A multi-modal information-based hemodialysis intelligent auxiliary diagnosis and treatment method and system

Abstract

The application discloses a kind of based on multi-modal information's hemodialysis wisdom auxiliary diagnosis and treatment method and system, it is related to the field of diagnosis and treatment system, the method includes the following steps: step 1, collect prescription data and sign data, store into hemodialysis knowledge base, and the prescription data and sign data are carried out ETL preprocessing, respectively generate dialysis log and sign log;Step 2, prescription data is trained by means of LDA theme model, obtains the parameter and drug distribution of efficacy theme, the efficacy theme distribution of each dialysis;Step 3, using K-means++ algorithm is carried out cluster analysis to the efficacy theme distribution of each dialysis, the efficacy theme distribution of each dialysis is changed into an efficacy combination label sequence;Step 4, the parameter and weight of convolutional neural network are determined;Step 5, output patient dialysis prescription recommendation.The system includes knowledge acquisition module, reasoning module, explanation module and man-machine interaction module, also contain based on multi-modal information's hemodialysis wisdom auxiliary diagnosis and treatment method.

Description

Technical Field

[0001] This invention relates to the field of diagnostic and treatment systems, and in particular to a smart assisted diagnostic and treatment method and system for hemodialysis based on multimodal information. Background Technology

[0002] The treatment of end-stage renal disease (ESRD) is a major medical challenge. End-stage renal disease, represented by uremia, is the final stage of chronic kidney disease. Due to its high morbidity, high mortality, and severe clinical harm, coupled with the long course of chronic kidney damage—causing extensive glomerular and tubular necrosis and sclerosis—ESRD leads to irreversible and continuous renal function decline in patients. Therefore, ESRD has become another major threat to human health worldwide, following cancer and cardiovascular disease. Hemodialysis is the primary treatment for ESRD. It involves draining blood from the body and passing it through a dialyzer. The blood exchanges substances with a solution containing electrolytes similar to the body's concentration (dialysis fluid) through diffusion, ultrafiltration, adsorption, and convection, achieving the effects of removing metabolic waste, maintaining electrolyte and acid-base balance, and removing excess water. With the number of hemodialysis patients in my country increasing at a rate far exceeding the growth of renal replacement therapy capacity, the demand for hospital dialysis beds is exceeding supply, leading to a rapid increase in medical costs. To enable patients to receive treatment regardless of location or work hours, combining artificial intelligence technology with the dialysis treatment process and using algorithms such as deep learning to improve the dialysis treatment process while ensuring the safety of dialysis is key to improving patients' quality of life. Among these, the design of a personalized intelligent assisted diagnosis and treatment system is of paramount importance.

[0003] Current dialysis treatment procedures have several shortcomings, primarily including: ① Traditional dialysis prescriptions are often delayed, affecting the adequacy of dialysis treatment and easily leading to dialysis complications; ② Dialysis patients exhibit significant individual differences in disease presentation, progression, and treatment response, and the experience and skill levels of dialysis physicians vary, making it easy for some dialysis patients to fall short of expectations in terms of clinical outcomes and prognosis; ③ With the rapid growth of the number of hemodialysis patients nationwide, most dialysis centers face problems such as bed saturation and high patient turnover, posing severe challenges to medical institutions and placing immense pressure on dialysis physicians in clinical decision-making. Therefore, constructing a personalized intelligent auxiliary diagnosis and treatment system based on multimodal information plays a crucial role in ensuring the effectiveness of dialysis treatment and improving patients' quality of life. It also helps to further explore the deep integration of artificial intelligence and medical systems, providing an effective tool for realizing smart healthcare.

[0004] In existing technologies, there is a lack of comprehensive system solutions for assisting dialysis prescription decision-making. Wang Guan et al., in their Chinese invention patent application "A Method for Assisting Decision-Making in Hemodialysis Treatment Plans" (application number CN201911197143.5), provided an auxiliary decision-making scheme for hemodialysis, but this only addressed the dialysate portion of the dialysis patient's prescription and did not propose decision-making schemes for the remaining prescription content. Li Jinsong et al., in their Chinese invention patent application "A Dry Weight Assisting Adjustment System for Hemodialysis Patients Based on Deep Reinforcement Learning" (application number CN202210404618.9), used a deep reinforcement learning scheme to predict the dry weight of dialysis patients, but similarly only proposed a decision-making scheme for a portion of the prescription content.

[0005] Therefore, those skilled in the art are dedicated to developing a new intelligent auxiliary diagnosis and treatment method and system for hemodialysis to address the many shortcomings of the existing technology. Summary of the Invention

[0006] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is how to solve the problem of insufficient hemodialysis and the inability to meet the personalized needs of special patients caused by the lag and repeatability of dialysis fluid prescription.

[0007] To achieve the above objectives, this invention provides a smart auxiliary diagnosis and treatment method and system for hemodialysis based on multimodal information. It is an auxiliary decision-making system based on process mining and deep convolutional neural networks (CNN), which can simulate hemodialysis experts in prescribing medications and improve the adaptability and computational efficiency of the auxiliary decision-making system. It provides patients with personalized and reliable diagnostic prescriptions and treatment plans, and ultimately realizes personalized customization of the treatment process for dialysis patients.

[0008] This invention provides a smart assisted diagnosis and treatment method for hemodialysis based on multimodal information, comprising the following steps:

[0009] Step 1: Collect prescription data and vital sign data, store them in the hemodialysis knowledge base, and perform ETL preprocessing on the prescription data and vital sign data to generate dialysis logs and vital sign logs respectively.

[0010] Step 2: Train the prescription data using an LDA topic model to obtain the parameters and drug distribution corresponding to the efficacy topics, and the efficacy topic distribution for each dialysis session;

[0011] Step 3: Use the K-means++ algorithm to perform cluster analysis on the efficacy theme distribution of each dialysis session. Each cluster generated by the clustering corresponds to an efficacy combination label, thus transforming the efficacy theme distribution of each dialysis session into a sequence of efficacy combination labels.

[0012] Step 4: Input the dialysis log, vital sign log, and efficacy combination label sequence as data into the convolutional neural network for training, determine the parameters and weights of the convolutional neural network, and input the current data of the patient to be predicted into the convolutional neural network for prediction to obtain the feature efficacy theme combination;

[0013] Step 5: Input the feature efficacy theme combination into the XG Boost multi-classification model to obtain the occurrence probability of the efficacy combination label corresponding to the subsequent treatment plan. Then, combine the parameters corresponding to the efficacy theme and the occurrence probability of the drug distribution to output the patient's dialysis prescription recommendation.

[0014] Furthermore, in step 2, the LDA topic model makes the following assumptions:

[0015] There exists a Dirichlet distribution with hyperparameter α that is the prior distribution of the document-topic distribution. A multinomial distribution θ is generated by sampling from this Dirichlet distribution with hyperparameter α. m , with θ m As the distribution of topics in document m; each time from the multinomial distribution θ m The topic z of the nth word in the document m is generated by sampling. m,n ;

[0016] Simultaneously, there exists a Dirichlet distribution with hyperparameter β that is the prior distribution of the topic-word distribution. The topic z is generated by sampling from the Dirichlet distribution with hyperparameter β. m,n Word distribution Then based on the word distribution Generate word w m,n ;

[0017] Under the above assumptions, given a set of documents, the joint probability distribution p of the document keywords is as follows:

[0018]

[0019] Where m and M represent the number of documents and the total number of documents, θ m This represents the topic distribution in the m-th document; k and K represent the topics and the total number of topics, respectively. This represents the word distribution in the k-th topic;

[0020] Using Gibbs sampling, the expected probability θ of topic k appearing in document m can be derived. m,k And the expected probability of word t appearing under topic k. They are respectively:

[0021]

[0022]

[0023] in, Let α represent the number of occurrences of the k-th topic in the m-th document. k is a hyperparameter of topic k, which is a Dirichlet distribution, where k and K represent topics and the total number of topics, respectively. β represents the number of words in the k-th topic, where β is the number of words in the t-th topic. k is a hyperparameter of topic k, which is a Dirichlet distribution, and t and V represent the t-th word and the total number of words, respectively.

[0024] Furthermore, in step 4, the convolutional neural network includes an input layer, several convolutional layers, several pooling layers, a fully connected layer, and an output layer;

[0025] in,

[0026] The input layer concatenates the data from the dialysis log and the vital signs log;

[0027] The convolutional layers and pooling layers will perform several convolutional operations and pooling operations of different sizes on the features, and all pooling layers will use max pooling.

[0028] The present invention also provides a smart auxiliary diagnosis and treatment system for hemodialysis based on multimodal information. The system includes a knowledge acquisition module, a reasoning module, an interpretation module, and a human-computer interaction module. The knowledge acquisition module and the reasoning module include the smart auxiliary diagnosis and treatment method for hemodialysis based on multimodal information as described in any one of claims 1 to 3.

[0029] Furthermore, the knowledge acquisition module is responsible for collecting the prescription data, the vital sign data, relevant literature in the field of dialysis, and the experience and knowledge gained by experts in practice, and storing them in the hemodialysis knowledge base;

[0030] in,

[0031] The prescription data includes demographic indicators, diagnostic information, and dialysis treatment data; the demographic indicators include patient gender, date of birth, date of death, admission time, and discharge time; the diagnostic information includes the dry weight set in the dialysis prescription, the material and area of ​​the dialysis membrane, the dialysis schedule, the amount of water removed per dialysis session, the rate of water removal per unit time, and the type of dialysate; the dialysis treatment data includes the patient's systolic blood pressure, diastolic blood pressure, and heart rate during dialysis.

[0032] Furthermore, the vital signs data include hemoglobin, serum sodium, serum potassium, urea, serum creatinine, predialysis body weight, predialysis diastolic blood pressure, and predialysis systolic blood pressure;

[0033] Each diagnostic information and vital sign data is labeled in chronological order to form the relationship between the data's influence on the current prescription prediction and the time interval from the current time.

[0034] The experience and knowledge gained by experts in practice are combined with medical theories, then processed into rules to form corresponding judgment rules, and then stored in the hemodialysis knowledge base.

[0035] Furthermore, the ETL preprocessing uses multiple interpolation to interpolate the missing prescription data and vital sign data.

[0036] Furthermore, the ETL preprocessing utilizes the Min-Max normalization method to normalize the prescription data and the vital sign data.

[0037] Furthermore, the explanation module connects the hemodialysis knowledge base and the reasoning module, and by displaying the efficacy combination matching situation related to the reasoning process, as well as describing the knowledge and reasoning conclusions used in the reasoning, it realizes the explanation of the prescription process simulated by the auxiliary diagnosis and treatment system.

[0038] Furthermore, the human-computer interaction module connects the knowledge acquisition module and the interpretation module, enabling communication between medical workers, patients, and the auxiliary diagnosis and treatment system; the patient dialysis prescription recommendation is fed back through the human-computer interaction module, and at the same time, the human-computer interaction module improves and enriches the hemodialysis knowledge base by receiving the latest case data.

[0039] The intelligent assisted diagnosis and treatment method and system for hemodialysis based on multimodal information provided by this invention has at least the following technical effects:

[0040] 1. The LDA topic extraction technology was applied to the decision support method and system for hemodialysis machines, which enabled the extraction of prescription parameters and drug efficacy topics from irregular electronic medical records and other information. Furthermore, the extracted topics and corresponding parameters were analyzed to establish a professional knowledge base for the auxiliary system.

[0041] 2. Using the K-means++ algorithm, cluster the distribution of efficacy themes for each dialysis session. Each combination corresponds to a label, thereby converting the patient's dialysis treatment process into a sequence of efficacy combination labels, which enables the extraction and conversion of the patient's dialysis prescription.

[0042] 3. By leveraging the parameter adaptation, massively parallel distributed processing, continuous-time nonlinear dynamics, and global collective action of deep CNN models, knowledge acquisition can be automated, parallel association and adaptive reasoning can be achieved, thereby improving the intelligence, real-time processing capability and robustness of the decision support system.

[0043] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0044] Figure 1 This is a system structure framework diagram of a preferred embodiment of the present invention;

[0045] Figure 2 for Figure 1 A schematic diagram of the prescription recommendation process of the inference module in the illustrated embodiment;

[0046] Figure 3 for Figure 1 A schematic diagram of the convolutional neural network used in the inference module of the illustrated embodiment. Detailed Implementation

[0047] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0048] The technical solution provided in this invention addresses the problems of high clinical pressure in dialysis prescription decision-making caused by the strong lag in traditional dialysis prescriptions, large individual differences among dialysis patients, and the shortage of medical resources. It utilizes artificial intelligence technology to provide decision support for medical treatment. At the same time, the development of a smart assisted diagnosis and treatment system plays a key role in ensuring the effectiveness of dialysis treatment and improving the quality of life of patients. It also helps to further explore the deep integration of artificial intelligence and medical systems, providing an effective tool for realizing smart healthcare.

[0049] Example 1

[0050] like Figure 2 As shown, the intelligent assisted diagnosis and treatment method for hemodialysis based on multimodal information provided by the present invention includes the following steps:

[0051] Step 1: Collect prescription data and vital sign data, store them in the hemodialysis knowledge base, and perform ETL preprocessing on the prescription data and vital sign data to generate dialysis logs and vital sign logs respectively.

[0052] Step 2: Use LDA topic model to train prescription data to obtain parameters and drug distribution corresponding to efficacy topics, and efficacy topic distribution for each dialysis session;

[0053] Step 3: Use the K-means++ algorithm to perform cluster analysis on the efficacy theme distribution of each dialysis session. Each cluster generated by the clustering corresponds to an efficacy combination label, thus transforming the efficacy theme distribution of each dialysis session into a sequence of efficacy combination labels.

[0054] Step 4: Input the dialysis log, vital signs log, and efficacy combination label sequence as data into the Convolutional Neural Network (CNN) for training, determine the parameters and weights of the CNN, and input the current data of the patient to be predicted into the CNN for prediction to obtain the feature efficacy theme combination.

[0055] Step 5: Input the feature efficacy theme combination into the XG Boost multi-classification model to obtain the occurrence probability of efficacy combination labels corresponding to subsequent treatment plans. Then, combine the parameters corresponding to efficacy themes and the occurrence probability of drug distribution to sort the prescription parameters and drugs respectively to obtain the recommendation results, and output the patient dialysis prescription recommendation.

[0056] Example 2

[0057] In step 2 of the method provided in Example 1, the LDA topic model is assumed to be:

[0058] There exists a Dirichlet distribution with hyperparameter α that is the prior distribution of the document-topic distribution. A multinomial distribution θ is generated by sampling from the Dirichlet distribution with hyperparameter α. m , with θ m As the distribution of topics in document m; each time from the multinomial distribution θ m The topic z of the nth word in document m is generated by sampling. m,n ;

[0059] Simultaneously, there exists a Dirichlet distribution with hyperparameter β that is the prior distribution of the topic-word distribution. Topic z is generated by sampling from the Dirichlet distribution with hyperparameter β. m,n Word distribution Then by word distribution Generate word w m,n ;

[0060] Under the above assumptions, given a set of documents, the joint probability distribution p of the document keywords is as follows:

[0061]

[0062] Where m and M represent the number of documents and the total number of documents, θ m This represents the topic distribution in the m-th document; k and K represent the topics and the total number of topics, respectively. This represents the word distribution in the k-th topic.

[0063] in, This represents the probability distribution of the topic.

[0064] According to the Dirichlet distribution:

[0065]

[0066] Represents the normalization factor. Let represent the prior probability distribution corresponding to the m-th topic.

[0067] therefore,

[0068]

[0069] Where, Δ(θ) m +α) is the Dirichlet distribution normalization factor.

[0070] Similarly,

[0071]

[0072] in, Δβ is the Dirichlet distribution normalization factor.

[0073] Using Gibbs sampling, we can inversely deduce the expected probability θ of topic k appearing in document m. m,k And the expected probability of word t appearing under topic k. They are respectively:

[0074]

[0075]

[0076] in, Let α represent the number of occurrences of the k-th topic in the m-th document. k is a hyperparameter of topic k, which is a Dirichlet distribution, where k and K represent topics and the total number of topics, respectively. β represents the number of words in the k-th topic, where β is the number of words in the t-th topic. k is a hyperparameter of topic k, which is a Dirichlet distribution, and t and V represent the t-th word and the total number of words, respectively.

[0077] In this embodiment, a patient's single dialysis medical log is analogized to a document, the prescription parameters and drugs in the medical log are analogized to words, and the medical logs of all dialysis records of all patients with the same disease are analogized to a document set. The "topic" obtained by the LDA topic model above actually corresponds to the efficacy topic of the parameters and drugs, the "topic-word distribution" corresponds to the probability of the parameters and drugs appearing under each efficacy, and the "document-topic distribution" is actually the efficacy distribution corresponding to the prescription for a certain patient in a certain dialysis session.

[0078] Example 3

[0079] like Figure 3 As shown, in step 4 of the method provided in Embodiment 1, the convolutional neural network includes an input layer, several convolutional layers, several pooling layers, a fully connected layer, and an output layer.

[0080] in,

[0081] The input layer concatenates data from dialysis logs and vital signs logs;

[0082] Several convolutional layers and several pooling layers will perform several convolutional operations and pooling operations of different sizes on the features, and all pooling layers will use max pooling.

[0083] Example 4

[0084] like Figure 1 As shown, the present invention also provides a smart auxiliary diagnosis and treatment system for hemodialysis based on multimodal information, including a knowledge acquisition module, a reasoning module, an interpretation module and a human-computer interaction module, wherein the knowledge acquisition module and the reasoning module include the smart auxiliary diagnosis and treatment method for hemodialysis based on multimodal information in Examples 1 to 3.

[0085] The knowledge acquisition module is responsible for collecting prescription data, vital sign data, relevant literature in the field of dialysis, and the experience and knowledge gained by experts in practice, and storing them in the hemodialysis knowledge base;

[0086] in,

[0087] Prescription data includes demographic indicators, diagnostic information, and dialysis treatment data. Demographic indicators include patient gender, date of birth, date of death, admission time, and discharge time. Diagnostic information includes the dry weight set in the dialysis prescription, the material and area of ​​the dialysis membrane, the dialysis schedule, the amount of water removed per dialysis session, the rate of water removal per unit time, and the type of dialysate. Dialysis treatment data includes the patient's systolic blood pressure, diastolic blood pressure, and heart rate during dialysis.

[0088] Vital data include hemoglobin, serum sodium, serum potassium, urea, serum creatinine, pre-dialysis weight, pre-dialysis diastolic blood pressure, and pre-dialysis systolic blood pressure.

[0089] Each diagnostic information and vital sign data is labeled in chronological order to establish the relationship between the data's influence on the current prescription prediction and the time interval from the current date. The experience and knowledge gained by experts in practice are combined with medical theories and then processed into rules to form corresponding judgment rules, which are then stored in the hemodialysis knowledge base.

[0090] The ETL preprocessing in the knowledge acquisition module uses multiple interpolation to interpolate the missing prescription and vital sign data, and uses the Min-Max normalization method to normalize the prescription and vital sign data.

[0091] The reasoning module uses logical and neural network reasoning on the input sample data based on the hemodialysis knowledge base to obtain network output values, which are then transformed into logical concepts for hemodialysis case diagnosis. The reasoning module includes steps 2 to 5 of the intelligent assisted diagnosis and treatment method for hemodialysis based on multimodal information described in Examples 1-3.

[0092] Example 5

[0093] Building upon Example 4, the explanation module connects the hemodialysis knowledge base and the reasoning module. By displaying the matching of efficacy combinations related to the reasoning process, and describing the knowledge and conclusions used in the reasoning, it explains the simulated prescription-issuing process of the assisted diagnostic system. Specific steps include selecting the level of explanation and representing the results of each level of explanation in text or symbols. First, the selection of the explanation level should not be too abstract to provide a substantive explanation of the problem, nor should it be too specific to make the explanation incomprehensible to the user. Therefore, selecting efficacy combination tags in the reasoning process as the basic level of explanation is appropriate, as each tag is an independent data structure unit with a clear meaning. Second, in the step of representing the results in text or symbols, the target efficacy combination tags matching the patient parameters to be predicted and the prescription data generated by the system are converted into text and fed back to the medical worker or patient in the human-computer interaction module.

[0094] Example 6

[0095] Building upon Example 4, the human-computer interaction module connects the knowledge acquisition module and the interpretation module, enabling communication between medical professionals, patients, and the auxiliary diagnosis and treatment system. Specific functions include inputting patient case information and expert diagnostic experience, searching the knowledge base, matching corresponding efficacy combinations to generate prescription suggestions, and then providing feedback on these suggestions via the human-computer interaction system. Simultaneously, the human-computer interaction module can receive the latest case data to improve and enrich the knowledge base.

[0096] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A smart assisted diagnosis and treatment method for hemodialysis based on multimodal information, characterized in that, The method includes the following steps: Step 1: Collect prescription data and vital sign data, store them in the hemodialysis knowledge base, and perform ETL preprocessing on the prescription data and vital sign data to generate dialysis logs and vital sign logs respectively. Step 2: Train the prescription data using an LDA topic model to obtain the parameters and drug distribution corresponding to the efficacy topics, and the efficacy topic distribution for each dialysis session; Step 3: Use the K-means++ algorithm to perform cluster analysis on the efficacy theme distribution of each dialysis session. Each cluster generated by the clustering corresponds to an efficacy combination label, thus transforming the efficacy theme distribution of each dialysis session into a sequence of efficacy combination labels. Step 4: Input the dialysis log, vital sign log, and efficacy combination label sequence as data into the convolutional neural network for training, determine the parameters and weights of the convolutional neural network, and input the current data of the patient to be predicted into the convolutional neural network for prediction to obtain the feature efficacy theme combination; Step 5: Input the feature efficacy theme combination into the XG Boost multi-classification model to obtain the occurrence probability of the efficacy combination label corresponding to the subsequent treatment plan. Then, combine the parameters corresponding to the efficacy theme and the occurrence probability of drug distribution to output the patient dialysis prescription recommendation. In step 2, the LDA topic model makes the following assumptions: There exists a hyperparameter as The Dirichlet distribution is a prior distribution of the document-topic distribution, given the hyperparameters as follows. A multinomial distribution is generated by sampling from the Dirichlet distribution. ,by As a document The distribution of the themes in the middle; each time from the multinomial distribution The document was generated by sampling from the middle. The first in The theme of each word ; At the same time, there exists a hyperparameter as The Dirichlet distribution is a prior distribution of the topic-word distribution, given the hyperparameters as follows. The topic is generated by sampling from the Dirichlet distribution. Word distribution Then based on the word distribution Generate words ; Under the above assumptions, given a set of documents, the joint probability distribution of document topic terms. As shown in the following formula: in, and This indicates the number of documents and the total number of documents. Indicates the first Thematic distribution in the document; and Indicates the topics and the total number of topics. Indicates the first Word distribution in each topic; The document can be derived by using Gibbs sampling. Themes appear in Expected probability and the theme The following words appear Expected probability They are respectively: in, Indicates the first The document in section number 1 The number of topics The topic The hyperparameters of are a Dirichlet distribution. and Indicates the topics and the total number of topics; Indicates the first The first topic The number of words, The topic The hyperparameters of are a Dirichlet distribution. and Indicates the first The number of words and the total number of words.

2. The intelligent assisted diagnosis and treatment method for hemodialysis based on multimodal information as described in claim 1, characterized in that, In step 4, the convolutional neural network includes an input layer, several convolutional layers, several pooling layers, a fully connected layer, and an output layer. in, The input layer concatenates the data from the dialysis log and the vital signs log; The convolutional layers and pooling layers will perform several convolutional operations and pooling operations of different sizes on the features, and all pooling layers will use max pooling.

3. A smart auxiliary diagnosis and treatment system for hemodialysis based on multimodal information, characterized in that, The system includes a knowledge acquisition module, a reasoning module, an explanation module, and a human-computer interaction module, wherein the knowledge acquisition module and the reasoning module include the intelligent assisted diagnosis and treatment method for hemodialysis based on multimodal information as described in any one of claims 1-2.

4. The intelligent auxiliary diagnosis and treatment system for hemodialysis based on multimodal information as described in claim 3, characterized in that, The knowledge acquisition module is responsible for collecting the prescription data, the vital sign data, relevant literature in the field of dialysis, and the experience and knowledge gained by experts in practice, and storing them in the hemodialysis knowledge base; in, The prescription data includes demographic indicators, diagnostic information, and dialysis treatment data; the demographic indicators include patient gender, date of birth, date of death, admission time, and discharge time; the diagnostic information includes the dry weight set in the dialysis prescription, the material and area of ​​the dialysis membrane, the dialysis schedule, the amount of water removed per dialysis session, the rate of water removal per unit time, and the type of dialysate; the dialysis treatment data includes the patient's systolic blood pressure, diastolic blood pressure, and heart rate during dialysis.

5. The intelligent auxiliary diagnosis and treatment system for hemodialysis based on multimodal information as described in claim 4, characterized in that, The vital signs data include hemoglobin, serum sodium, serum potassium, urea, serum creatinine, predialysis weight, predialysis diastolic blood pressure, and predialysis systolic blood pressure. Each diagnostic information and vital sign data is labeled in chronological order to form the relationship between the data's influence on the current prescription prediction and the time interval from the current time. The experience and knowledge gained by experts in practice are combined with medical theories, then processed into rules to form corresponding judgment rules, and then stored in the hemodialysis knowledge base.

6. The intelligent auxiliary diagnosis and treatment system for hemodialysis based on multimodal information as described in claim 3, characterized in that, The ETL preprocessing uses multiple interpolation to interpolate the missing prescription data and vital sign data.

7. The intelligent assisted diagnosis and treatment system for hemodialysis based on multimodal information as described in claim 6, characterized in that, The ETL preprocessing uses the Min-Max normalization method to normalize the prescription data and the vital sign data.

8. The intelligent auxiliary diagnosis and treatment system for hemodialysis based on multimodal information as described in claim 3, characterized in that, The explanation module connects the hemodialysis knowledge base and the reasoning module. By displaying the efficacy combination matching situation related to the reasoning process, as well as describing the knowledge and reasoning conclusions used in the reasoning, it realizes the explanation of the prescription process simulated by the auxiliary diagnosis and treatment system.

9. The intelligent auxiliary diagnosis and treatment system for hemodialysis based on multimodal information as described in claim 3, characterized in that, The human-computer interaction module connects the knowledge acquisition module and the interpretation module, enabling communication between medical workers, patients, and the auxiliary diagnosis and treatment system. The patient's dialysis prescription recommendation is fed back through the human-computer interaction module. At the same time, the human-computer interaction module improves and enriches the hemodialysis knowledge base by receiving the latest case data.

Images