global feature weights and local feature weights of all enhanced features; The breadth - first decision tree weight coefficient is 1 minus the depth - first decision tree weight coefficient.

[0075] Finally, multiply the prediction result of the depth - first decision tree by the depth - first decision tree weight coefficient, and add the product of the prediction result of the breadth - first decision tree and the breadth - first decision tree weight coefficient to obtain the first result.

[0076] Furthermore, the first result includes the predicted values of survival probability, local recurrence risk, distant metastasis risk at 6 months, 12 months, and 24 months, as well as the predicted value of the occurrence risk of complications (which can be the 3 most common complications, such as infectious complications, cardiovascular complications, and respiratory complications). It should be noted that for each predicted value in the first result, its credibility is determined according to the following rules:

[0077] First, calculate the confidence interval stability score for each predicted value, specifically by subtracting the difference between the upper and lower limits of the confidence interval of each predicted value from each predicted value, and then dividing by each predicted value;

[0078] Second, calculate the similarity score between the predicted distribution and the true distribution of each predicted value, specifically by calculating the negative exponent of the KL divergence between the predicted distribution and the true distribution;

[0079] Finally, according to the comparison result of the confidence interval stability score with the third preset threshold, and the comparison results of the similarity score with the fourth preset threshold and the fifth preset threshold (where the fifth preset threshold is greater than the fourth preset threshold), divide the credibility into four levels:

[0080] When the confidence interval stability score is higher than the third preset threshold, and the similarity score is higher than the fifth preset threshold, the credibility is level 3;

[0081] When the confidence interval stability score is higher than the third preset threshold, and the similarity score is higher than the fourth preset threshold but not higher than the fifth preset threshold, the credibility level is 2;

[0082] When the confidence interval stability score is higher than the third preset threshold, the credibility level is 1;

[0083] In other cases, the credibility level is 0.

[0084] Preferably, in tumor prognosis assessment, the present invention realizes the dynamic trade-off between global-local features and adaptive feature selection through a dual-path decision tree ensemble model. Specifically, it is embodied as follows: by determining the feature stability through the ratio of within-group variance to total variance, and combining the comparison of global-local feature weights, the present invention realizes an intelligent switching mechanism that preferentially considers global representations (depth-first) within the stable interval and emphasizes local representations (breadth-first) within the unstable interval. This mechanism can not only effectively cope with the heterogeneity of individual characteristics of tumor patients but also adaptively adjust the prediction strategy. In addition, the present invention introduces a multi-level credibility assessment mechanism based on confidence interval stability and distribution similarity, and provides a more fine-grained reliability guarantee for the prediction results through the similarity threshold constraints in two dimensions of the validation set and the training set. The present invention significantly improves the robustness and interpretability of the prognosis assessment model in heterogeneous data scenarios, and provides a more reliable quantitative reference basis for clinical decision-making.

[0085] S1.3: Judge the risk category according to the first result.

[0086] First, calculate the comprehensive score of the survival probability prediction value. Specifically, take the 24-month survival probability prediction value as the benchmark score, and when the decline rates of the 12-month and 6-month survival probability prediction values exceed the decline rate threshold, multiply the benchmark score by the weighted penalty coefficient.

[0087] The comprehensive score of the survival probability prediction value is expressed as:

[0088] ;

[0089] Wherein, is the comprehensive score of the survival probability prediction value; is the 24-month survival probability prediction value (i.e., the benchmark score); P is the penalty coefficient.

[0090] The penalty coefficient P is calculated by the following formula:

[0091] ;

[0092] Wherein, is the decline rate of the 12-month survival probability; is the decline rate of the 6-month survival probability; is the descent rate threshold, used to determine whether the descent rate exceeds the standard; is the penalty adjustment coefficient, with a value range of (0, 1), used to adjust the penalty intensity for the descent rate exceeding the standard.

[0093] The descent rate is calculated by the following formula:

[0094] ;

[0095] ;

[0096] Among them, is the predicted value of the 12-month survival probability; is the predicted value of the 6-month survival probability.

[0097] Secondly, based on the local recurrence risk prediction value, distant metastasis risk prediction value, and the risk prediction values of complications (infectious complications, cardiovascular complications, respiratory complications), calculate the comprehensive score of risk events. Specifically, perform a weighted sum of all risk prediction values, and the weights are determined according to the influence degree of each risk event on the survival period.

[0098] Calculation of the comprehensive score of risk events:

[0099] ;

[0100] ;

[0101] Among them, RS is the comprehensive score of risk events; is the weight coefficient corresponding to each risk prediction value (k = 1, 2, 3, 4, 5); refers to the local recurrence risk prediction value, distant metastasis risk prediction value, and the risk prediction values of complications (infectious complications, cardiovascular complications, respiratory complications), : Local recurrence risk prediction value, : Distant metastasis risk prediction value, : Infectious complication risk prediction value, : Cardiovascular complication risk prediction value, : Respiratory complication risk prediction value.

[0102] Finally, perform a weighted combination of the comprehensive score of the survival probability prediction value and the comprehensive score RS of risk events, and combine the credibility levels of each prediction value to calculate the final risk classification score, and classify the tumor radiotherapy patients into three categories: low risk, medium risk, and high risk.

[0103] Specifically, the calculation of the final risk classification score:

[0104] ;

[0105] Among them, FS is the final risk classification score; is the survival probability weight, with a value range of (0.5, 1), which is used to balance the survival probability and the impact of risk events.

[0106] When the final risk classification score FS is greater than or equal to the sixth preset threshold T2 and the credibility is greater than or equal to level 2, it is determined as low risk; when the final risk classification score FS is greater than or equal to the seventh preset threshold T3 and less than the sixth preset threshold T2, or the credibility is level 1, it is determined as medium risk; when the final risk classification score FS is less than the seventh preset threshold T3, or the credibility = 0, it is determined as high risk.

[0107] S2: Obtain the symptom information of tumor radiotherapy patients, evaluate the severity of symptoms according to the symptom information, and generate a second result.

[0108] Specifically, the symptom information includes the symptom name, symptom onset time, symptom duration, symptom location, and symptom manifestation.

[0109] According to the symptom information, evaluate the symptom severity by referring to the radiotherapy acute reaction grading standard, and generate a second result including the symptom severity level.

[0110] It should be noted that the radiotherapy acute reaction grading standard can be: 1. RTOG / EORTC radiotherapy acute reaction grading standard; 2. CTCAE (Common Terminology Criteria for Adverse Events) adverse reaction grading standard; 3. WHO radiotherapy adverse reaction grading standard; 4. NCI-CTC (National Cancer Institute Common Toxicity Criteria) toxicity reaction grading standard.

[0111] S3: Conduct symptom management according to the second result and the risk category, and at the same time generate prognostic data, and update the first result through the prognostic data.

[0112] S3.1: Extract the corresponding radiotherapy adjustment plan and nursing intervention measures from the supporting symptom management database according to the symptom severity level and risk category in the second result.

[0113] S3.2: Obtain each treatment record and symptom severity level collected in the radiotherapy treatment room, and generate prognostic data including the symptom change trend.

[0114] Among them, the treatment record includes the actual irradiation dose, cumulative irradiation dose, symptom duration, and complication record.

[0115] S3.3: Perform format conversion on the prognostic data, specifically, standardize the prognostic data. If there are missing data in the prognostic data, fill in the missing values. Input the standardized prognostic data into the dual contrast learning network. If the similarity between the obtained new global feature representation matrix and the new local feature representation matrix is lower than the similarity of the corresponding matrix in the enhanced feature matrix, recalculate the feature weights and perform feature contrast learning. Input the optimized enhanced feature matrix into the dual-path decision tree ensemble model to update the survival probability prediction value, local recurrence risk prediction value, distant metastasis risk prediction value, and complication occurrence risk prediction value in the first result.

[0116] Preferably, in the existing radiotherapy, the patient symptom management mainly relies on empirical judgment and single symptom severity assessment, making it difficult to effectively grasp the correlation between symptom change trends and treatment prognosis, and lacking a flexible treatment plan adjustment mechanism. Through the synergistic effect of the dual contrast learning network and the dual-path decision tree ensemble model, the present invention realizes the dynamic association between symptom information and prognosis prediction: Firstly, based on the symptom severity level and risk category, timely adjustment of the treatment plan is carried out to ensure treatment safety; Secondly, multi-dimensional information such as irradiation dose and symptom changes during the treatment process is converted into standardized prognostic data; Finally, through the feature similarity comparison mechanism, when it is found that the quality of the feature representation decreases, the feature weights are optimized in a timely manner, so that the prognosis prediction result can be dynamically updated with symptom changes, greatly improving the accuracy of prognosis assessment and the scientific nature of treatment plan adjustment during radiotherapy, and providing effective technical support for the realization of precise and personalized radiotherapy management in clinical practice.

[0117] In summary, through the innovative architecture of constructing a dual contrast learning network and a dual-path decision tree ensemble model, the present invention realizes the dynamic association analysis and adaptive feature enhancement of multi-dimensional medical data of tumor radiotherapy patients. The present invention can not only accurately quantify the prognostic risk of patients, but also realize personalized treatment plan adjustment based on symptom severity grading, and has the following remarkable advantages: Firstly, by using time-decay-based standardization processing and weighted target coding conversion, the feature expression ability of heterogeneous medical data is improved; Secondly, through the synergistic effect of global and local contrast learning, the semantic correlation between features is enhanced, and the generalization performance of the model is improved; Thirdly, based on the ensemble strategy of the dual-path decision tree and the multi-level credibility evaluation mechanism, the accuracy and reliability of the prediction result are significantly improved; Finally, the dynamic prognostic data update mechanism enables the model to adaptively adjust the feature weights, realizing the continuous optimization of symptom management and providing more scientific and accurate decision support for clinical practice.

[0118] Example 2, an embodiment of the present invention, provides a symptom management and prognosis assessment system for tumor radiotherapy patients, which consists of a prognosis assessment module, a symptom assessment module, and a dynamic management module.

[0119] A prognosis evaluation module, configured to obtain basic information of a tumor radiotherapy patient, perform prognosis evaluation based on the basic information, generate a first result, and determine a risk category according to the first result; a symptom evaluation module, configured to obtain symptom information of a tumor radiotherapy patient, evaluate the severity of the symptoms according to the symptom information, and generate a second result; a dynamic management module, configured to perform symptom management according to the second result and the risk category, and at the same time generate prognosis data, and update the prognosis evaluation result through the prognosis data.

[0120] Example 3, referring to Figure 2 This is an embodiment of the present invention. The difference from the previous embodiment is that if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0121] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0122] More specific examples (a non-exhaustive list) of computer-readable media include the following: electrical connections (electronic devices) having one or more wirings, portable computer disk cartridges (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing it in a suitable manner if necessary, and then storing it in a computer memory.

[0123] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well-known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGA), field-programmable gate arrays (FPGA), etc.

[0124] Example 4, an embodiment of the present invention, provides a method for symptom management and prognosis evaluation of tumor radiotherapy patients. To verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0125] To verify the effectiveness of the present invention, 120 tumor radiotherapy patients admitted to the radiotherapy department of a certain tertiary hospital from January 2023 to December 2023 were selected as the research objects. First, the basic information of the patients, including age, gender, BMI index, KPS score, tumor stage, pathological type, radiotherapy planned dose, etc., was collected through the hospital information system. A standardized processing method based on time decay was used to process numerical features, and the decay coefficient was set to 0.85. At the same time, a weighted target encoding method was used to transform categorical features, and the encoding weights were determined based on feature importance analysis.

[0126] Construct a dual contrast learning network, where the global contrast learning module adopts a three-layer fully connected neural network structure, with the number of neurons in the hidden layers being 256, 128, and 64 respectively, and the activation function being ReLU; the local contrast learning module adopts a two-layer convolutional neural network enhanced by an attention mechanism, with the convolutional kernel size being 3×3 and the stride being 1. Set the first preset threshold to 0.6 for distinguishing global features and local features. Feature correlation analysis shows that 45% of the features belong to the first type of features (global features), and 55% belong to the second type of features (local features). In the dual-path decision tree ensemble model, the maximum depth of the depth-first decision tree is set to 8, and the minimum number of samples for splitting is 5; the maximum depth of the breadth-first decision tree is 5, and the minimum number of samples for splitting is 8. Set the second preset threshold to 0.3 for judging feature stability.

[0127] The research objects were randomly divided into an experimental group (60 cases) and a control group (60 cases). The experimental group used the method of the present invention for prognosis evaluation and symptom management, while the control group used the traditional evaluation method. There were no statistically significant differences in baseline characteristics such as age, gender, and disease stage between the two groups of patients (p>0.05). All patients completed an 8-week follow-up observation.

[0128] Table 1 Comparison of model performance evaluation indicators

[0129] Evaluation Index Results of Experimental Group Results of Control Group Improvement Rate (%) P Value Prediction Accuracy Rate (%) 92.5 77.8 18.9 <0.001 Specificity (%) 90.8 75.2 20.7 <0.001 Sensitivity (%) 91.2 76.5 19.2 <0.001 AUC Value 0.915 0.782 17.0 <0.001 F1 Score 0.898 0.742 21.0 <0.001 Kappa Coefficient 0.885 0.728 21.6 <0.001

[0130] As shown in Table 1, compared with the prior art, the present invention shows significant performance improvements in six core evaluation indicators. The prediction accuracy rate is increased from 77.8% to 92.5% (p<0.001), and the specificity and sensitivity reach 90.8% and 91.2% respectively, with increases of 20.7% and 19.2% compared with the control group respectively, fully demonstrating the advantages of the dual contrast learning network in feature enhancement. The AUC value reaches 0.915, with an increase of 17.0% compared with the control group, indicating that the model has excellent classification ability. The significant increases in the F1 score and Kappa coefficient (21.0% and 21.6% respectively) further verify the accuracy and consistency of the model prediction results. The p-values of all evaluation indicators are less than 0.001, with extremely strong statistical significance, confirming the outstanding advantages of the present invention in prediction performance.

[0131] Table 2 Comparison of symptom management effects

[0132] Observation Index Before Treatment 2 Weeks after Treatment 4 Weeks after Treatment 6 Weeks after Treatment 8 Weeks after Treatment Improvement Rate (%) P Value Incidence of Radiation Dermatitis (%) 0 15.2 28.5 32.1 25.4 21.3 <0.001 Incidence of Radiation Stomatitis (%) 0 12.8 25.6 28.9 20.1 23.5 <0.001 Incidence of Dysphagia (%) 0 10.5 22.3 25.7 18.2 25.8 <0.001 Fatigue Degree Score 2.5 3.2 4.1 3.8 2.9 28.4 <0.001 Quality of Life Score 75.2 72.1 69.8 71.5 76.8 31.2 <0.001

[0133] As shown in Table 2, through the analysis of the dynamic monitoring data during the entire treatment process (8 weeks), it was found that the incidence rates of the three main radiation complications showed a trend of first increasing and then decreasing, and the peak values were significantly lower than the levels reported in the literature. The incidence rate of radiation dermatitis increased from 15.2% at 2 weeks to 32.1% at 6 weeks and then decreased to 25.4% at 8 weeks, with an overall improvement rate of 21.3%. The changing trends of the incidence rates of radiation stomatitis and dysphagia were similar, with improvement rates of 23.5% and 25.8% respectively. The highest score of fatigue level appeared at the 4th week of treatment (4.1 points), and then gradually improved, with an overall improvement rate of 28.4%. The quality of life score decreased during the mid-term of treatment, but significantly rebounded to 76.8 points in the late stage of treatment, which was 31.2% higher than the lowest point. The improvements of all indicators were statistically significant (p < 0.001), indicating that the present invention has significant clinical value in symptom management.

[0134] Table 3 Evaluation of Clinical Application Effects

[0135] Clinical Index Experimental Group Control Group Difference Value P Value Treatment Plan Completion Rate (%) 94.5 85.2 9.3 <0.001 Treatment Compliance (%) 92.8 78.5 14.3 <0.001 Timely Intervention Rate of Adverse Reactions (%) 93.6 76.9 16.7 <0.001 Patient Satisfaction Score 92.1 77.8 14.3 <0.001 Length of Hospital Stay (days) 18.5 25.8 -7.3 <0.001

[0136] As shown in Table 3, the evaluation data of clinical application effects show that the present invention significantly reduces the consumption of medical resources while improving the treatment effect. The treatment plan completion rate (94.5%) and treatment compliance (92.8%) of the experimental group were 9.3 and 14.3 percentage points higher than those of the control group respectively, and the timely intervention rate of adverse reactions increased by 16.7%. These improvements directly reflect the clinical value of the intelligent management system. In terms of the utilization of medical resources, the average length of hospital stay in the experimental group was shortened by 7.3 days. The patient satisfaction score increased by 14.3 points to reach 92.1 points, comprehensively reflecting the advantages of the present invention in improving the quality of medical services and patient experience. The differences of all indicators were statistically significant (p < 0.001), fully confirming the comprehensive advantages of the present invention in clinical practice.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for symptom management and prognosis assessment in patients undergoing radiotherapy for tumors, characterized in that: include: Obtaining basic information of a patient undergoing tumor radiotherapy, performing a prognosis assessment based on the basic information, generating a first result, and determining a risk category based on the first result; Acquiring symptom information of a patient undergoing radiotherapy for a tumor, assessing the severity of the symptoms according to the symptom information, and generating a second result; performing symptom management according to the second result and the risk category, generating prognostic data, and updating the first result by using the prognostic data; The process of performing a prognosis assessment according to the basic information to generate a first result includes the following steps: Constructing a patient feature matrix, standardizing and encoding the numerical features and categorical features in the basic information, respectively, to generate a standardized feature matrix; Constructing a double contrast learning network to perform feature enhancement on the standardized feature matrix to generate an enhanced feature matrix; Based on the enhanced feature matrix, prognostic evaluation modeling is performed through an integrated tree model and a first result is output; The step of constructing a patient feature matrix, respectively standardizing and encoding the numerical features and categorical features in the basic information to generate a standardized feature matrix, includes the following steps: The basic information of the patient is constructed into an original feature matrix X including a demographic feature vector, a disease-related feature vector, a treatment-related feature vector, a medical history feature vector, a lifestyle feature vector, and a laboratory test feature vector; The numerical features in the original feature matrix X are normalized using the Z-score based on time decay: ； in, is the standardized value of the jth feature; is the original value of the jth feature of the patient to be evaluated; is the mean value of the jth characteristic of the patient to be evaluated; is the standard deviation of the jth feature of historical patients; is the current time point; The data collection time points for historical patients; is the time attenuation coefficient; The weighted target encoding conversion is used for the categorical features in the original feature matrix X: ； in, is the coded value of category c, is the target variable value of the rth historical patient in category c, is the historical number of patients in category c, is the time weight coefficient of the rth historical patient, calculated by the following formula: ； in, is the time weight attenuation coefficient, and its value range is [0.2, 0.6]; is the data collection time point for the rth historical patient; The standardized numerical eigenvalues ​​and the categorical eigenvalues ​​after encoding conversion are recombined according to the corresponding positions of the original features to generate a standardized feature matrix A: ； in, is the jth standardized feature value of the i-th patient to be evaluated. If the jth feature is a numerical feature, then ; If the jth feature is a categorical feature, then ; The dual contrast learning network includes a global contrast learning module and a local contrast learning module; Among them, the loss function of the global contrastive learning module is: ； in, is the characteristic representation of the i-th patient to be evaluated; is the feature representation corresponding to the positive matching historical patient of the i-th patient to be evaluated; is the feature representation of the vth negative matching historical patient; is the cosine similarity function; is the global temperature parameter, with a value range of [0.1, 1.0]; N is the number of negative matching historical patients; Among them, the loss function of the local contrast learning module is: ； in, Representation of local features of the patient to be evaluated; is the local feature representation of the wth nearest historical patient; is the local feature representation of the zth non-neighbor historical patient; K is the number of neighbor historical patients; M is the number of non-neighbor historical patients; is the local temperature parameter.

2. The method for symptom management and prognosis assessment of tumor radiotherapy patients according to claim 1, characterized in that: Constructing a double contrast learning network to perform feature enhancement on the standardized feature matrix to generate an enhanced feature matrix includes the following steps: Calculating the Pearson correlation coefficient between every two features in the standardized feature matrix to generate a feature correlation coefficient matrix; Traversing all correlation coefficients in the feature correlation coefficient matrix, and dividing the features into first-category features and second-category features according to the absolute values ​​of the correlation coefficients; wherein the condition for satisfying the first-category features is that in the feature correlation coefficient matrix, there is at least one feature whose absolute value of the correlation coefficient is greater than a first preset threshold; the condition for satisfying the second-category features is that in the feature correlation coefficient matrix, the absolute values ​​of all the correlation coefficients are less than or equal to the first preset threshold; Inputting the first type of features into the global contrast learning module to generate a global feature representation matrix; inputting the second type of features into the local contrast learning module to generate a local feature representation matrix; The ratio of the number of first-category features to the total number of features is used as the global feature weight; the ratio of the number of second-category features to the total number of features is used as the local feature weight; Generate an enhanced feature matrix through feature concatenation operation according to the global feature representation matrix and the local feature representation matrix, the global feature weights, and the local feature weights; Based on the enhanced feature matrix, prognostic evaluation modeling is performed through an integrated tree model and a first result is output, comprising the following steps: For each enhanced feature in the enhanced feature matrix, a dual-path decision tree integration model is constructed according to the corresponding value of each enhanced feature in the global feature representation matrix and the local feature representation matrix, as well as the global feature weight and the local feature weight: Calculating the product of the global feature weight and the corresponding enhanced eigenvalue in the global feature representation matrix, and the product of the local feature weight and the corresponding enhanced eigenvalue in the local feature representation matrix; Calculating the total variance, the intra-group variance and the inter-group variance of each enhanced feature in the enhanced feature matrix; comparing the ratio of the intra-group variance to the total variance with a second preset threshold to determine whether each enhanced feature value is in a feature stability interval; Classify and judge each enhanced feature in the enhanced feature matrix: if the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix is ​​greater than the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix, and the enhanced feature value is within the feature stability interval, then use the depth-first decision tree for prediction; if the product of the local feature weight and the corresponding enhanced feature value in the local feature representation matrix is ​​greater than the product of the global feature weight and the corresponding enhanced feature value in the global feature representation matrix, or the enhanced feature value exceeds the feature stability interval, then use the breadth-first decision tree for prediction; Calculate the weight coefficient of the depth-first decision tree and the weight coefficient of the breadth-first decision tree, multiply the prediction result of the depth-first decision tree by the weight coefficient of the depth-first decision tree, and add the product of the prediction result of the breadth-first decision tree and the weight coefficient of the breadth-first decision tree to obtain the survival probability prediction value, local recurrence risk prediction value, distant metastasis risk prediction value, and complication risk prediction value, respectively, to form the first result.

3. The method for symptom management and prognosis assessment of tumor radiotherapy patients according to claim 2, characterized in that: For each predicted value in the first result, the credibility calculation rule is as follows: Calculate the confidence interval stability score for each predicted value; Calculate the similarity score between the predicted distribution and the true distribution for each predicted value; According to the comparison result of the confidence interval stability score with the third preset threshold, and the comparison result of the similarity score with the fourth preset threshold and the fifth preset threshold, the credibility is divided into four levels: When the confidence interval stability score is higher than the third preset threshold and the similarity score is higher than the fifth preset threshold, the credibility is level 3; When the confidence interval stability score is higher than the third preset threshold, and the similarity score is higher than the fourth preset threshold but not higher than the fifth preset threshold, the credibility is level 2; When the confidence interval stability score is higher than the third preset threshold, the credibility is level 1; In other cases, the credibility is level 0.

4. The method for symptom management and prognosis assessment of tumor radiotherapy patients according to claim 3, characterized in that: Determining the risk category according to the first result includes the following steps: Calculate the composite score of the predicted value of survival probability; The comprehensive score of risk events was calculated based on the risk prediction values ​​of local recurrence, distant metastasis, and complications. The survival probability prediction value composite score and the risk event composite score are weighted and combined with the credibility to calculate the final risk classification score, and tumor radiotherapy patients are divided into three categories: low risk, medium risk and high risk: When the final risk classification score is greater than or equal to the sixth preset threshold, and the credibility is greater than or equal to level 2, it is judged as low risk; When the final risk classification score is greater than or equal to the seventh preset threshold and less than the sixth preset threshold, or the credibility is level 1, it is judged as medium risk; when the final risk classification score is less than the seventh preset threshold, or the credibility is 0, it is judged as high risk.

5. The method for symptom management and prognosis assessment of tumor radiotherapy patients according to claim 4, characterized in that: The symptom information includes symptom name, symptom onset time, symptom duration, symptom location and symptom manifestation; According to the symptom information, the symptom severity is evaluated according to the radiation therapy acute reaction grading standard to generate a second result including the symptom severity level.

6. The method for symptom management and prognosis assessment of tumor radiotherapy patients according to claim 5, characterized in that: Performing symptom management according to the second result and the risk category, generating prognostic data, and updating the first result by using the prognostic data, comprises the following steps: According to the symptom severity level and the risk category in the second result, extracting corresponding radiotherapy adjustment plans and nursing intervention measures from a supporting symptom management database; Obtaining each treatment record collected by the radiotherapy treatment room and the symptom severity level, and generating prognostic data including the symptom change trend; the treatment record includes the actual irradiation dose, the cumulative irradiation dose, the symptom duration, and the complication record; The prognostic data are standardized, and if there are missing data in the prognostic data, the missing values ​​are filled; the standardized prognostic data are input into the double contrast learning network, and if the similarity between the obtained new global feature representation matrix and the new local feature representation matrix is ​​lower than the similarity of the corresponding matrix in the enhanced feature matrix, the feature weights are recalculated and feature contrast learning is performed, and the optimized enhanced feature matrix is ​​input into the dual-path decision tree integrated model, and the survival probability prediction value, local recurrence risk prediction value, distant metastasis risk prediction value and complication risk prediction value in the first result are updated.

7. A system using the method for symptom management and prognosis assessment of tumor radiotherapy patients as described in any one of claims 1 to 6, characterized in that: include: A prognosis assessment module, used to obtain basic information of patients undergoing tumor radiotherapy, perform prognosis assessment based on the basic information, generate a first result, and determine the risk category based on the first result; A symptom assessment module, used to obtain symptom information of patients undergoing tumor radiotherapy, assess the severity of symptoms based on the symptom information, and generate a second result; A dynamic management module is used to perform symptom management according to the second result and the risk category, generate prognostic data, and update the first result through the prognostic data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for symptom management and prognosis assessment of tumor radiotherapy patients described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for symptom management and prognosis assessment of tumor radiotherapy patients described in any one of claims 1 to 6 are implemented.

Images