Tumor patient adverse drug reaction intelligent follow-up visit system based on Internet hospital

Abstract

The invention relates to the technical field of medical information, in particular to a tumor patient adverse drug reaction intelligent follow-up visit system based on an internet hospital, which comprises a patient file module for acquiring medical data, identifying bad data and recording the bad data in a medical record file to form an abnormal symptom report; the intelligent analysis module is used for comprehensively analyzing the abnormal symptom report by utilizing a trained machine learning model and outputting analysis data; according to the system and the method, continuous monitoring and active early warning of extrahospital adverse reactions of tumor patients are realized, the timeliness of finding the adverse reactions is remarkably improved, intervention measures can be taken before symptoms of the patients deteriorate, and the treatment safety of the patients is improved. And secondly, through intelligent analysis and grading, the burden of manually screening the patient condition by medical staff is greatly reduced, the system automatically finds abnormity from massive follow-up visit data, and the efficiency and accuracy of follow-up visit management are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and particularly to an intelligent follow-up system for adverse drug reactions of cancer patients based on an Internet hospital. Background Art

[0002] In recent years, with the development of cancer treatment technologies, means such as immunotherapy, targeted therapy, and chemotherapy drugs have been widely applied in clinical practice. However, these treatment means are often accompanied by various adverse reactions. Immunotherapy such as immune checkpoint inhibitors may trigger immune-related adverse reactions (irAE), including rash, colitis, pneumonia, endocrine disorders, etc.; targeted drug therapy may cause side effects such as cardiovascular toxicity, hypertension, skin reactions, etc.; traditional chemotherapy is commonly accompanied by adverse reactions such as nausea, vomiting, fever, leukopenia, and abnormal liver and kidney functions. The above-mentioned adverse reactions may affect the quality of life of patients to a lesser extent and may endanger life in severe cases. Therefore, in the out-of-hospital stage after patients receive treatment, effective monitoring and timely intervention of adverse reactions are of great significance for ensuring patient safety and treatment effects.

[0003] Existing Internet hospital services mostly respond passively to patients' help requests. When patients do not actively report discomfort, the platform will not automatically collect patients' symptom data, nor analyze and prompt potential risks. Developing an intelligent follow-up system for adverse drug reactions of cancer patients based on an Internet hospital is still a key problem that urgently needs to be solved in the field of medical information technology. Summary of the Invention

[0004] The object of the present invention is to solve the problem in the prior art that adverse reactions of cancer patients receiving immunotherapy, targeted therapy, or chemotherapy cannot be continuously monitored, intelligently analyzed, and timely intervened in the out-of-hospital stage.

[0005] To achieve the above object, the present invention provides an intelligent follow-up system for adverse drug reactions of cancer patients based on an Internet hospital, including: a patient file module, which acquires medical data, identifies abnormal data, and records it in the medical record file to form an abnormal symptom report; An intelligent analysis module, which comprehensively analyzes the abnormal symptom report by using a trained machine learning model and outputs analysis data; An automatic warning module, which determines whether the warning condition is reached according to the analysis data. If so, it notifies the Internet hospital to push a warning message to the corresponding doctor terminal; A remote intervention module, which contacts the patient for remote intervention according to the warning message through the reserved information in the medical record file, outputs medical orders, and updates the medical record file; A dynamic adjustment module, which dynamically adjusts the weight parameters of the machine learning model according to the updated medical record file.

[0006] Furthermore, the operation process of the patient file module includes: Obtain medical data, which is used as input. The medical data includes symptom descriptions, physical sign data (body temperature, blood pressure), and the time when discomfort occurs. At this stage, a multivariate gauge field is used to represent different types of medical data in the medical data, and Lie algebra is used for modeling. Through the Ricci-Calabi manifold theory (time-domain evolution), the medical data is evolved in the time domain, and the expression formula is: , where represents the partial derivative of the target variable time with respect to the tensor , is a quantity related to the objective function, represents a regularization coefficient, respectively represent the gradients of the tensor in the and directions, is the logarithm of the determinant of the tensor , is the curvature describing the tensor transformation, represents the penalty coefficient, is the linear rectification unit, is the result of the function acting on the input , is the medical data. When reporting the real-time medical data, an instanton number constraint is introduced for encryption, and the medical data is combined with the adverse reaction knowledge base classification rules to identify and evaluate the symptoms; if the risk level of the symptoms exceeds the preset threshold (the threshold refers to the national medical standard) during the evaluation process, then the symptom is identified as an abnormal symptom and reported and recorded in the medical record file.

[0007] Furthermore, the operation process of the intelligent analysis module includes: Set the machine learning model as: . After the abnormal symptom report is input into the machine learning model, the identified adverse reaction type is output , where is the probability of the th type of adverse reaction occurring under the condition of the given symptom feature , and the expression formula is: , where is the probability of the th type of adverse reaction occurring under the condition of the given symptom feature , is the sum over all types of adverse reaction types, is the sum over all types of adverse reaction types, It is Gaussian kernel function value of adverse reactions, yes and The square of the Euclidean distance, It is The average symptom vector of the adverse reaction class, It is The variance of symptom characteristics of adverse reactions, It is Gaussian kernel function value of adverse reactions, yes and The square of the Euclidean distance, It is The average symptom vector of the adverse reaction class, It is The variance of symptom characteristics of this type of adverse reaction.

[0008] Furthermore, the operation process of the intelligent analysis module includes: According to the identified adverse reaction type, the adverse reaction is graded according to the severity. Each adverse reaction is set to have a corresponding severity function. The output of the severity function is a score between 0 and 1. The identified adverse reaction type and severity classification are combined with the time series prediction model to output the analysis data. The expression formula is: ,in It is The analysis data of time steps, The output range of the Sigmoid activation function is between (0,1). is the weight matrix representing the linear mapping transformation parameters from the input features to the update gate, It is Symptom features of time steps, is the bias vector of the time series forecasting model.

[0009] Furthermore, the operation process of the automatic warning module includes: The warning conditions trigger different levels of warnings according to the severity of the symptoms. A multidimensional Morse function is constructed to measure the priority and severity of the symptoms. The expression formula is: ,in is a function of the input vector The output, is the summation symbol from arrive It means to sum all variables. It is The weight parameter of the item, is the first The square of an input vector, is the rectified linear unit activation function, represents the weighted sum, is the weight vector, is for the analysis data of the time step, is the analysis data of the time step, and the severity is obtained by calculating the critical points of the Morse function to determine whether the symptom reaches the warning condition. The expression formula is: where is the gradient vector of the function the gradient is zero, is the push, represents the weighted sum over all variables, is the row and column element in the coefficient matrix, is the input vector, represents the partial derivative of the input vector with respect to the vector represents the element at the , in the Hessian matrix of the function

[0010] Furthermore, the operation process of the automatic warning module includes: The reaching of the warning condition notifies the Internet hospital to push a warning message to the corresponding doctor's end. The homology class theory is used to classify different symptom severities and determine the level that triggers the warning. The expression formula is: where is the th-order Morse homology group, is the addition operation representing "no intersection" between multiple subspaces, is all sets of critical points of the function is the set of critical points of the is the binary modulo 2 integer ring, is the vector space generator based on the critical point . After calculating the homology class, the warning level is determined by the rank of the homology class. The expression formula for judging the warning level is: where is the warning level, is a matrix is the rank of is a second-order covariance matrix used to describe the degree of association between abnormal symptom indicators is the output value of the severity function of the type of adverse reaction in the interval. When the warning condition is met, a warning message is automatically pushed to the doctor's end of the Internet hospital. The content of the warning message includes patient identification, treatment plan, reported symptoms, and the analyzed severity level

[0011] Furthermore, the operation process of the remote intervention module includes The remote intervention performs personalized intervention on symptoms through a quantum optimization algorithm. The symptom data uses a quantum communication protocol and is transmitted in real time through a fiber bundle structure. A personalized prescription is formulated for the patient using a quantum instanton solution optimization model. The personalized prescription is based on the patient's symptom data and medical history, and adjusts the drug selection and treatment plan according to the principles of quantum mechanics. The expression formula where is the curvature connecting is the Hodge dual of is the self-dual part of represents the standard way to extract the "self-dual" component from is a spinor Weyl spin field is the complex conjugate of is the tensor product of and represents taking the "trace-free" part from the tensor product is the imaginary unit is the externally added "self-dual perturbation term"

[0012] Furthermore, the operation process of the remote intervention module includes Using a quantum error correction algorithm to check the integrity of the personalized prescription. The expression formula where represents the quantum error correction code space is the space of the tensor product of qubit represents the error correction process of information ​​It is a quantum state describing prescription information. According to the warning message, the doctor conducts a video consultation with the patient to understand detailed symptom information and further evaluate the condition. Based on the patient's feedback, the doctor remotely prescribes prescription drugs (including but not limited to painkillers, antipyretics, and glucocorticoid drugs for immune-related adverse reactions) and issues examination and test application forms (including but not limited to blood routine and imaging examinations) on an external Internet hospital, and gives subsequent treatment suggestions. When it is judged that the patient's condition is serious and requires hospital treatment, the doctor sends a notice of advice to the patient to seek medical treatment, outputs medical orders, and updates the medical record file.

[0013] Furthermore, the operation process of the dynamic adjustment module includes: Dynamically adjust the weight parameters of the machine learning model according to the medical record file updated by follow-up. Introduce the Malliavin derivative manifold diffusion equation to describe the response dynamics evolution process of the weight parameters to the input feedback. The expression formula is: , where is the Ricci tensor on the Kähler manifold, is the thermal noise perturbation introduced by Brownian motion, is the coupling tensor between symptom indicators in the high-dimensional feature space, is the corresponding Wiener process, represents at time the th moment of the tiny change in the th machine learning model parameter represents the intensity of the "thermal noise" in the machine learning model, is the th independent path in Brownian motion representing random perturbation, is the Gaussian noise perturbation term,

[0014] Furthermore, the operation process of the dynamic adjustment module includes: Introduce a feedback weight expectation calibration mechanism driven by Gromov-Witten invariants to evaluate the comprehensive adjustment direction of the machine learning model response mechanism to different feedback combinations from geometric and topological perspectives, and then inversely transmit it to the weight parameters. The expression formula is: , where is the follow-up feedback feature term representing the quantum intervention curvature response related to types of symptoms, respectively model the topological feedback variations between the medical record and the weight for the A-roof class, Chern characteristic class, and Todd class, is the symptom-behavior-drug joint space composed of the current feedback, is the symptom evolution bundle and state bundle related to time represents time Take the partial derivative with respect to is a quantum correlation function is the state space metric of the current machine learning model is the corresponding feedback history path in the medical record file is for the feedback space Integrate is a geometric invariant of the tangent space on the feedback manifold, monitors the geometric convergence of the weight parameter changes during the evolution process, and simultaneously performs geometric verification on the evolution path of the weight parameter to judge the physical realizability and adjustability of the machine learning model after dynamic adjustment

[0015] Advantageous effects Adopting the technical solution provided by the present invention, compared with the known public technology, it has the following advantageous effects The present invention realizes continuous monitoring and active warning of the off-hospital adverse reactions of tumor patients, significantly improves the timeliness of discovering adverse reactions, can take intervention measures before the patient's symptoms deteriorate, and improves the treatment safety of the patient. Secondly, through intelligent analysis and grading, the burden on medical staff for manually screening the patient's condition is greatly reduced, and the system automatically discovers abnormalities from a large amount of follow-up data, improving the efficiency and accuracy of follow-up management. Thirdly, the present invention makes full use of the existing platform foundation of the Internet hospital. Patients can obtain medical monitoring through familiar online channels without installing new applications or making frequent trips to the hospital; doctors can also use the existing remote consultation and electronic prescription systems for intervention, reducing the threshold for system deployment and promotion. Finally, the system is specifically optimized for the adverse reaction characteristics of new treatment methods such as immunotherapy and targeted therapy, can more accurately identify relevant adverse reaction types and provide personalized follow-up plans, thereby better ensuring the treatment continuity and efficacy of tumor patients Brief description of the drawings

[0016] Figure 1 is the system diagram of the intelligent follow-up system for drug adverse reactions of tumor patients based on the Internet hospital of the present invention Detailed implementation manners

[0017] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the 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

[0018] It should be noted that the terms "first", "second", etc. in the description, claims and drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0019] The present invention will be further described in detail below with reference to the accompanying drawings: Embodiment: As Figure 1 shown, the present invention provides an intelligent follow-up system for adverse drug reactions of cancer patients based on an Internet hospital, including: a patient file module, which acquires medical data, identifies adverse data and records it in the medical record file to form an abnormal symptom report; Further, the operation process of the patient file module includes: Acquiring medical data, the medical data is used as input, and the medical data includes symptom descriptions, physical sign data (body temperature, blood pressure, various blood reports and imaging examinations), and the time when discomfort occurs. At this stage, a multivariate gauge field is used to represent different types of medical data in the medical data, and Lie algebra is used for modeling. Through the Ricci-Calabi manifold theory (time-domain evolution), the medical data is evolved in the time domain, and the expression formula is: , where represents the partial derivative of the target variable time with respect to the tensor , is a quantity related to the objective function, represents a regularization coefficient, respectively represent the gradients of the tensor in the and directions, is the logarithm of the determinant of the tensor , is the curvature describing the tensor transformation, represents a penalty coefficient, is a linear rectification unit, is the result of the function acting on the input , It is medical data. When reporting medical data in real time, instanton number constraints are introduced for encryption. According to preset rules, bad data is screened out from the medical data. For example, when the body temperature exceeds the normal human temperature, the body temperature data is regarded as bad data; the bad data is combined with the grading rules of the adverse reaction knowledge base to identify and evaluate the symptoms; during the evaluation process, if the risk level of the symptoms exceeds the preset threshold (the threshold refers to national medical standards), then the symptoms are identified as abnormal symptoms and reported and recorded in the medical record file. Taking immune pneumonia as an example, the grading rules for immune therapy-related adverse reactions are shown in Table 1: Table 1

[0020] Specifically, before the patient starts immune therapy, targeted therapy or chemotherapy, the patient's basic information, diagnosis, treatment plan and potential high-risk adverse reactions and other data are entered into the Internet hospital platform, and a follow-up management service is opened for the patient. The patient fills in the adverse reaction questionnaire regularly through the patient's smart wearable device and smartphone device according to the follow-up plan outside the hospital. The patient data is uploaded in real time through the smart wearable device to ensure that the doctor can obtain accurate health information in the first time. The real-time monitoring and feedback system can effectively reduce the risk of sudden illness. By introducing instanton number constraint encryption technology, the patient's personal health data is effectively protected during the transmission process, ensuring the privacy and security of users and enhancing the patient's trust in the smart health monitoring system.

[0021] The intelligent analysis module uses the trained machine learning model to comprehensively analyze the abnormal symptom report and outputs analysis data. Furthermore, the operation process of the intelligent analysis module includes: Set the machine learning model as: , after the abnormal symptom report is input into the machine learning model, the identified adverse reaction type is output , where is the given symptom feature under the condition of the probability of the th type of adverse reaction occurring, expression formula: , where is the given symptom feature under the condition of the probability of the th type of adverse reaction occurring, is the sum over all types of adverse reaction types, is the Gaussian kernel function value of the th type of adverse reaction, is and The square of the Euclidean distance, is the average symptom vector of the type of adverse reaction, is the variance of the symptom characteristics of the type of adverse reaction, is the Gaussian kernel function value of the type of adverse reaction, is and The square of the Euclidean distance, is the average symptom vector of the type of adverse reaction, is the Variance of the symptom characteristics of the type of adverse reaction; Furthermore, the operation process of the intelligent analysis module includes: According to the identified type of adverse reaction, grade it according to the severity, set a corresponding severity function for each adverse reaction, the output of the severity function is a score between 0 and 1, and use the identified type of adverse reaction and severity grading, combined with the time series prediction model, to output analysis data, expression formula: where is the analysis data at the th time step, is the output range of the Sigmoid activation function between (0,1), is the weight matrix representing the linear mapping transformation parameters from the input features to the update gate, is the symptom characteristics at the th time step, is the bias vector of the time series prediction model; Specifically, the intelligent analysis module of this system comprehensively analyzes the abnormal symptom reports through a machine learning model, identifies the potential types of adverse reactions of the patient, and grades them according to their severity. Moreover, the analysis process can combine the patient's medical history and recent test results to improve the accuracy. By accurately identifying the types of adverse reactions and evaluating their occurrence probabilities, it provides key clinical decision-making support for doctors. Each adverse reaction has a quantified severity score, enabling doctors to clearly understand the criticality of the symptoms and avoid omissions or misjudgments caused by human factors. Combined with the time series prediction model, this system can not only identify the current adverse reactions but also predict the future disease trend, helping doctors make adjustments in advance to prevent the symptoms from deteriorating.

[0022] The automatic warning module determines whether the warning condition is reached according to the analysis data. If so, it notifies the Internet hospital to push a warning message to the corresponding doctor terminal; Furthermore, the operation process of the automatic warning module includes: The warning conditions trigger warnings at different levels according to the severity of the symptoms. A multi-dimensional Morse function is constructed to measure the priority and severity of the symptoms. The expression formula is: , where is the output of the function with respect to the input vector ; is the summation symbol from to indicating the summation over all variables; is the weight parameter of the -th term; is the square of the -th input vector in the input vector; is the rectified linear unit activation function; represents the weighted summation; is the weight vector; is the analysis data for the -th time step after softmax normalization; is the analysis data for the -th time step. The severity is obtained by calculating the critical points of the Morse function to determine whether the symptoms reach the warning conditions. The expression formula is: , where is the gradient vector of the function ; means the gradient is zero; is "implies"; represents the weighted summation over all variables; is the element in the -rd row and -th column of the coefficient matrix; is the -th input vector; represents the partial derivative of the -th input vector with respect to the vector ; represents the -th element in the Hessian matrix of the function , -th; Furthermore, the operation process of the automatic warning module includes: When the warning conditions are met, it notifies the Internet hospital to push a warning message to the corresponding doctor terminal. The homology class theory is used to classify different symptom severities and determine the warning level triggered. The expression formula is: , where is the -th Morse homology group; is the addition operation indicating "no intersection" between multiple subspaces; is all The set composed of critical points of order is the -th order critical point set of the function ; is the binary modulo 2 integer ring ; and is the generator of the vector space based on the critical point . After calculating the homology class, the warning level is determined by the rank of the homology class. The expression formula for judging the warning level is: , where is the warning level, is the rank of the matrix , is the second-order covariance matrix used to describe the correlation degree between abnormal symptom indicators, is the -th output value of the severity function of adverse reactions of the category in the interval . When the warning condition is satisfied, a warning message is automatically pushed to the doctor's end of the Internet hospital. The content of the warning message includes patient identification, treatment plan, reported symptoms, and the analyzed severity level;

[0023] The remote intervention module contacts the patient for remote intervention according to the warning message through the reserved information in the medical record file, outputs medical orders, and updates the medical record file; Furthermore, the operation process of the remote intervention module includes: The remote intervention performs personalized intervention on symptoms through a quantum optimization algorithm. The symptom data adopts a quantum communication protocol and is transmitted in real time through a fiber bundle structure. A personalized prescription is formulated for the patient using a quantum instanton solution optimization model. The personalized prescription is based on the patient's symptom data and medical history, and adjusts drug selection and treatment plan according to the principles of quantum mechanics. The expression formula is: , where is the curvature connecting , is the Hodge dual of The self-dual part, denotes the standard way to extract the "self-dual" component from a spinor Weyl spin field, which is the complex conjugate of and is the tensor product of with denoting the extraction of the "trace-free" part from the tensor product, where \(i\) is the imaginary unit, and \(\xi\) is the externally added "self-dual perturbation term"; Furthermore, the operation process of the remote intervention module includes: Using a quantum error correction algorithm to check the integrity of the personalized prescription, with the expression formula: , where represents the quantum error correction code space, is the space of the tensor product of \(n\) qubits, represents the error correction process of the information, is the set of error patterns, and \(\vert\psi\rangle\) is the quantum state describing the prescription information. According to the warning message, the doctor conducts a video consultation with the patient to understand the detailed symptom information and further evaluate the condition. Based on the patient's feedback, the doctor remotely prescribes prescription drugs (including but not limited to painkillers, antipyretics, and glucocorticoid drugs for immune-related adverse reactions) and issues inspection and test requisition forms (including but not limited to blood routine and imaging examinations) on the external Internet hospital, and gives follow-up treatment suggestions. When it is judged that the patient's condition is serious and requires hospital treatment, the doctor sends a notice of advice to the patient to seek medical treatment, outputs medical orders and updates the medical record file; Specifically, the remote intervention module in this system uses intelligent technology, quantum optimization algorithms, and quantum communication protocols to perform precise intervention when the patient's condition is abnormal. Based on the patient's historical data and the latest symptoms in the medical record file, it provides a personalized treatment plan and interacts with the doctor through remote video consultation, which is beneficial to further optimizing the treatment process and improving the treatment effect. Through the quantum error correction algorithm, it improves the accuracy and integrity of the personalized prescription, avoids potential errors in data transmission, and ensures patient safety. After the initial intervention, this system continues to follow up and monitor the patient. For the early warning events that have been issued, this system will remind the patient to feedback the symptom changes after a set time interval until the adverse reaction is relieved or stabilized. The doctor can adjust the treatment plan or follow-up frequency according to the patient's subsequent feedback to ensure that the adverse reaction is under closed-loop management.

[0024] The dynamic adjustment module dynamically adjusts the weight parameters of the machine learning model according to the updated medical record file; Furthermore, the operation process of the dynamic adjustment module includes: Dynamically adjust the weight parameters of the machine learning model according to the medical record file updated by follow-up, and introduce the Malliavin derivative manifold diffusion equation to describe the response dynamics evolution process of the weight parameters to the input feedback. The expression formula is: , where is the Ricci tensor on the Kähler manifold, is the thermal noise perturbation introduced by Brownian motion, is the coupling tensor between symptom indicators in the high-dimensional feature space, is the corresponding Wiener process, represents at time the th small change in the machine learning model parameter , represents the intensity of "thermal noise" in the machine learning model, is the th independent path in Brownian motion representing random perturbation, is the Gaussian noise perturbation term, is a structural perturbation term; Furthermore, the operation process of the dynamic adjustment module includes: Introduce a feedback weight expectation calibration mechanism driven by Gromov-Witten invariants to evaluate the comprehensive adjustment direction of the machine learning model response mechanism to different feedback combinations from geometric and topological perspectives, and then backpropagate it to the weight parameters. The expression formula is: , where is the follow-up feedback feature term representing the quantum intervention curvature response related to types of symptoms, respectively model the topological feedback variations between the medical record - weight for the A-roof class, Chern characteristic class, and Todd class, is the symptom-behavior-drug joint space formed by the current feedback, is the symptom evolution bundle and state bundle related to time, represents taking the partial derivative with respect to time , is a quantum correlation function, is the state space metric of the current machine learning model, is the corresponding feedback history path in the medical record file, is the integration over the feedback space , It is a geometric invariant of the tangent space on the feedback manifold, monitoring the geometric convergence of the weight parameter changes during the evolution process, and at the same time performing geometric verification on the evolution path of the weight parameters to judge the physical realizability and adjustability of the machine learning model after dynamic adjustment; Specifically, the design of the dynamic adjustment module in this system is to intelligently adjust the weight parameters of the machine learning model according to the updated medical records of patient follow-up, so as to improve the accuracy of disease prediction, treatment effect evaluation and personalized diagnosis and treatment. Consider a real clinical scenario: Suppose there is a lung cancer patient receiving PD-1 immunotherapy, who regularly uploads immune-related adverse reaction indicators (such as thyroid function TSH / FT4 levels, creatine kinase CK-MB values), treatment response characteristics (tumor shrinkage ratio in CT images), and symptom complaints (number of diarrhea, severity of rash) through intelligent wearable devices. Each updated medical record will be deeply analyzed by this system, and the warning threshold will be dynamically optimized based on the immune therapy-related adverse reaction database and current data characteristics. When the patient presents with progressive skin erythema accompanied by itching, this system will combine T cell proliferation-related biomarkers (such as IL-6 levels) and pathological biopsy results, and automatically adjust the decision boundary weight of the mucosal toxicity characteristics in the classification model through the introduced manifold diffusion equation model. For example, if the rash area of the patient expands to 30% of the body surface within 72 hours and is accompanied by fever symptoms, the system will increase the weight coefficients of inflammatory cytokine characteristics (IL-17α, IFN-γ) in the epidermal toxicity grading model, so as to more sensitively identify the progression risk of immune-related skin adverse reactions.

[0025] Through the topological feedback mechanism, this system performs geometric manifold mapping on the multi-dimensional characteristics of treatment toxicity reactions: when it detects a spatial association between the pathological characteristics of immune enteritis shown in colonoscopy (such as crypt abscess) and fecal calprotectin > 500 μg / g, the system will establish an intestinal mucosal injury gradient field through the Chern characteristic class, and reconstruct the curvature tensor relationship of intestinal symptoms - inflammation indicators - imaging characteristics. Based on the quantum correlation function analysis of the symptom remission trajectory after hormone shock therapy (such as the quantum state decay process of diarrhea frequency), accurately evaluate the efficacy of different intervention plans (methylprednisolone dose gradient) on intestinal toxicity, and realize personalized step-down adjustment of treatment plans.

[0026] This dynamic adjustment mechanism enables the system to continuously track the toxicity evolution process in tumor treatment. When the patient's neutrophil count persists < 1.5×10 9When it is / L, the system will optimize the prediction model of the G-CSF (granulocyte colony-stimulating factor) usage strategy through the self-dual perturbation equation, and at the same time dynamically adjust the chemotherapy dose recommendation parameters in combination with the CD34+ cell count. This adaptive learning based on multi-modal data can ensure the early identification of dose-limiting toxicity and ensure that the treatment plan maintains the best balance between efficacy and safety.

[0027] This system will optimize the drug adjustment strategy according to the joint space analysis of historical symptoms-drugs-behaviors, and accurately evaluate the impact of different treatment methods on the patient's condition through the quantum correlation function, which is conducive to realizing accurate disease prediction and personalized treatment plan adjustment, responding in real time to the evolution of the patient's condition, ensuring that the treatment plan is always in the best state. Based on the automatic adjustment of the machine learning model in this system, doctors can understand the changes and potential risks of patients faster, thus making timely intervention decisions, saving medical resources and improving the diagnosis and treatment efficiency.

[0028] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent follow-up system for adverse drug reactions in cancer patients based on Internet hospitals, characterized by: include: Patient file module, which obtains medical data, identifies adverse data and records them in medical records to form abnormal symptom reports; An intelligent analysis module uses a trained machine learning model to conduct a comprehensive analysis of the abnormal symptom report and output analysis data; The automatic warning module determines whether the warning condition is met based on the analysis data, and if so, notifies the Internet hospital to push a warning message to the corresponding doctor; A remote intervention module, which contacts the patient through the reserved information in the medical record file according to the warning message to perform remote intervention, outputs medical advice and updates the medical record file; A dynamic adjustment module dynamically adjusts the weight parameters of the machine learning model according to the updated medical record file; The operation process of the patient file module includes: Acquiring medical data, the medical data being used as input, the medical data comprising actively acquired symptom descriptions, physical sign data, and the time when the discomfort occurred; encrypting the medical data; According to preset rules, bad data is obtained from medical data and recorded in medical records; Combining the adverse reaction data with the adverse reaction knowledge base classification rules to identify and evaluate symptoms; If the risk level of a symptom exceeds a preset threshold during the assessment process, the symptom will be identified as an abnormal symptom report and recorded in the medical record.

2. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 1 is characterized in that: The encrypting the medical data comprises: Using multivariate gauge fields to represent different types of medical data in the medical data, and using Lie algebra to model, and through Ricci-Calabi manifold theory, the medical data is evolved in the time domain; When reporting medical data in real time, instantaneous sub-number constraints are introduced for encryption.

3. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 2 is characterized in that: The operation process of the intelligent analysis module includes: The machine learning model is set as: After the abnormal symptom report is input into the machine learning model, the identified adverse reaction type is output ,in Is a given symptom characteristic The first Adverse reactions The probability of is expressed as: ， in Is a given symptom characteristic The first Adverse reactions The probability of For all The sum of the adverse reaction types is It is Gaussian kernel function value of adverse reactions, yes and The square of the Euclidean distance, It is The average symptom vector of the adverse reaction class, It is The variance of symptom characteristics of adverse reactions, It is Gaussian kernel function value of adverse reactions, yes and The square of the Euclidean distance, It is The average symptom vector of the adverse reaction class, It is The variance of symptom characteristics of this type of adverse reaction.

4. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 3 is characterized in that: The operation process of the intelligent analysis module includes: Based on the type of adverse reaction identified, grade it according to its severity; It is assumed that each adverse reaction has a corresponding severity function, the output of which is a score between 0 and 1. The identified adverse reaction type and severity grade are combined with a time series prediction model to output analysis data.

5. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 4 is characterized in that: The operation process of the automatic warning module includes: The warning conditions trigger different levels of warnings according to the severity of the symptoms, and a multidimensional Morse function is constructed to measure the priority and severity of the symptoms; The severity is obtained by calculating the critical point of the Morse function to determine whether the symptoms meet the warning conditions.

6. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 5 is characterized in that: The operation process of the automatic warning module includes: When the warning condition is met, the Internet hospital is notified to push a warning message to the corresponding doctor, and the homology theory is used to classify the severity of different symptoms and determine the level of triggering the warning; After calculating the homology class, the warning level is determined by the rank of the homology class. The warning level expression formula for determining the warning level is: in It is the warning level. is a matrix rank, is the second-order covariance matrix used to describe the degree of correlation between abnormal symptom indicators. It is The output value of the severity function of the adverse reaction is in the interval ; When the warning conditions are met, a warning message is automatically pushed to the doctor side of the Internet hospital. The content of the warning message includes the patient identification, treatment plan, reported symptoms and the severity level obtained by analysis.

7. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 6 is characterized in that: The operation process of the remote intervention module includes: The remote intervention uses a quantum optimization algorithm to perform personalized intervention on symptoms. Symptom data is transmitted in real time through a fiber bundle structure using a quantum communication protocol, and a quantum instanton solution optimization model is used to formulate personalized prescriptions for patients. The personalized prescription is based on the patient's symptom data and medical history, and adjusts the drug selection and treatment plan according to the principles of quantum mechanics, extracting the expression formula from the dual component: in Yes Connect The curvature of yes Hodge duality, The self-dual part of is, Indicates from The standard way to extract the dual component is, is a spin or Weyl spin field, yes The complex conjugate of yes and The tensor product of means taking out the part with zero trace from the tensor product, is an imaginary unit, is an added self-dual perturbation term.

8. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 7 is characterized in that: The remote intervention module uses a quantum error correction algorithm to check the integrity of the personalized prescription. The formula of the quantum error correction code space V is: in represents the quantum error correction code space, yes The space of tensor products of qubits, Represents the error correction process of information, is the set of error patterns, is the quantum state that describes the prescription information; According to the warning message, the doctor conducts a video consultation with the patient to understand the detailed symptom information and further evaluate the condition. Based on the patient's feedback, the doctor remotely prescribes drugs and issues examination and test application forms on the external Internet hospital and gives follow-up treatment suggestions. If it is judged that the patient's condition is serious and needs to be treated in the hospital, the doctor will issue a medical advice notice to the patient, output medical advice and update the medical record file.

9. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 8 is characterized in that: The operation process of the dynamic adjustment module includes: The weight parameters of the machine learning model are dynamically adjusted according to the updated medical records, and the small changes The expression formula is: in is the Ricci tensor on the Kähler manifold, is the thermal noise disturbance introduced by Brownian motion, is the coupling tensor between symptom indicators in the high-dimensional feature space, is the corresponding Wiener process, represents the intensity of thermal noise in the machine learning model, It is the first The independent paths represent random perturbations, is the Gaussian noise disturbance term, is the structural disturbance term.

10. The intelligent follow-up system for adverse drug reactions in tumor patients based on Internet hospitals according to claim 9 is characterized in that: The operation process of the dynamic adjustment module includes: The feedback weight expectation calibration mechanism driven by the Gromov-Witten invariant is introduced, and the expression formula is: in is the follow-up feedback feature item representative of Quantum intervention curvature response related to symptom-like conditions, Model the topological feedback variation between medical records and weights for A-roof class, Chern feature class and Todd class respectively. is the symptom-behavior-drug joint space composed of current feedback, It is a time-related symptom evolution cluster and state cluster, Indicates time Find the partial derivative, is a quantum correlation function, is the state space metric of the current machine learning model, is the corresponding feedback history path in the medical record file, It is the feedback space To perform integration, is a geometric invariant of the tangent space on the feedback manifold.

Images