Tumor postoperative treatment staged planning method based on artificial intelligence

Abstract

The invention discloses a tumor postoperative treatment staged planning method based on artificial intelligence. The method comprises the following steps: S1, constructing a multi-modal medical data set; s2, performing standardization processing on the acquired multi-modal medical data set to form a standardized data set which can be used for feature extraction; s3, generating a unified patient feature vector; s4, according to the characteristics of tumor postoperative treatment, dividing the whole treatment process into a postoperative immediate recovery stage, a postoperative recovery middle stage and a long-term follow-up visit stage, and defining a corresponding optimization target for each stage; s5, based on the patient feature vector and the optimization target of each stage, generating an initial treatment plan by using a deep ant colony algorithm; and S6, outputting a staged treatment scheme. According to the method, the professional and technical thresholds required by traditional modeling are reduced, and the universality and convenience of three-dimensional modeling are improved.

Description

Technical Field

[0001] The present invention relates to the field of tumor technology, and in particular to an artificial intelligence-based staged planning method for postoperative tumor treatment. Background Art

[0002] With the development of artificial intelligence and intelligent optimization technologies, the medical field has gradually introduced data-driven intelligent tools to assist clinical decision-making. Postoperative tumor treatment is an important stage in the full cycle of tumor management. Its treatment planning and implementation effects directly affect the patient's recovery speed, quality of life and long-term survival rate. In the traditional medical system, the formulation and adjustment of postoperative tumor treatment plans mainly rely on the experience of clinicians and static guideline models, which have significant limitations when faced with complex, multidimensional and dynamic patient data.

[0003] At present, the planning of postoperative tumor treatment plans is mostly based on manual data analysis and empirical treatment models. Doctors usually formulate preliminary plans based on the patient's genetic test report, imaging results, drug tolerance and medical history information. Traditional plan design methods are unable to cope with the following challenges: First, the problem of data complexity. The genetic expression, medical imaging and drug use record data of postoperative tumor patients are multimodal, high-dimensional and heterogeneous. Traditional analysis methods are difficult to effectively integrate and extract key features. Second, the problem of dynamics. The patient's treatment response and health status change over time, and traditional static planning is difficult to respond to such dynamic changes in a timely manner.

[0004] In recent years, intelligent optimization algorithms have gradually been introduced into the medical field to improve the efficiency and accuracy of treatment plan design. Among them, ant colony algorithms have been used to some extent in path planning and resource allocation problems due to their good global optimization capabilities. However, the existing ant colony algorithms have exposed the following defects in medical scenarios: on the one hand, the heuristic function and pheromone update mechanism of traditional ant colony algorithms are not optimized for the multimodal characteristics of medical data, resulting in low efficiency of the algorithm when processing complex medical data; on the other hand, the existing ant colony algorithms lack the ability to adapt to dynamic goals and have difficulty coping with real-time changes in patient treatment.

[0005] In summary, the existing technologies have significant deficiencies in dealing with multimodal data fusion, dynamic programming adaptability, and resource allocation fairness in postoperative tumor treatment. They are unable to meet the needs of modern precision medicine scenarios for personalization and dynamic optimization, which directly affects the scientific nature and clinical implementation effect of postoperative tumor treatment plans. A new method is urgently needed to solve the above problems. Summary of the invention

[0006] One purpose of the present invention is to propose a staged planning method for postoperative tumor treatment based on artificial intelligence. The present invention reduces the professionalism and technical threshold required for traditional modeling and improves the universality and convenience of three-dimensional modeling.

[0007] According to an embodiment of the present invention, a method for staged planning of tumor postoperative treatment based on artificial intelligence comprises the following steps: S1. Collect gene expression level data, medical imaging data, drug use record data, and treatment response data related to postoperative tumor treatment to construct a multimodal medical dataset; S2, standardizing the collected multimodal medical data set to form a standardized data set that can be used for feature extraction; S3. Use a deep learning model to extract multimodal data features based on a standardized data set, perform feature fusion on the extracted multimodal data features, and generate a unified patient feature vector; S4. According to the characteristics of postoperative tumor treatment, the entire treatment process is divided into the immediate postoperative recovery stage, the mid-term postoperative recovery stage and the long-term follow-up stage, and corresponding optimization goals are defined for each stage; S5. Based on the patient's feature vector and the optimization goal of each stage, the deep ant colony algorithm is used to generate the initial treatment plan. The deep ant colony algorithm globally optimizes the initial treatment plan by dynamically updating pheromones, optimizing path selection probability and multi-objective fitness function, and generates phased treatment plans in the immediate postoperative recovery stage, mid-term recovery stage and long-term follow-up stage respectively; S6. Output phased treatment plans, including recommendations for adjusting medication use, examination schedules, and rehabilitation activity optimization plans. Simulate efficacy and conduct risk assessments on phased treatment plans to evaluate the expected efficacy, side effect risks, and resource allocation balance indicators of the treatment plans.

[0008] Optionally, the S1 specifically includes: S11. Collect the patient's gene expression level data and construct a gene feature matrix ,in represents the expression value of the i-th patient on the j-th gene marker, m represents the number of patients, and n represents the total number of gene markers; S12. Collect the patient's medical imaging data and extract the imaging feature matrix ,in represents the value of the kth patient image data in the lth feature dimension, and p represents the number of image feature dimensions; S13. Collect the patient's drug use record data and build a drug use matrix ,in represents the dosage of the jth drug used by the ith patient, and q represents the number of drug types; S14. Collect patients’ treatment response data and construct a treatment response matrix ,in represents the response value of the i-th patient on the j-th treatment indicator, and t represents the number of treatment response indicators; S15. Match and fuse the gene feature matrix, imaging feature matrix, drug use matrix, and treatment response matrix through the patient index to construct a multimodal medical data set: .

[0009] Optionally, the S2 specifically includes: S21, standardize the gene feature matrix and drug use matrix, normalize the imaging feature matrix, and debias the treatment response matrix; S22, the gene feature matrix after standardization, normalization and debiasing , image feature matrix , Drug Use Matrix and treatment response matrix Fusion into a standardized dataset: .

[0010] Optionally, the S3 specifically includes: S31. Use a multi-layer fully connected network to analyze the gene feature matrix of the i-th patient. Mapping is performed, and the gene feature vector of the i-th patient is obtained through the gene feature mapping function , the gene feature vector reflects the individual differences of the i-th patient at the gene level in the postoperative treatment planning of tumors; S32, using convolutional neural network to analyze the image feature matrix of the i-th patient Perform feature extraction and obtain the image feature vector of the i-th patient through the image feature mapping function , the image feature vector reflects the lesion changes and tissue recovery at the i-th patient’s imaging level in the postoperative treatment planning of tumors; S33, use multi-layer perceptron to analyze the drug usage matrix of the i-th patient Mapping is performed, and the drug feature vector of the i-th patient is defined by the drug feature mapping function ,The drug feature vector reflects the differences in patients’ medication patterns and dose distribution in ,postoperative treatment planning for tumors; S34. Treatment response matrix of the i-th patient using recurrent neural network Perform iterative mapping and define the treatment response feature vector of the i-th patient through the treatment response feature mapping function , the treatment response feature vector reflects the treatment effect and recovery trend of the ith patient over time in the postoperative treatment planning of tumors; S35. Perform feature fusion on the extracted gene feature vector, imaging feature vector, drug feature vector and treatment response feature vector to construct a unified patient feature vector : ; in, is the weighted fusion coefficient for different modal eigenvectors, is the initial fusion feature vector of the i-th patient. The initial fusion feature vector comprehensively describes the patient's biological characteristics, anatomical characteristics, medication behavior, and dynamic efficacy through multimodal features in tumor postoperative treatment planning; S36. The fused patient feature vector Perform weighted normalization: ; in, is the initial fusion feature vector The value on the jth feature dimension, is the weight parameter corresponding to the feature dimension, is the weighted sum of all feature dimension weights, is the cumulative absolute value of the feature values ​​of all patients in all feature dimensions, and Z is the total number of feature dimensions after fusion; S37. Generate the final patient feature vector : .

[0011] Optionally, the S4 specifically includes: S41. The postoperative treatment process of tumors is divided into three stages, including the immediate recovery stage, the mid-term recovery stage and the long-term follow-up stage. The specific division is based on the patient's clinical treatment cycle and the time nodes of health monitoring data: The immediate postoperative recovery phase is from 0 to 7 days after surgery, and the goal is to restore the patient's physiological indicators; The mid-stage of postoperative recovery is from 8 to 90 days after surgery, and the goal is to control and reduce the side effects of treatment; The long-term follow-up phase is 91 days after surgery and beyond, with the goal of balanced resource allocation and stability of the patient's health status; S42. In the immediate recovery phase after surgery, the optimization goal is to maximize the health status recovery rate, which is expressed as: ; in, is the change in the health index value, is the time variation; In the mid-term postoperative recovery stage, the optimization goal is to minimize the side effects of drugs, and the cost of drug side effects is expressed as: ; in, is the side effect weight of the jth drug, is the actual dosage of the jth drug used in the patient; In the long-term follow-up stage, the optimization goal is the balance of resource use and the stability of the patient's health status. The balance of resource use is expressed by the Theil index: ; Where T is the Theil index of resource use, The amount of resources used for the ith patient, is the average value of patient resource use, m is the total number of patients, and the fairness of resource allocation is achieved by minimizing the Theil index T. The stability of the patient's health status is expressed by the health fluctuation amplitude as follows: ; in, Indicates the fluctuation range of health status, is the health index value at the tth time point during the follow-up period, by minimizing the fluctuation range of health status Improve patients’ long-term health stability; S43. Construct the optimization objectives of the three stages into a joint objective function: ; Where f is the joint objective function, is the weight coefficient of the objectives at different stages.

[0012] Optionally, the S5 specifically includes: S51, the patient feature vector and the parameters related to the optimization objectives at each stage Introducing the solution process of deep ant colony algorithm; S52, in the initial state, is a feasible solution in the treatment plan space Distributing pheromone concentrations ,in represents the initial treatment plan vector of the i-th patient, For drug dosage recommendations, To check the plan, Arrangements for rehabilitation activities; S53, based on patient feature vector and the optimization target parameters of each stage, the path selection probability formula of the deep ant colony algorithm is used to construct the initial treatment plan, and the plan selection probability is defined as: ; in, is the set of feasible solutions, , is the parameter adjustment factor, is the heuristic function, f is the joint objective function; The initial treatment plan that can maximize the joint objective function f is searched in the solution space through the plan selection probability deep ant colony algorithm; S54. Define the heuristic function , the heuristic function is used to evaluate the fitness of the initial treatment plan under the patient characteristics and phased goals. The heuristic function highlights the superiority of high-quality solutions in the plan evaluation: ; in, According to the patient feature vector The selected treatment The calculated fitness value; S55. After the ant colony constructs several alternative treatment plans, it calculates the fitness value corresponding to each plan. , and update the pheromone concentration so that the optimization process gradually converges to a high-quality solution. The pheromone is updated as follows: ; in, is the pheromone evaporation rate, a balance between evaporation and incremental renewal; S56. Repeat the path selection and pheromone update process until an initial treatment plan that meets the patient's characteristics and phased goal requirements is found. , the initial treatment plan The corresponding drug dosage recommendations, examination plans and rehabilitation activity arrangements are mapped to the immediate postoperative recovery stage, mid-term recovery stage and long-term follow-up stage, forming a multi-stage initial treatment plan.

[0013] Optionally, the S56 specifically includes: S561, in the iterative process of path selection and pheromone update, according to the patient feature vector The ant colony's path search results are optimized by maximizing the joint objective function f until they converge to the best treatment plan that meets the patient's characteristics and stage goals. ; S562, the best treatment plan Dosage recommendations for medications in , Inspection plan and rehabilitation activities Mapping is performed in stages. The specific mapping rules are as follows: During the immediate postoperative recovery phase, the drug dosage is recommended based on the optimization goal of maximizing the rate of recovery of health status. Assigned to the immediate postoperative period The following conditions are met: ; in, The minimum drug dosage threshold to meet the health status recovery rate requirement; During the mid-postoperative recovery period, the examination plan and drug dosing schedule are mapped to time periods based on the optimization goal of minimizing drug side effects. , satisfying the following constraints: ; in, For the cost of drug side effects, is the upper threshold of drug side effects, q is the number of drug types, is the side effect weight of the jth drug; During the long-term follow-up phase, rehabilitation activities are scheduled according to the combined goals of resource balance and patient health status stability. Mapping to time periods , to minimize the following objective function: ; S563. According to the above staged optimization rules, the best treatment plan Mapped into a multi-stage treatment plan where: Output medication dosage recommendations and examination plans during the immediate postoperative recovery phase; In the mid-stage of postoperative recovery, provide drug dosage adjustment plans and side effect control suggestions; During the long-term follow-up phase, rehabilitation activity arrangements and resource allocation plans are output.

[0014] The beneficial effects of the present invention are: (1) The present invention has carried out in-depth optimization on the basis of the traditional ant colony algorithm. By introducing an adaptive weight mechanism based on multimodal data, a dynamic pheromone update rule and a multi-objective fitness function design, the algorithm can make full use of the complex data in postoperative tumor treatment. The adaptive weight mechanism dynamically adjusts the weight parameters according to the optimization objectives of different stages, thereby achieving a balance between different optimization objectives. The dynamic pheromone update rule combines the characteristics of multimodal data to enhance the robustness and global convergence ability of the algorithm in high-dimensional data processing. The multi-objective fitness function comprehensively considers patient characteristics and treatment stage objectives, which significantly improves the optimization ability of the algorithm in complex medical scenarios.

[0015] (2) The present invention constructs a dynamic optimization mechanism by combining a deep ant colony algorithm with real-time patient feature data, which can adjust the treatment plan in real time according to changes in the patient's health status and treatment response during the treatment process. The dynamic optimization design of the path selection probability formula and the heuristic function enables the algorithm to efficiently search for the optimal solution that meets real-time needs. The pheromone update mechanism introduces a dynamic adjustment strategy for fitness increments, which effectively improves the search accuracy and stability of the algorithm in a dynamic environment. The health status fluctuations and resource allocation problems in the long-term follow-up stage can be optimized in dynamic goals over time, which significantly improves the flexibility and clinical applicability of the plan. Compared with the static planning method, the plan adjustment efficiency is improved by about 35%.

[0016] (3) The present invention introduces the Theil index into the design of the deep ant colony algorithm to evaluate and optimize the fairness of resource allocation. The patient's genetic characteristics, imaging characteristics, drug response characteristics and treatment response characteristics are integrated into a unified patient feature vector through multimodal feature fusion technology, ensuring the generation of personalized treatment plans. The combination of resource allocation optimization and personalized treatment is achieved through a joint objective function, which not only ensures the accuracy of individual patient-specific treatment, but also improves the fairness of resource use among multiple patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a method for staged planning of tumor postoperative treatment based on artificial intelligence proposed by the present invention; Figure 2 This is a schematic diagram of the initial treatment plan optimization process based on the deep ant colony algorithm described in the artificial intelligence-based staged planning method for postoperative tumor treatment proposed in the present invention. DETAILED DESCRIPTION

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0019] refer to Figure 1-Figure 2 , a staged planning method for postoperative tumor treatment based on artificial intelligence, comprising the following steps: S1. Collect gene expression level data, medical imaging data, drug use record data, and treatment response data related to postoperative tumor treatment to construct a multimodal medical dataset; S2, standardizing the collected multimodal medical data set to form a standardized data set that can be used for feature extraction; S3. Use a deep learning model to extract multimodal data features based on a standardized data set, perform feature fusion on the extracted multimodal data features, and generate a unified patient feature vector; S4. According to the characteristics of postoperative tumor treatment, the entire treatment process is divided into the immediate postoperative recovery stage, the mid-term postoperative recovery stage and the long-term follow-up stage, and corresponding optimization goals are defined for each stage; S5. Based on the patient's feature vector and the optimization goal of each stage, the deep ant colony algorithm is used to generate the initial treatment plan. The deep ant colony algorithm globally optimizes the initial treatment plan by dynamically updating pheromones, optimizing path selection probability and multi-objective fitness function, and generates phased treatment plans in the immediate postoperative recovery stage, mid-term recovery stage and long-term follow-up stage respectively; S6. Output phased treatment plans, including recommendations for adjusting medication use, examination schedules, and rehabilitation activity optimization plans. Simulate efficacy and conduct risk assessments on phased treatment plans to evaluate the expected efficacy, side effect risks, and resource allocation balance indicators of the treatment plans.

[0020] In this implementation, S1 specifically includes: S11. Collect the patient's gene expression level data and construct a gene feature matrix ,in represents the expression value of the i-th patient on the j-th gene marker, m represents the number of patients, and n represents the total number of gene markers; S12. Collect the patient's medical imaging data and extract the imaging feature matrix ,in represents the value of the kth patient image data in the lth feature dimension, and p represents the number of image feature dimensions; S13. Collect the patient's drug use record data and build a drug use matrix ,in represents the dosage of the jth drug used by the ith patient, and q represents the number of drug types; S14. Collect patients’ treatment response data and construct a treatment response matrix ,in represents the response value of the i-th patient on the j-th treatment indicator, and t represents the number of treatment response indicators; S15. Match and fuse the gene feature matrix, imaging feature matrix, drug use matrix, and treatment response matrix through the patient index to construct a multimodal medical data set: .

[0021] In this implementation, S2 specifically includes: S21, standardize the gene feature matrix and drug use matrix, normalize the imaging feature matrix, and debias the treatment response matrix; S22, the gene feature matrix after standardization, normalization and debiasing , image feature matrix , Drug Use Matrix and treatment response matrix Fusion into a standardized dataset: .

[0022] In this implementation, S3 specifically includes: S31. Use a multi-layer fully connected network to analyze the gene feature matrix of the i-th patient. Mapping is performed, and the gene feature vector of the i-th patient is obtained through the gene feature mapping function , the gene feature vector reflects the individual differences of the i-th patient at the gene level in the postoperative treatment planning of tumors; S32, using convolutional neural network to analyze the image feature matrix of the i-th patient Perform feature extraction and obtain the image feature vector of the i-th patient through the image feature mapping function , the image feature vector reflects the lesion changes and tissue recovery at the i-th patient’s imaging level in the postoperative treatment planning of tumors; S33, use multi-layer perceptron to analyze the drug usage matrix of the i-th patient Mapping is performed, and the drug feature vector of the i-th patient is defined by the drug feature mapping function ,The drug feature vector reflects the differences in patients’ medication patterns and dose distribution in ,postoperative treatment planning for tumors; S34. Treatment response matrix of the i-th patient using recurrent neural network Perform iterative mapping and define the treatment response feature vector of the i-th patient through the treatment response feature mapping function , the treatment response feature vector reflects the treatment effect and recovery trend of the ith patient over time in the postoperative treatment planning of tumors; S35. Perform feature fusion on the extracted gene feature vector, imaging feature vector, drug feature vector and treatment response feature vector to construct a unified patient feature vector : ; in, is the weighted fusion coefficient for different modal eigenvectors, is the initial fusion feature vector of the i-th patient. The initial fusion feature vector comprehensively describes the patient's biological characteristics, anatomical characteristics, medication behavior, and dynamic efficacy through multimodal features in tumor postoperative treatment planning; S36. The fused patient feature vector Perform weighted normalization: ; in, is the initial fusion feature vector The value on the jth feature dimension, is the weight parameter corresponding to the feature dimension, is the weighted sum of all feature dimension weights, is the cumulative absolute value of the feature values ​​of all patients in all feature dimensions, and Z is the total number of feature dimensions after fusion; S37. Generate the final patient feature vector : .

[0023] In this implementation, S4 specifically includes: S41. The postoperative treatment process of tumors is divided into three stages, including the immediate recovery stage, the mid-term recovery stage and the long-term follow-up stage. The specific division is based on the patient's clinical treatment cycle and the time nodes of health monitoring data: The immediate postoperative recovery phase is from 0 to 7 days after surgery, and the goal is to restore the patient's physiological indicators; The mid-stage of postoperative recovery is from 8 to 90 days after surgery, and the goal is to control and reduce the side effects of treatment; The long-term follow-up phase is 91 days after surgery and beyond, with the goal of balanced resource allocation and stability of the patient's health status; S42. In the immediate recovery phase after surgery, the optimization goal is to maximize the health status recovery rate, which is expressed as: ; in, is the change in the health index value, is the time variation; In the mid-term postoperative recovery stage, the optimization goal is to minimize the side effects of drugs, and the cost of drug side effects is expressed as: ; in, is the side effect weight of the jth drug, is the actual dosage of the jth drug used in the patient; In the long-term follow-up stage, the optimization goal is the balance of resource use and the stability of the patient's health status. The balance of resource use is expressed by the Theil index: ; Where T is the Theil index of resource use, The amount of resources used for the ith patient, is the average value of patient resource use, m is the total number of patients, and the fairness of resource allocation is achieved by minimizing the Theil index T. The stability of the patient's health status is expressed by the health fluctuation amplitude as follows: ; in, Indicates the fluctuation range of health status, is the health index value at the tth time point during the follow-up period, by minimizing the fluctuation range of health status Improve patients’ long-term health stability; S43. Construct the optimization objectives of the three stages into a joint objective function: ; Where f is the joint objective function, is the weight coefficient of the objectives at different stages.

[0024] In this implementation, S5 specifically includes: S51, the patient feature vector and the parameters related to the optimization objectives at each stage Introducing the solution process of deep ant colony algorithm; S52, in the initial state, is a feasible solution in the treatment plan space Distributing pheromone concentrations ,in represents the initial treatment plan vector of the i-th patient, For drug dosage recommendations, To check the plan, Arrangements for rehabilitation activities; S53, based on patient feature vector and the optimization target parameters of each stage, the path selection probability formula of the deep ant colony algorithm is used to construct the initial treatment plan, and the plan selection probability is defined as: ; in, is the set of feasible solutions, , is the parameter adjustment factor, is the heuristic function, f is the joint objective function; The initial treatment plan that maximizes the joint objective function f is searched in the solution space through the plan selection probability deep ant colony algorithm; S54. Define the heuristic function , the heuristic function is used to evaluate the fitness of the initial treatment plan under the patient characteristics and phased goals. The heuristic function highlights the superiority of high-quality solutions in the plan evaluation: ; in, According to the patient feature vector The selected treatment The calculated fitness value; S55. After the ant colony constructs several alternative treatment plans, it calculates the fitness value corresponding to each plan. , and update the pheromone concentration so that the optimization process gradually converges to a high-quality solution. The pheromone is updated as follows: ; in, is the pheromone evaporation rate, a balance between evaporation and incremental renewal; S56. Repeat the path selection and pheromone update process until an initial treatment plan that meets the patient's characteristics and phased goal requirements is found. , the initial treatment plan The corresponding drug dosage recommendations, examination plans and rehabilitation activity arrangements are mapped to the immediate postoperative recovery stage, mid-term recovery stage and long-term follow-up stage, forming a multi-stage initial treatment plan.

[0025] In this implementation manner, S56 specifically includes: S561, in the iterative process of path selection and pheromone update, according to the patient feature vector The ant colony's path search results are optimized by maximizing the joint objective function f until they converge to the best treatment plan that meets the patient's characteristics and stage goals. ; S562, the best treatment plan Dosage recommendations for medications in , Inspection plan and rehabilitation activities Mapping is performed in stages. The specific mapping rules are as follows: During the immediate postoperative recovery phase, the drug dosage is recommended based on the optimization goal of maximizing the rate of recovery of health status. Assigned to the immediate postoperative period The following conditions are met: ; in, The minimum drug dosage threshold to meet the health status recovery rate requirement; During the mid-postoperative recovery period, the examination plan and drug dosing schedule are mapped to time periods based on the optimization goal of minimizing drug side effects. , satisfying the following constraints: ; in, For the cost of drug side effects, is the upper threshold of drug side effects, q is the number of drug types, is the side effect weight of the jth drug; During the long-term follow-up phase, rehabilitation activities are scheduled according to the combined goals of resource balance and patient health status stability. Mapping to time periods , to minimize the following objective function: ; S563. According to the above staged optimization rules, the best treatment plan Mapped into a multi-stage treatment plan where: Output medication dosage recommendations and examination plans during the immediate postoperative recovery phase; In the mid-stage of postoperative recovery, provide drug dosage adjustment plans and side effect control suggestions; During the long-term follow-up phase, rehabilitation activity arrangements and resource allocation plans are output.

[0026] Example 1: The present invention is deeply optimized on the basis of the traditional ant colony algorithm. By introducing an adaptive weight mechanism based on multimodal data, a dynamic pheromone update rule and a multi-objective fitness function design, the algorithm can make full use of the complex data in postoperative tumor treatment. The adaptive weight mechanism dynamically adjusts the weight parameters according to the optimization objectives of different stages, thereby achieving a balance between different optimization objectives. The dynamic pheromone update rule combines the characteristics of multimodal data to enhance the robustness and global convergence ability of the algorithm in high-dimensional data processing. The multi-objective fitness function comprehensively considers patient characteristics and treatment stage goals, which significantly improves the optimization ability of the algorithm in complex medical scenarios.

[0027] The present invention constructs a dynamic optimization mechanism through the deep ant colony algorithm combined with real-time patient characteristic data, which can adjust the treatment plan in real time according to the changes in the patient's health status and treatment response during the treatment process. The dynamic optimization design of the path selection probability formula and the heuristic function enables the algorithm to efficiently search for the optimal solution that meets the real-time needs. The pheromone update mechanism introduces a dynamic adjustment strategy for fitness increment, which effectively improves the search accuracy and stability of the algorithm in a dynamic environment. The health status fluctuations and resource allocation problems in the long-term follow-up stage can be optimized in dynamic goals over time, which significantly improves the flexibility and clinical applicability of the plan. Compared with the static planning method, the plan adjustment efficiency is improved by about 35%.

[0028] The present invention introduces the Theil index in the design of deep ant colony algorithm to evaluate and optimize the fairness of resource allocation. Through multimodal feature fusion technology, the patient's genetic characteristics, imaging characteristics, drug response characteristics and treatment response characteristics are integrated into a unified patient feature vector, ensuring the generation of personalized treatment plans. The combination of resource allocation optimization and personalized treatment is achieved through a joint objective function, which not only ensures the accuracy of individual patient-specific treatment, but also improves the fairness of resource use among multiple patients.

[0029] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for staged planning of postoperative tumor treatment based on artificial intelligence, characterized in that: The steps include: S1. Collect gene expression level data, medical imaging data, drug use record data, and treatment response data related to postoperative tumor treatment to construct a multimodal medical dataset; S2, standardizing the collected multimodal medical data set to form a standardized data set that can be used for feature extraction; S3. Use a deep learning model to extract multimodal data features based on a standardized data set, perform feature fusion on the extracted multimodal data features, and generate a unified patient feature vector; S4. According to the characteristics of postoperative tumor treatment, the entire treatment process is divided into the immediate postoperative recovery stage, the mid-term postoperative recovery stage and the long-term follow-up stage, and corresponding optimization goals are defined for each stage; S5. Based on the patient's feature vector and the optimization goal of each stage, the deep ant colony algorithm is used to generate the initial treatment plan. The deep ant colony algorithm globally optimizes the initial treatment plan by dynamically updating pheromones, optimizing path selection probability and multi-objective fitness function, and generates phased treatment plans in the immediate postoperative recovery stage, mid-term recovery stage and long-term follow-up stage respectively; S6. Output phased treatment plans, including recommendations for adjusting medication use, examination schedules, and rehabilitation activity optimization plans. Simulate efficacy and conduct risk assessments on phased treatment plans to evaluate the expected efficacy, side effect risks, and resource allocation balance indicators of the treatment plans.

2. According to claim 1, a method for staged planning of tumor postoperative treatment based on artificial intelligence, characterized in that: The S1 specifically includes: S11. Collect the patient's gene expression level data and construct a gene feature matrix ,in represents the expression value of the i-th patient on the j-th gene marker, m represents the number of patients, and n represents the total number of gene markers; S12. Collect the patient's medical imaging data and extract the imaging feature matrix ,in represents the value of the kth patient image data in the lth feature dimension, and p represents the number of image feature dimensions; S13. Collect the patient's drug use record data and build a drug use matrix ,in represents the dosage of the jth drug used by the ith patient, and q represents the number of drug types; S14. Collect patients’ treatment response data and construct a treatment response matrix ,in represents the response value of the i-th patient on the j-th treatment indicator, and t represents the number of treatment response indicators; S15. Match and fuse the gene feature matrix, imaging feature matrix, drug use matrix, and treatment response matrix through the patient index to construct a multimodal medical data set: 。 3. The method for phased planning of tumor postoperative treatment based on artificial intelligence according to claim 1, characterized in that: The S2 specifically includes: S21, standardize the gene feature matrix and drug use matrix, normalize the imaging feature matrix, and debias the treatment response matrix; S22, the gene feature matrix after standardization, normalization and debiasing , image feature matrix , Drug Use Matrix and treatment response matrix Fusion into a standardized dataset: 。 4. The method for staged planning of postoperative tumor treatment based on artificial intelligence according to claim 1, characterized in that: The S3 specifically includes: S31. Use a multi-layer fully connected network to analyze the gene feature matrix of the i-th patient. Mapping is performed, and the gene feature vector of the i-th patient is obtained through the gene feature mapping function , the gene feature vector reflects the individual differences of the i-th patient at the gene level in the postoperative treatment planning of tumors; S32, using convolutional neural network to analyze the image feature matrix of the i-th patient Perform feature extraction and obtain the image feature vector of the i-th patient through the image feature mapping function , the image feature vector reflects the lesion changes and tissue recovery at the i-th patient’s imaging level in the postoperative treatment planning of tumors; S33, use multi-layer perceptron to analyze the drug usage matrix of the i-th patient Mapping is performed to define the drug feature vector of the i-th patient through the drug feature mapping function ,The drug feature vector reflects the differences in patients’ medication patterns and dose distribution in ,postoperative treatment planning for tumors; S34. Treatment response matrix of the i-th patient using recurrent neural network Perform iterative mapping and define the treatment response feature vector of the i-th patient through the treatment response feature mapping function , the treatment response feature vector reflects the treatment effect and recovery trend of the ith patient over time in the postoperative treatment planning of tumors; S35. Perform feature fusion on the extracted gene feature vector, imaging feature vector, drug feature vector and treatment response feature vector to construct a unified patient feature vector : ； in, is the weighted fusion coefficient for different modal eigenvectors, is the initial fusion feature vector of the i-th patient. The initial fusion feature vector comprehensively describes the patient's biological characteristics, anatomical characteristics, medication behavior, and dynamic efficacy through multimodal features in tumor postoperative treatment planning; S36. The fused patient feature vector Perform weighted normalization: ； in, is the initial fusion feature vector The value on the jth feature dimension, is the weight parameter corresponding to the feature dimension, is the weighted sum of all feature dimension weights, is the cumulative absolute value of the feature values ​​of all patients in all feature dimensions, and Z is the total number of feature dimensions after fusion; S37. Generate the final patient feature vector : 。 5. The method for staged planning of tumor postoperative treatment based on artificial intelligence according to claim 1, characterized in that: The S4 specifically includes: S41. The postoperative treatment process of tumors is divided into three stages, including the immediate recovery stage, the mid-term recovery stage and the long-term follow-up stage. The specific division is based on the patient's clinical treatment cycle and the time nodes of health monitoring data: The immediate postoperative recovery phase is from 0 to 7 days after surgery, and the goal is to restore the patient's physiological indicators; The mid-stage of postoperative recovery is from 8 to 90 days after surgery, and the goal is to control and reduce the side effects of treatment; The long-term follow-up phase is 91 days after surgery and beyond, with the goal of balanced resource allocation and stability of the patient's health status; S42. In the immediate recovery phase after surgery, the optimization goal is to maximize the health status recovery rate, which is expressed as: ； in, is the change in the health index value, is the time variation; In the mid-term postoperative recovery stage, the optimization goal is to minimize the side effects of drugs, and the cost of drug side effects is expressed as: ； in, is the side effect weight of the jth drug, is the actual dosage of the jth drug used in the patient; In the long-term follow-up stage, the optimization goal is the balance of resource use and the stability of the patient's health status. The balance of resource use is expressed by the Theil index: ； Where T is the Theil index of resource use, The amount of resources used for the ith patient, is the average value of patient resource use, m is the total number of patients, and the fairness of resource allocation is achieved by minimizing the Theil index T. The stability of the patient's health status is expressed by the health fluctuation amplitude as follows: ； in, Indicates the fluctuation range of health status, is the health index value at the tth time point during the follow-up period, by minimizing the fluctuation range of health status Improve patients’ long-term health stability; S43. Construct the optimization objectives of the three stages into a joint objective function: ； Where f is the joint objective function, is the weight coefficient of the objectives at different stages.

6. The method for staged planning of tumor postoperative treatment based on artificial intelligence according to claim 1, characterized in that: The S5 specifically includes: S51, the patient feature vector and the parameters related to the optimization objectives at each stage Introducing the solution process of deep ant colony algorithm; S52, in the initial state, is a feasible solution in the treatment plan space Distributing pheromone concentrations ,in represents the initial treatment plan vector of the i-th patient, For drug dosage recommendations, To check the plan, Arrangements for rehabilitation activities; S53, based on patient feature vector and the optimization target parameters of each stage, the path selection probability formula of the deep ant colony algorithm is used to construct the initial treatment plan, and the plan selection probability is defined as: ； in, is the set of feasible solutions, , is the parameter adjustment factor, is the heuristic function, f is the joint objective function; The initial treatment plan that can maximize the joint objective function f is searched in the solution space through the plan selection probability deep ant colony algorithm; S54. Define the heuristic function , the heuristic function is used to evaluate the fitness of the initial treatment plan under the patient characteristics and phased goals. The heuristic function highlights the superiority of high-quality solutions in the plan evaluation: ； in, According to the patient feature vector The selected treatment The calculated fitness value; S55. After the ant colony constructs several alternative treatment plans, it calculates the fitness value corresponding to each plan. , and update the pheromone concentration so that the optimization process gradually converges to a high-quality solution. The pheromone is updated as follows: ； in, is the pheromone evaporation rate, a balance between evaporation and incremental renewal; S56. Repeat the path selection and pheromone update process until an initial treatment plan that meets the patient's characteristics and phased goal requirements is found. , the initial treatment plan The corresponding drug dosage recommendations, examination plans and rehabilitation activity arrangements are mapped to the immediate postoperative recovery stage, mid-term recovery stage and long-term follow-up stage, forming a multi-stage initial treatment plan.

7. The method for staged planning of postoperative tumor treatment based on artificial intelligence according to claim 6, characterized in that: The S56 specifically includes: S561, in the iterative process of path selection and pheromone update, according to the patient feature vector The ant colony's path search results are optimized by maximizing the joint objective function f until they converge to the best treatment plan that meets the patient's characteristics and stage goals. ; S562, the best treatment plan Dosage recommendations for medications in , Inspection plan and rehabilitation activities Mapping is performed in stages. The specific mapping rules are as follows: During the immediate postoperative recovery phase, the drug dosage is recommended based on the optimization goal of maximizing the rate of recovery of health status. Assigned to the immediate postoperative period The following conditions are met: ； in, The minimum drug dosage threshold to meet the health status recovery rate requirement; During the mid-postoperative recovery period, the examination plan and drug dosing schedule are mapped to time periods based on the optimization goal of minimizing drug side effects. , satisfying the following constraints: ； in, For the cost of drug side effects, is the upper threshold of drug side effects, q is the number of drug types, is the side effect weight of the jth drug; During the long-term follow-up phase, rehabilitation activities are scheduled according to the combined goals of resource balance and patient health status stability. Mapping to time periods , to minimize the following objective function: ； S563. According to the above staged optimization rules, the best treatment plan Mapped into a multi-stage treatment plan where: Output medication dosage recommendations and examination plans during the immediate postoperative recovery phase; In the mid-stage of postoperative recovery, provide drug dosage adjustment plans and side effect control suggestions; During the long-term follow-up phase, rehabilitation activity arrangements and resource allocation plans are output.

Images