Data analysis system and method for intraperitoneal hyperthermic perfusion chemotherapy

Abstract

The invention provides a data analysis system and method for intraperitoneal hyperthermic perfusion chemotherapy, and relates to the field of data intelligent analysis. Basic information and illness state characteristics of a patient are subjected to embedded coding by using a data processing technology based on deep learning; splicing the embedded patient basic information embedded features and the illness state embedded features, and meanwhile, carrying out semantic embedded coding on various adaptation disease screening rules; therefore, whether the intraperitoneal hyperthermic perfusion chemotherapy is suitable for being used as a first-line treatment strategy or an auxiliary treatment means in an operation or not is intelligently analyzed according to the selection range ratio query response representation between the patient information-illness state cascade coding feature and the screening rule semantic embedding feature of each indication. Therefore, more accurate and timely treatment strategy suggestions can be provided, and doctors can be helped to make more scientific and reasonable decisions.

Description

Technical Field

[0001] The present application relates to the field of intelligent data analysis, and more specifically, to a data analysis system and method for intraperitoneal hyperthermic perfusion chemotherapy. Background Art

[0002] Hyperthermic intraperitoneal chemotherapy is a local treatment method used to treat certain types of cancer (such as colorectal cancer, ovarian cancer, etc.). It combines surgical removal of the tumor with direct infusion of heated chemotherapy drugs into the abdominal cavity during surgery to increase the drug's ability to kill tumor cells while reducing systemic side effects.

[0003] Traditional techniques often rely on general and fixed rules when evaluating the suitability of hyperthermic intraperitoneal chemotherapy, which mainly determines the treatment strategy based on the general classification of cancer type and stage. However, this approach ignores the significant individual differences between patients. In addition to cancer type and stage, the patient's basic health status also has a key impact on the treatment effect. Traditional technologies make it difficult to comprehensively consider these complex individual factors for personalized analysis, resulting in inaccurate treatment decisions. In addition, the patient's condition changes dynamically, and new situations may arise during the treatment process. For example, the patient's suitability for hyperthermic intraperitoneal chemotherapy changes after receiving other treatments, and the patient's physical condition changes over time. Traditional technologies lack flexibility in the face of these dynamic changes and cannot adjust treatment strategy recommendations in a timely manner.

[0004] Therefore, an optimized data analysis protocol for hyperthermic intraperitoneal chemotherapy is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a data analysis system and method for hyperthermic intraperitoneal chemotherapy.

[0006] According to one aspect of the present application, a data analysis system for intraperitoneal hyperthermic chemotherapy is provided, comprising: Patient information acquisition module, used to obtain the patient's basic information and disease characteristics; A screening rule extraction module, used to extract a set of indication screening rules; A basic information-condition characteristic embedding fusion module is used to embed the basic information and condition characteristics of the patient to obtain a patient basic information embedding coding vector and a condition characteristic embedding coding vector, and concatenate the patient basic information embedding coding vector and the condition characteristic embedding coding vector to obtain a patient information-condition cascade coding vector; An indication screening rule semantic embedding module, used for performing semantic embedding coding on each indication screening rule in the set of indication screening rules to obtain a set of indication screening rule semantic embedding coding vectors; A patient condition-screening rule query response module, used for performing a dynamic semantic query response of a selection range ratio on the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule query response coding vector; The analysis result generation module is used to query the response coding vector based on the patient condition-screening rule to obtain an analysis result indicating whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an auxiliary treatment means during surgery.

[0007] Furthermore, the basic information includes ECOG score, nutritional assessment, whether there are underlying diseases, whether chemotherapy, surgery or radiotherapy has been received, and whether there is drug resistance or chemotherapy insensitivity; the disease characteristics include cancer type, cancer stage, tumor volume burden and pathological grade, whether there are completely resectable macroscopic lesions and tumor markers.

[0008] Furthermore, the basic information-disease characteristics are embedded in a fusion module, including: A patient basic information and condition characteristics embedding coding unit, used for embedding the patient's basic information and condition characteristics using one-hot coding to obtain the patient basic information embedding coding vector and the condition characteristics embedding coding vector; The patient information-condition cascade coding feature generating unit is used to concatenate the patient basic information embedded coding vector and the condition feature embedded coding vector to obtain the patient information-condition cascade coding vector.

[0009] Furthermore, the indication screening rule semantic embedding module is used to: use a semantic embedding encoder based on the Bert model to perform semantic embedding encoding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding encoding vectors of the indication screening rules.

[0010] Furthermore, the patient condition-screening rule query response module includes: A patient condition-screening rule anchoring unit, used for dynamically anchoring the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule dynamic search anchoring vector; an indication screening rule selection range ratio determination unit, configured to determine the indication screening rule selection range ratio based on the characteristic distribution characteristics of the patient condition-screening rule dynamic search anchor vector, wherein the vector at the starting position of the indication screening rule selection range ratio is the patient condition-screening rule dynamic search anchor vector, and each indication screening rule semantic embedding coding vector in the indication screening rule selection range ratio is defined as an indication screening rule dynamic search optimization library feature vector; The dynamic semantic search unit is used to perform dynamic semantic search encoding on the patient information-condition cascade encoding vector and each indication screening rule dynamic search optimization library feature vector in the indication screening rule selection range ratio to obtain the patient condition-screening rule query response encoding vector.

[0011] Furthermore, the patient condition-screening rule anchoring unit is used to: Calculating the internal relationship score value of the patient information-condition cascade coding vector relative to each indication screening rule semantic embedding coding vector in the set of indication screening rule semantic embedding coding vectors to obtain a set of patient condition-screening rule internal relationship score values; The indication screening rule semantic embedding coding vector corresponding to the largest patient condition-screening rule internal relationship score value in the set of patient condition-screening rule internal relationship score values ​​is used as the patient condition-screening rule dynamic search anchor vector.

[0012] Furthermore, the dynamic semantic search unit is used to: Respectively calculating the cosine similarity between the patient information-condition cascade coding vector and the feature vector of each indication screening rule dynamic search optimization library in the indication screening rule selection range ratio to obtain a set of patient condition-screening rule semantic similarity; Taking the set of patient condition-screening rule semantic similarities as a weighted sequence, calculating the weighted sum of semantic differences between the patient information-condition cascade coding vector and the feature vectors of each indication screening rule dynamic search optimization library in the indication screening rule selection range ratio to obtain a patient condition-screening rule semantic measurement vector; The patient condition-screening rule semantic metric vector and the patient condition-screening rule dynamic search anchor vector are added point by point to obtain the patient condition-screening rule query response encoding vector.

[0013] Further, it is used to: input the patient condition-screening rule query response encoding vector into a classifier-based data analysis auxiliary decision-making module to obtain the analysis result, and the analysis result is used to indicate whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an auxiliary treatment method during surgery.

[0014] According to another aspect of the present application, a data analysis method for intraperitoneal hyperthermic chemotherapy is provided, comprising: Obtain the patient's basic information and disease characteristics; Extracting a set of indication screening rules; Embedding and coding the basic information and condition characteristics of the patient to obtain a patient basic information embedded coding vector and a condition characteristic embedded coding vector, and concatenating the patient basic information embedded coding vector and the condition characteristic embedded coding vector to obtain a patient information-condition cascade coding vector; Performing semantic embedding coding on each indication screening rule in the set of indication screening rules to obtain a set of indication screening rule semantic embedding coding vectors; Performing a dynamic semantic query response of a selection range ratio on the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule query response coding vector; Based on the patient condition-screening rule query response encoding vector, an analysis result is obtained to indicate whether intraperitoneal hyperthermic chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment means during surgery.

[0015] Furthermore, each indication screening rule in the set of indication screening rules is semantically embedded and encoded to obtain a set of semantic embedded encoding vectors of indication screening rules, including: using a semantic embedding encoder based on the Bert model to semantically embed code each indication screening rule in the set of indication screening rules to obtain a set of semantic embedded encoding vectors of indication screening rules.

[0016] Compared with the prior art, the data analysis system and method for intraperitoneal hyperthermic perfusion chemotherapy provided by the present application uses a data processing technology based on deep learning to embed and encode the patient's basic information and disease characteristics, then splices the embedded patient basic information embedding features and disease characteristics embedding features, and at the same time, semantic embedding codes are performed on each indication screening rule, so as to intelligently analyze whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment during surgery based on the selection range ratio between the patient information-disease condition cascade coding features and the semantic embedding features of each indication screening rule according to the query response representation. In this way, it is conducive to providing more accurate and timely treatment strategy recommendations, thereby helping doctors make more scientific and reasonable decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 4 is a system block diagram of a data analysis system for intraperitoneal hyperthermic chemotherapy according to an embodiment of the present application.

[0018] Figure 2 The present invention is a block diagram of a basic information-disease characteristic embedding fusion module in a data analysis system for intraperitoneal hyperthermic chemotherapy according to an embodiment of the present application.

[0019] Figure 3 The block diagram is a patient condition-screening rule query response module in a data analysis system for intraperitoneal hyperthermic chemotherapy according to an embodiment of the present application.

[0020] Figure 4 The present invention is a flowchart of a data analysis method for hyperthermic intraperitoneal chemotherapy according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0022] Hyperthermic intraperitoneal chemotherapy is a localized treatment for certain cancer types, such as colorectal and ovarian cancer. This treatment combines tumor removal surgery with the intraoperative administration of heated chemotherapy drugs directly into the abdominal cavity, aiming to enhance the effect of anticancer drugs and reduce systemic side effects.

[0023] Traditional techniques often follow a broad and fixed set of criteria when evaluating the suitability of hyperthermic intraperitoneal chemotherapy, mainly based on the type and clinical stage of cancer to formulate treatment plans. However, this approach fails to fully take into account individual differences. In addition to the type and stage of cancer, the patient's overall health status also has a significant impact on treatment outcomes. Traditional techniques are difficult to integrate these complex and variable personal factors for detailed personalized analysis, which may lead to inaccurate treatment choices. In addition, given that the disease process is dynamic, new conditions may be encountered during treatment. For example, a patient's suitability for hyperthermic intraperitoneal chemotherapy may change after receiving other forms of treatment; or the patient's physical condition may change over time. Traditional techniques are often not flexible enough in the face of such changing situations, and it is difficult to adjust treatment strategy recommendations in a timely manner.

[0024] In response to the above technical problems, the technical concept of the present application is to obtain the basic information and disease characteristics of the patient, and extract a set of indication screening rules, and use data analysis and coding methods based on deep learning to embed the basic information and disease characteristics. Then, the embedded patient basic information embedding features and disease characteristics embedding features are spliced, and each indication screening rule is semantically embedded and encoded, so as to intelligently analyze whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment during surgery based on the selection range ratio between the patient information-disease cascade coding features and the semantic embedding features of each indication screening rule according to the query response representation. Unlike traditional technologies that rely on general and fixed rules, the present application can fully obtain the basic information and disease characteristics of the patient, and can ensure that the unique information of each patient is fully considered. At the same time, the application can quickly respond to changes in the patient's condition and physical condition, and provide the latest treatment strategy recommendations, thereby helping doctors make more scientific and reasonable decisions.

[0025] Figure 1 FIG. 4 is a system block diagram of a data analysis system for intraperitoneal hyperthermic chemotherapy according to an embodiment of the present application. Figure 1As shown, in the data analysis system 100 for intraperitoneal hyperthermic chemotherapy, it includes: a patient information acquisition module 110, which is used to acquire the basic information and disease characteristics of the patient; a screening rule extraction module 120, which is used to extract a set of indication screening rules; a basic information-disease characteristic embedding fusion module 130, which is used to embed the basic information and disease characteristics of the patient to obtain a patient basic information embedding coding vector and a disease characteristic embedding coding vector, and splice the patient basic information embedding coding vector and the disease characteristic embedding coding vector to obtain a patient information-disease cascade coding vector; an indication screening rule semantic embedding module 140, which is used to embed the basic information and disease characteristics of the patient to obtain a patient basic information embedding coding vector and a disease characteristic embedding coding vector, and splice the patient basic information embedding coding vector and the disease characteristic embedding coding vector to obtain a patient information-disease cascade coding vector; Each indication screening rule in the set of indication screening rules is semantically embedded and encoded to obtain a set of indication screening rule semantic embedded coding vectors; a patient condition-screening rule query response module 150 is used to perform a dynamic semantic query response of a selection range ratio on the patient information-condition cascade coding vector and the set of indication screening rule semantic embedded coding vectors to obtain a patient condition-screening rule query response coding vector; an analysis result generation module 160 is used to obtain an analysis result indicating whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an auxiliary treatment means during surgery based on the patient condition-screening rule query response coding vector.

[0026] In the embodiment of the present application, the patient information acquisition module 110 is used to obtain the basic information and disease characteristics of the patient. Specifically, in the embodiment of the present application, the basic information includes ECOG score, nutritional assessment, whether there is an underlying disease, whether chemotherapy, surgery or radiotherapy has been received, whether there is drug resistance or chemotherapy insensitivity; the disease characteristics include cancer type, cancer stage, tumor volume load and pathological grade, whether there are completely resectable macroscopic lesions and tumor markers.

[0027] It should be understood that the ECOG score is an indicator of the patient's functional status, reflecting the patient's ability to carry out daily activities and physical state. When considering hyperthermic intraperitoneal chemotherapy, the ECOG score is very important, as it can help doctors predict the patient's tolerance to treatment. For example, a patient with an ECOG score of 0 (able to move normally) may be more able to bear the physical burden of hyperthermic intraperitoneal chemotherapy than a patient with an ECOG score of 3 (only partially able to take care of himself and need to stay in bed or in a wheelchair for more than half of the day). A lower ECOG score may indicate that the patient is in better physical condition and is more likely to benefit from aggressive treatment methods (such as hyperthermic intraperitoneal chemotherapy), while a higher ECOG score may indicate that the patient is physically weak and unable to withstand the side effects of complex treatments, increasing the risk of treatment. Nutritional status is an important indicator of the patient's overall physical health. A good nutritional status helps patients better tolerate chemotherapy drugs and surgical procedures. For hyperthermic intraperitoneal chemotherapy, patients need sufficient nutritional reserves to cope with side effects such as nausea, vomiting, and loss of appetite that may occur during treatment, and to promote postoperative recovery. Malnourished patients may have adverse effects on the metabolism and response of chemotherapy drugs, which may also affect wound healing and the body's overall recovery ability, increase the risk of infection, and thus affect the safety and effectiveness of treatment. Underlying diseases can affect patients' tolerance and treatment effects of hyperthermic intraperitoneal chemotherapy. Previous treatment experience can affect patients' response to hyperthermic intraperitoneal chemotherapy. Patients who have undergone chemotherapy, surgery, or radiotherapy may already be in a state of stress, and their tolerance to further treatment may change. Understanding whether a patient is resistant or insensitive to chemotherapy is crucial to determining whether hyperthermic intraperitoneal chemotherapy is an appropriate treatment option. If a patient has developed resistance to conventional chemotherapy drugs, it may be necessary to reconsider the choice of drug or evaluate the efficacy of hyperthermic intraperitoneal chemotherapy, because the same drug may not achieve the expected tumor killing effect and may increase the patient's side effect burden without treatment benefit.

[0028] Accordingly, considering that different types of cancer have significant differences in biological behavior, sensitivity to chemotherapy drugs, and treatment strategies. For example, tumors such as ovarian cancer and pseudomyxoma peritonei are prone to spread and implant in the abdominal cavity. Intraperitoneal hyperthermic perfusion chemotherapy can enable chemotherapy drugs to directly act on tumor cells in the abdominal cavity and effectively control local tumor progression. Therefore, for this type of cancer, intraperitoneal hyperthermic perfusion chemotherapy is an important treatment method. For some cancer types that are naturally insensitive to chemotherapy drugs, such as pancreatic cancer, intraperitoneal hyperthermic perfusion chemotherapy alone may not achieve ideal results, and it is necessary to combine other treatment methods, such as surgical resection, radiotherapy, targeted therapy and other comprehensive treatments. Understanding the type of cancer helps to reasonably select treatment plans based on the biological characteristics of the tumor and judge the role of intraperitoneal hyperthermic perfusion chemotherapy in overall treatment. Cancer staging reflects the degree of tumor development, including tumor size, infiltration range, whether there is lymph node metastasis and distant metastasis. For early cancer, the tumor is limited. Surgical resection combined with intraperitoneal hyperthermic perfusion chemotherapy as an intraoperative adjuvant treatment can effectively remove residual micro-lesions, reduce the risk of local recurrence, and improve the cure rate. For locally advanced cancer, preoperative hyperthermic intraperitoneal chemotherapy as a neoadjuvant therapy can reduce tumor volume, improve surgical resection rate, kill micrometastases, and improve patient prognosis. For advanced cancer, if extensive distant metastasis occurs, hyperthermic intraperitoneal chemotherapy is mainly used to control intraperitoneal symptoms and relieve patient pain as a palliative treatment rather than a radical treatment option. Tumor volume load refers to the size, number and distribution of the tumor in the abdominal cavity, which affects the distribution and penetration of chemotherapy drugs in tumor tissue. Large tumor volume or high-load tumor may make it difficult for drugs to be evenly distributed to all tumor cells, affecting the treatment effect. Tumor pathological grade reflects the degree of differentiation and malignancy of tumor cells. High-grade tumor cells are highly malignant, grow rapidly, have strong invasion and metastasis capabilities, and may be less sensitive to chemotherapy drugs, but they also need more active treatment to control tumor progression. For example, for patients with large tumor volume, heavy load and high pathological grade, it may be necessary to combine systemic chemotherapy, targeted therapy and other methods on the basis of hyperthermic intraperitoneal chemotherapy to improve the treatment effect. For patients with smaller tumors and low pathological grades, the necessity of hyperthermic intraperitoneal chemotherapy as an adjuvant treatment after simple surgical resection may be relatively low. Whether there are completely resectable macroscopic lesions is directly related to the feasibility of surgical treatment and the timing of hyperthermic intraperitoneal chemotherapy. If there are completely resectable macroscopic lesions, surgical resection is the main treatment, and postoperative hyperthermic intraperitoneal chemotherapy can further remove residual cancer cells and reduce the risk of recurrence. If the macroscopic lesions cannot be completely removed, hyperthermic intraperitoneal chemotherapy may need to be used in combination with other treatment methods, such as first performing hyperthermic intraperitoneal chemotherapy to shrink the tumor and create an opportunity for surgical resection, or combining it with targeted therapy and immunotherapy to control tumor growth. In addition, whether the macroscopic lesions can be completely removed also affects the doctor's judgment of the patient's prognosis and the overall planning of the treatment plan.Tumor markers are a type of substance that reflects the presence and growth of tumors. Changes in their levels are closely related to tumor occurrence, development, and treatment response. Before intraperitoneal hyperthermic perfusion chemotherapy, detecting tumor marker levels can help understand tumor burden and biological activity, and provide a reference for the formulation of treatment plans. During treatment, dynamic monitoring of tumor marker levels can evaluate the treatment effect. If the tumor marker level drops significantly after treatment, it indicates that the treatment is effective, and intraperitoneal hyperthermic perfusion chemotherapy can continue to be part of the treatment plan; if the tumor marker level continues to increase or does not change significantly, it may mean that the treatment effect is poor and the treatment plan needs to be adjusted.

[0029] In general, the patient's basic information lays the foundation for judging whether intraperitoneal hyperthermic perfusion chemotherapy is feasible from the two aspects of overall physical condition and previous treatment background; the disease characteristics determine the specific role of intraperitoneal hyperthermic perfusion chemotherapy in treatment based on the characteristics and development level of the tumor itself; the basic information and disease characteristics complement each other and provide a comprehensive and accurate basis for judging whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or intraoperative adjuvant treatment method.

[0030] In an embodiment of the present application, the screening rule extraction module 120 is used to extract a set of indication screening rules. It should be understood that the set of indication screening rules includes pathological feature rules, previous treatment history rules, contraindication rules, etc. The indication screening rules can strictly evaluate patients and exclude patients who cannot accept tolerant treatment, which can effectively reduce the risk of serious complications (such as infection, organ failure, etc.) during treatment and help protect the patient's life safety. At the same time, by strictly following the verified indication rules, patients who can truly benefit from intraperitoneal hyperthermic perfusion chemotherapy can be better identified, thereby improving the treatment effect.

[0031] The following is a detailed description of a specific implementation process of "extracting a set of indication screening rules": First of all, in the process of constructing a set of indication screening rules, it is necessary to start from internationally recognized medical journals, oncology research papers, and guidelines issued by professional organizations, and widely collect relevant information on the indications of intraperitoneal hyperthermic perfusion chemotherapy. These resources contain the crystallization of knowledge accumulated by top scientists and clinical experts around the world over the years, and are an important basis for formulating high-quality screening rules. These institutions regularly publish recommendations on treatment strategies for different types of cancer, which provide a solid theoretical foundation for indication screening rules. In addition, a series of standard operating procedures summarized by major medical institutions based on their own practical experience can also be valuable reference materials. By in-depth mining of these materials, rich original data can be obtained, which lays a good foundation for subsequent work.

[0032] Next, in order to enable the unified management and utilization of these information from different sources, they need to be standardized. This step aims to remove redundant content and save it in a consistent format to facilitate the parsing and application of computer systems. For example, each rule can be stored in a structured data format such as XML or JSON to ensure that it is clear, understandable and easy to retrieve. At the same time, detailed definitions and semantic annotations are required for the terms and concepts in the rules to ensure that the system understands the information accurately. For example, the term "completely resectable macroscopic lesions" needs to have a clear specific meaning and judgment criteria so that the system can correctly identify and apply the rule. In addition, a logical framework needs to be built to connect the independent rules to form a complete screening system. This means that not only the effectiveness of individual rules should be considered, but also the relationship between them, such as conditional branches, weight distribution and other factors, to ensure that the entire rule set is an organic whole that can operate stably in a complex and changing clinical environment.

[0033] However, relying solely on a static rule base cannot meet the needs of modern medical services. Therefore, a mechanism for continuous updating and verification must be established so that the rule base can keep pace with the times and always remain up to date and most effective. To this end, a special team can be set up to track the latest research results and technological advances, and to adjust and improve the content of the rule base in a timely manner. In this process, regular review has become an indispensable part, which ensures that all rules reflect the latest medical knowledge and technical levels. At the same time, in order to further improve the quality of the rules, well-known experts in the field can be invited to participate in the review work. With their deep academic background and rich clinical experience, they can strictly control the professionalism and practicality of the rules. In addition, it is necessary to attach great importance to feedback from practical applications, and to test the effectiveness of existing rules through the analysis of a large amount of actual case data, and make necessary improvements accordingly. This evidence-based optimization method not only helps to improve the accuracy of the rules, but also provides more reliable support for personalized medical decision-making. In summary, a set of reliable indication screening rules can be extracted.

[0034] In the embodiment of the present application, the basic information-disease characteristic embedding fusion module 130 is used to embed the basic information and the disease characteristic of the patient to obtain the patient basic information embedding coding vector and the disease characteristic embedding coding vector, and concatenate the patient basic information embedding coding vector and the disease characteristic embedding coding vector to obtain the patient information-disease cascade coding vector. Specifically, Figure 2 FIG. 1 is a block diagram of a basic information-disease characteristic embedding fusion module in a data analysis system for intraperitoneal hyperthermic perfusion chemotherapy according to an embodiment of the present application. Figure 2As shown, the basic information-condition characteristic embedding fusion module 130 includes: a patient basic information and condition characteristic embedding coding unit 131, which is used to embed the basic information and condition characteristics of the patient using one-hot coding to obtain the patient basic information embedding coding vector and the condition characteristic embedding coding vector; a patient information-condition cascade coding feature generating unit 132, which is used to splice the patient basic information embedding coding vector and the condition characteristic embedding coding vector to obtain the patient information-condition cascade coding vector.

[0035] In an embodiment of the present application, the patient basic information and condition characteristics embedding coding unit 131 is used to embed the patient's basic information and condition characteristics using unique hot coding to obtain the patient's basic information embedding coding vector and the condition characteristics embedding coding vector. Accordingly, considering that the patient's basic information and condition characteristics contain many categorical variables, in order to be able to convert these categorical variables into binary vectors so that the model can conveniently process these data, the present application embeds the patient's basic information and condition characteristics to convert these complex and potentially redundant information into low-dimensional dense vectors, thereby obtaining the patient's basic information embedding coding vector and the condition characteristics embedding coding vector. In particular, in a specific embodiment of the present application, the patient's basic information and condition characteristics are embedded and coded using unique hot coding to obtain the patient's basic information embedding coding vector and the condition characteristics embedding coding vector. It should be understood that unique hot coding provides a standardized way to process categorical data, converting each category of the categorical variable into an independent binary vector, so that all data can be represented and processed in a relatively unified format, which is convenient for the subsequent comprehensive use of this information under the same data analysis framework.

[0036] The following is a detailed description of a specific implementation process of "using one-hot encoding to embed the basic information and condition characteristics of the patient to obtain the patient basic information embedding encoding vector and the condition characteristic embedding encoding vector": First, for the basic information of the patient, this part usually contains qualitative or categorical variables, such as ECOG score, nutritional assessment results, whether there is an underlying disease, whether other treatments have been received (such as chemotherapy, surgery or radiotherapy), and whether there is drug resistance or insensitivity to chemotherapy. Each variable has several possible ranges of values. Take the ECOG score as an example. It is used to measure the physical condition of the patient, ranging from 0 to 4 points, and the higher the score, the worse the physical function. In order to convert such categorical information into a form that can be processed by the computer, it is necessary to create a corresponding binary vector to represent each possible value. Specifically, for each categorical variable, it is necessary to construct a zero vector with matching dimensions according to the number of all possible categories, and set the elements at the corresponding position to 1 to indicate the existence of the category. For example, if the ECOG score has 5 levels, a vector of length 5 will be generated; when a patient's score is 2, the corresponding one-hot encoding vector will be [0,0,1,0,0]. The same method is applied to other basic attributes, such as nutritional assessment can be good, general, and poor, then a one-hot encoding vector of length 3 will be obtained. In this way, the different basic information of the patient can be converted into a series of independent and easy-to-calculate binary vector sets, and the patient's basic information embedded coding vector can be obtained by splicing these binary vectors in the set.

[0037] At the same time, the patient's disease characteristic data is processed. This part of information is also composed of multiple categorical variables, which reflect more of the characteristics of the disease itself, such as cancer type, stage, tumor volume load and pathological grade, whether there are completely resectable macroscopic lesions and tumor marker levels. These characteristics not only determine the severity and development stage of the disease, but also directly affect the choice and effect of treatment. Therefore, it is particularly important to encode them appropriately. Similar to the processing of basic information, a corresponding one-hot encoding is created for each disease characteristic. For example, for a multi-classification variable such as cancer type, assuming that colorectal cancer, ovarian cancer and several other common types of cancer are considered, a longer binary vector will be generated, in which only one position has a value of 1 and the rest are 0, to identify the specific cancer type. Similarly, for cancer stage (I to IV), a one-hot encoding vector of length 4 will be generated; and for tumor volume load, if it is divided into several intervals according to size, there will be a one-hot encoding vector of corresponding length accordingly. In addition, Boolean features such as the presence or absence of completely resectable macroscopic lesions can be directly represented by a vector of length 2, such as [1,0] for presence and [0,1] for absence. It is worth noting that some continuous variables (such as tumor marker levels) may need to be discretized first, divided into several intervals before applying one-hot encoding. This is done to ensure that all inputs are in the same format for unified management and analysis. Similarly, the vectors of each disease feature data need to be concatenated to obtain the final disease feature embedding encoding vector.

[0038] After completing the above one-hot encoding, two independent vectors can be obtained: the patient’s basic information embedding encoding vector and the disease characteristic embedding encoding vector.

[0039] In an embodiment of the present application, the patient information-condition cascade coding feature generation unit 132 is used to embed the patient basic information into the coding vector and the condition characteristic into the coding vector to obtain the patient information-condition cascade coding vector. It should be understood that the patient basic information and condition characteristics are two important types of data that describe the patient's condition from different angles. The patient basic information embedding coding vector reflects the patient's overall physical basic conditions, such as the ECOG score reflects the physical state, the nutritional assessment is related to the body's reserve capacity, and the underlying disease, past treatment history, etc. affect the tolerance and adaptability to treatment. The condition characteristic embedding coding vector focuses on the characteristics of the disease itself, such as cancer type, stage, tumor load, etc., which determine the severity of the disease and the direction of treatment. Therefore, in order to integrate these scattered but closely related information and fully present the complete information related to the patient and the disease, in the technical solution of the present application, the patient basic information embedding coding vector and the condition characteristic embedding coding vector are spliced, and the multi-faceted information is fused into a unified feature representation to obtain the patient information-condition cascade coding vector.

[0040] In an embodiment of the present application, the indication screening rule semantic embedding module 140 is used to perform semantic embedding encoding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding encoding vectors of indication screening rules. Specifically, in an embodiment of the present application, the indication screening rule semantic embedding module 140 is used to: use a semantic embedding encoder based on the Bert model to perform semantic embedding encoding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding encoding vectors of indication screening rules. It should be understood that each screening rule contains rich semantic information, and each screening rule contains deep semantic features and potential semantic association dependencies. Based on this, in the technical solution of the present application, semantic embedding encoding is performed on each indication screening rule in the set of indication screening rules to convert these rules into numerical representations, so as to better capture their intrinsic meanings and obtain a set of semantic embedding encoding vectors of indication screening rules. In particular, in a specific embodiment of the present application, each indication screening rule in the set of indication screening rules can be semantically embedded and encoded by using a semantic embedding encoder based on the Bert model to obtain a set of semantic embedding encoding vectors of the indication screening rules. That is, the Bert model is based on the Transformer architecture, which can encode text in both directions and fully capture contextual information. Indication screening rules usually exist in the form of natural language text, which is semantically rich and context-dependent. The Bert model can be used to mine complex semantic relationships between words, so that the semantic embedding encoding of the screening rules is more accurate and the complete semantic information of the rules is retained.

[0041] The following is a detailed description of a specific implementation process of "using a semantic embedding encoder based on a Bert model to perform semantic embedding encoding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding encoding vectors of the indication screening rules": First, data preparation and preprocessing are crucial. For example, there may be a rule like this: "For patients with colorectal cancer who have no distant metastasis and whose ECOG score is between 0 and 1, if their tumor volume is less than a certain value (such as 5 cm³) and their pathological grade is moderately or poorly differentiated, intraperitoneal hyperthermic perfusion chemotherapy can be carefully considered as an adjuvant treatment." After the collection is completed, these rule texts should be carefully cleaned to remove irrelevant punctuation marks, extra spaces, and non-standard character expressions that may interfere with the coding process, so as to ensure the standardization and consistency of the text in form and content.

[0042] Then, the cleaned text data is converted into a format that can be smoothly accepted by the model. In this process, the rule text is segmented with the help of natural language processing tools (such as NLTK or SpaCy libraries in Python). Taking the above example rules as an example, after the segmentation process, a series of tokens such as "for", "none", "distant metastasis", "and", "ECOG score", "0-1 points", "of", "colorectal cancer", "patients", "if", "tumor volume", "less than", "5cm³", "and", "pathological grade", "for", "moderately poor differentiation", "may be considered", "intraperitoneal hyperthermic perfusion chemotherapy", "as", "adjuvant treatment means" may be obtained. Then, a unique index is assigned to each token and the corresponding vocabulary is constructed. For example, "for" may correspond to index 1, "none" may correspond to index 2, and so on. At the same time, each word in the text is replaced with its corresponding index to generate an index sequence. Considering that the Bert model usually has certain restrictions on the input length (such as the maximum input length is 512 tokens), the index sequence needs to be padded or truncated. If the length of the rule text after word segmentation is less than 512, a special padding marker (such as [PAD]) is added to the end of the sequence; if it exceeds 512, the excess part is truncated to ensure that its length meets the model input requirements.

[0043] After completing data preparation and preprocessing, we will enter the stage of selecting and loading the Bert model. When selecting the Bert model version, we need to comprehensively weigh the specific needs of the actual application and the available computing resources. If the computing resources are relatively limited and the accuracy requirements for semantic understanding are not at an extremely high level, it is more appropriate to choose a smaller model such as "bert-base-uncased"; on the contrary, if there are sufficient computing resources and more powerful semantic representation capabilities are expected, you can consider choosing "bert-large-uncased" or other Bert variant models that have been pre-trained in specific fields. After selecting the model version, use the model loading function in the deep learning framework (such as TensorFlow or PyTorch) to load the selected Bert model and its weight parameters from the pre-trained model library. Taking PyTorch as an example, the model loading operation can be implemented through code statements such as "from transformersimport BertModel" and "model = BertModel.from_pretrained ('bert-base-uncased')".

[0044] Next is the core semantic embedding encoding process. In the input data preparation stage, the preprocessed regular text index sequence is first converted into a tensor form acceptable to the model. In PyTorch, the "torch.tensor()" function can be used to convert the index sequence into a tensor of shape "[batch_size, sequence_length]", where "batch_size" represents the number of regular texts processed at a time (which can be reasonably set according to computing resources and memory conditions, such as 32 or 64, etc.), and "sequence_length" is the length of the text after padding or truncation. At the same time, the corresponding attention mask tensor needs to be generated, which is used to indicate to the model which positions are actual text content and which are padded positions. For the actual text position, the mask value is set to 1 and the padded position is 0. For example, if a regular text is 100 in length after processing and padded to 512 in length, the attention mask value of the first 100 positions is 1, and the following 412 positions are 0. After preparing the input tensor and attention mask tensor, input them into the loaded Bert model for forward propagation calculation. In PyTorch, this can be achieved through the following code: "outputs=model input_ids=input_tensor, attention_mask=attention_mask)". The Bert model performs complex operations on the input regular text based on its multi-layer Transformer architecture, and finally outputs the hidden state representation of each word unit. Usually, the hidden state of the last layer is selected as the semantic representation. For example, for a regular text, the hidden state shape of the model output may be "[batch_size,sequence_length,hidden_size]", where "hidden_size" is the hidden layer dimension of the Bert model (such as 768 or 1024, etc.). In order to obtain the semantic representation vector of the entire regular text, a specific method is generally used. It is common to take the vector corresponding to the "[CLS]" (classification tag, usually added at the beginning of the text) tag in the last layer of hidden state as the semantic representation of the sentence. In PyTorch, the vector can be extracted through indexing operations, such as "semantic_embedding = outputs.last_hidden_state[:,0,:]". In this way, a semantic embedding coding vector of a fixed dimension can be obtained for each indication screening rule. By summarizing the vectors of all rules, a set of semantic embedding coding vectors of indication screening rules can be successfully constructed.

[0045] After the encoding process is completed, there are still subsequent processing and application links. On the one hand, the obtained semantic embedding coding vector should be standardized, such as using the L2 norm for normalization operation to make the vector length 1, which helps to ensure the consistency and comparability of the data in subsequent calculations and comparisons. Subsequently, the standardized vector is properly stored in the database so that it can be quickly read and efficiently used in the subsequent analysis of patient information and indication screening rules. On the other hand, in view of the continuous development of the medical field and the continuous accumulation of clinical practice experience, new indication screening rules may come into being, or existing rules may need to be adjusted and improved. Therefore, it is necessary to regularly re-collect and organize the rule data and repeat the above encoding process to update the semantic embedding coding vector set, thereby ensuring the accuracy and timeliness of the entire system in the analysis of intraperitoneal hyperthermic perfusion chemotherapy.

[0046] In the embodiment of the present application, the patient condition-screening rule query response module 150 is used to perform a dynamic semantic query response of a selection range ratio on the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule query response coding vector. Specifically, Figure 3 FIG. 1 is a block diagram of a patient condition-screening rule query response module in a data analysis system for intraperitoneal hyperthermic chemotherapy according to an embodiment of the present application. Figure 3 As shown, the patient condition-screening rule query response module 150 includes: a patient condition-screening rule anchoring unit 151, which is used to dynamically anchor the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule dynamic search anchoring vector; an indication screening rule selection range ratio determination unit 152, which is used to determine the indication screening rule selection range ratio based on the feature distribution characteristics of the patient condition-screening rule dynamic search anchoring vector, wherein the vector at the starting position of the indication screening rule selection range ratio is the patient condition-screening rule dynamic search anchoring vector, and each indication screening rule semantic embedding coding vector in the indication screening rule selection range ratio is defined as an indication screening rule dynamic search optimization library feature vector; a dynamic semantic search unit 153, which is used to perform dynamic semantic search coding on the patient information-condition cascade coding vector and each indication screening rule dynamic search optimization library feature vector in the indication screening rule selection range ratio to obtain the patient condition-screening rule query response coding vector.

[0047] It should be understood that the patient information-condition cascade coding vector represents the specific condition of the patient, and each indication screening rule semantic embedding coding vector contains the semantic information of multiple applicable condition rules. Therefore, in order to accurately analyze the degree of match between the patient's condition and each screening rule, to quantify the degree of fit between different rules and the patient's actual situation, and to avoid errors that may be caused by simple matching, this application introduces a dynamic semantic query response method of a selection range ratio to query and respond to the patient information-condition cascade coding vector and the set of the indication screening rule semantic embedding coding vector to obtain the patient condition-screening rule query response coding vector. That is, this method can automatically identify the vector that best represents the patient information-condition cascade characteristics, i.e., the dynamic search anchor point, by introducing a dynamic selection mechanism and semantic coding technology. This means that the most critical factors for judging the applicability of intraperitoneal hyperthermic chemotherapy can be accurately located from the massive and complex patient condition and rule information, thereby providing support for clinical decision-making.

[0048] Specifically, in the embodiment of the present application, the patient condition-screening rule anchoring unit 151 is used to: calculate the internal relationship score value of the patient information-condition cascade coding vector relative to each indication screening rule semantic embedding coding vector in the set of indication screening rule semantic embedding coding vectors to obtain a set of patient condition-screening rule internal relationship score values. This process can be expressed as: ; ; in, is the set of semantic embedding encoding vectors of the indication screening rules, , , and is the first, second, and third in the set of semantic embedding encoding vectors of the indication screening rule. and The semantic embedding encoding vector of the indication screening rule, is the weight matrix of indication screening rules, is the patient information-condition cascade coding vector, is the weight matrix of indication screening rules, is the hyperbolic tangent function, is the weight coefficient vector, is the first value in the set of patient condition-screening rule internal relationship score values. Patient condition-screening rule internal relationship score value; The semantic embedding coding vector of the indication screening rule corresponding to the largest patient condition-screening rule internal relationship score value in the set of the patient condition-screening rule internal relationship score values ​​is used as the patient condition-screening rule dynamic search anchor vector. The process can be expressed as: ; in, is the first in the set of semantic embedding encoding vectors of the indication screening rule The semantic embedding encoding vector of the indication screening rule, is the set of semantic embedding encoding vectors of the indication screening rules, is the first value in the set of patient condition-screening rule internal relationship score values. Patient condition-screening rule internal relationship score, is the vector at the location of the maximum value, is the patient condition-screening rule dynamic search anchor vector.

[0049] It should be understood that the patient information-condition cascade coding vector comprehensively reflects the individual characteristics and condition information of the patient, and the indication screening rule semantic embedding coding vector set represents the various standards and condition information applicable to intraperitoneal hyperthermic perfusion chemotherapy. In order to comprehensively evaluate the degree of match between the patient's actual situation and each indication screening rule, it is necessary to calculate the internal relationship score value of the patient information-condition cascade coding vector relative to each indication screening rule semantic embedding coding vector in the set of indication screening rule semantic embedding coding vectors in this application to obtain a set of patient condition-screening rule internal relationship score values. Compared with a simple direct similarity comparison that can only reflect the similarity between the two on the surface, the internal relationship score value can consider more potential factors, and can deeply analyze the fit between the patient and each rule, avoiding making a one-sided judgment based only on surface similarity.

[0050] Accordingly, considering the complexity and diversity of the patient's situation, there are also interrelationships and influences between the various indication screening rules. If only a single rule with the maximum score is used, the important information reflected by other related rules may be ignored, and the relationship between the patient's overall condition and all indication screening rules cannot be fully considered. For example, when evaluating the applicability of intraperitoneal hyperthermic perfusion chemotherapy, a rule may score the highest due to a prominent feature of the patient, but from the overall cancer type, stage, basic physical condition and other aspects, the rule may not be the most representative. Considering the global distribution characteristics, it can ensure that the selected rule is the most appropriate in a wider context, avoiding inappropriate decisions due to the one-sided pursuit of the highest score. Based on this, in this application, the semantic embedding coding vector of the indication screening rule corresponding to the largest patient condition-screening rule internal relationship score value in the set of the patient condition-screening rule internal relationship score value is required as the patient condition-screening rule dynamic search anchor vector. By selecting representative rules as anchor points, the relationship between the overall condition of the patient and the treatment indication can be better reflected.

[0051] Next, based on the characteristic distribution characteristics of the patient condition-screening rule dynamic search anchor vector, the indication screening rule selection range ratio is determined, wherein the vector at the starting position of the indication screening rule selection range ratio is the patient condition-screening rule dynamic search anchor vector, and each indication screening rule semantic embedding coding vector in the indication screening rule selection range ratio is defined as the indication screening rule dynamic search optimization library feature vector. The above process can be expressed as: ; ; in, is the patient condition-screening rule dynamic search anchor vector, and They are The variance and mean of is the modulation parameter, To obtain The maximum value in Indicates rounding up. is the size of the selection range ratio of the indication screening rule, is a set of characteristic vectors of the dynamic search optimization library of the indication screening rule in the indication screening rule selection range ratio, and Select the first and Dynamic search and optimization of library feature vectors for indication screening rules.

[0052] It should be understood that the set of semantic embedding encoding vectors of indication screening rules may contain a lot of information. If all vectors are fully analyzed, the calculation amount is huge and inefficient. By determining the selection range ratio of indication screening rules and taking the patient condition-screening rule dynamic search anchor vector as the starting position, we can focus on the feature vectors with high relevance to the current patient query, which can greatly reduce the search space for subsequent analysis and help improve the analysis efficiency. It should be noted that medical data is dynamically changing, the patient's condition may change during the treatment process, and new clinical research results may also update the indication screening rules. Therefore, as new data continues to flow in, it is necessary to keep the selection range and the optimized library feature vectors within it dynamically updated. This ensures that the system is always analyzed based on the latest and most relevant information, maintains the timeliness and relevance of the system, and provides a more reliable basis for clinical decision-making.

[0053] Specifically, in an embodiment of the present application, the dynamic semantic search unit 153 is used to: respectively calculate the cosine similarity between the patient information-condition cascade coding vector and the feature vectors of the dynamic search optimization library of each indication screening rule in the indication screening rule selection range ratio to obtain a set of patient condition-screening rule semantic similarities; using the set of patient condition-screening rule semantic similarities as a weighted sequence, calculate the weighted sum of the semantic differences between the patient information-condition cascade coding vector and the feature vectors of the dynamic search optimization library of each indication screening rule in the indication screening rule selection range ratio to obtain a patient condition-screening rule semantic measurement vector; and perform position point comparison on the patient condition-screening rule semantic measurement vector and the patient condition-screening rule dynamic search anchor vector to obtain the patient condition-screening rule query response coding vector. The above process can be expressed as: ; ; in, Select the first Indication screening rules dynamically search and optimize library feature vectors, is the Euclidean norm of the vector, is the patient information-condition cascade coding vector, for and The patient condition-screening rule semantic similarity between is the patient condition-screening rule dynamic search anchor vector, is the patient condition-screening rule query response encoding vector.

[0054] It should be understood that the patient information-condition cascade encoding vector represents the comprehensive information of a specific patient, while the characteristic vector set of the indication screening rule dynamic search optimization library covers a series of treatment indication rule information related to the patient's condition. Dynamic semantic search encoding of the two can effectively integrate this information and mine the deep semantic connection between patient information and treatment rules. And considering that the semantics between patient information and indication rules are relatively similar, but the slight difference information may have a significant impact on treatment decisions. For example, two patients have similar cancer types and stages, but one patient has a specific underlying disease. This subtle difference may determine whether intraperitoneal hyperthermic perfusion chemotherapy is applicable. In the process of dynamic semantic search encoding, cosine similarity can be used to enhance the difference information between the two, thereby ensuring that these subtle differences will not be ignored, while capturing the overall semantic information, thereby providing comprehensive information support for accurate treatment decisions. The final output patient condition-screening rule query response encoding vector has a high degree of abstract generalization ability, which can express the complex semantic relationship between patient information and indication rules in a concise and effective way. This abstract summary enables the rapid and accurate use of this information for decision-making in downstream tasks, such as determining whether hyperthermic intraperitoneal chemotherapy is suitable as a first-line treatment strategy or as an intraoperative adjuvant therapy.

[0055] In an embodiment of the present application, the analysis result generation module 160 is used to obtain an analysis result indicating whether hyperthermic intraperitoneal chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment during surgery based on the patient condition-screening rule query response coding vector. Specifically, in an embodiment of the present application, the analysis result generation module 160 is used to: input the patient condition-screening rule query response coding vector into a data analysis auxiliary decision module based on a classifier to obtain the analysis result, and the analysis result is used to indicate whether hyperthermic intraperitoneal chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment during surgery. That is, the patient condition-screening rule query response coding vector obtained by querying the response using the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector is classified, so as to utilize the professional classification ability of the classifier to effectively classify the complex information contained in the vector, thereby obtaining an accurate analysis result, thereby providing a key basis for formulating a treatment plan. In particular, in a specific embodiment, the patient condition-screening rule query response encoding vector is input into a classifier-based data analysis decision support module to obtain the analysis result, and the analysis result is used to indicate whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an intraoperative auxiliary treatment method, including: using the fully connected layer of the classifier to fully connect the patient condition-screening rule query response encoding vector to obtain a patient condition-screening rule query response fully connected encoding classification feature vector; inputting the patient condition-screening rule query response fully connected encoding classification feature vector into the Softmax classification function of the classifier to obtain the analysis result.

[0056] Here, since the patient information-condition cascade coding vector and the set of indication screening rule semantic embedded coding vectors respectively represent the embedded splicing coding features of the patient's basic information and condition characteristics and the embedded semantic coding features of the set of indication screening rules, when performing feature dynamic semantic search encoding based on selection range ratio anchoring, the feature population attributes of different semantics will have dynamic semantic search fairness differences based on the selection range ratio anchoring level, thereby affecting the coding feature distribution response inclusiveness of the patient condition-screening rule query response coding vector, and reducing the accuracy of the analysis results obtained by inputting the classifier-based data analysis decision support module.

[0057] In a preferred example, inputting the patient condition-screening rule query response encoding vector into a classifier-based data analysis auxiliary decision module to obtain an analysis result includes: Determine the mean of the eigenvalues ​​corresponding to the patient condition-screening rule query response encoding vector and the standard deviation of the eigenvalues ; The patient condition-screening rule query response encoding vector is multiplied by the point-subtraction vector of the eigenvalue mean and the eigenvalue standard deviation to obtain a first patient condition-screening rule query response encoding fairness target vector. This process can be expressed as: ; in, represents the patient condition-screening rule query response encoding vector, express The corresponding eigenvalue mean, express The corresponding eigenvalue standard deviation is, It means point multiplication by position. It means to subtract by position point. representing said first patient condition-screening rule query response encoding a fairness objective vector; The point-wise multiplication of the patient condition-screening rule query response encoding vector and the point-wise subtraction vector of the eigenvalue standard deviation and the eigenvalue mean is performed to obtain a second patient condition-screening rule query response encoding fairness target vector. This process can be expressed as: ; in, represents the patient condition-screening rule query response encoding vector, express The corresponding eigenvalue mean, express The corresponding eigenvalue standard deviation is, It means point multiplication by position. It means to subtract by position point. The second patient condition-screening rule query response encodes a fairness objective vector.

[0058] After multiplying the bit-by-bit inverse of the second patient condition-screening rule query response encoding fairness target vector by the first patient condition-screening rule query response encoding fairness target vector, taking the bit-by-bit logarithm with base 2 to obtain the patient condition-screening rule query response encoding information correction vector, the process can be expressed as: ; in, represents the first patient condition-screening rule query response encoding a fairness target vector, represents the second patient condition-screening rule query response encoding fairness objective vector, Representation calculation The inverse of each eigenvalue in , is the logarithmic function value with base 2, It means point multiplication by position. A correction vector representing the patient condition-screening rule query response encoding information.

[0059] The eigenvalue mean Divide by the standard deviation of the eigenvalue The square root of the quotient is multiplied by the weight hyperparameter, and then the optimized patient condition-screening rule query response encoding vector is obtained by adding the patient condition-screening rule query response encoding information correction vector point. The process can be expressed as: ; in, express The corresponding eigenvalue mean, express The corresponding eigenvalue standard deviation is, represents the weighted hyperparameter, Indicates adding by position point. represents the patient condition-screening rule query response encoding information correction vector, a patient condition-screening rule query response encoding vector representing the optimized patient condition-screening rule query response encoding vector; The optimized patient condition-screening rule query response encoding vector is input into the classifier-based data analysis auxiliary decision module to obtain the analysis result.

[0060] That is, taking into account the attribute level fairness differences of the data population corresponding to the sequence response fusion features of the patient condition-screening rule query response encoding vector, in order to improve the response inclusiveness under the feature distribution diversity of the patient condition-screening rule query response encoding vector, the crossover probability value constraint based on the patient condition-screening rule query response encoding vector is used as the interactive fairness target representation to correct the group feature information interactive propagation of the patient condition-screening rule query response encoding vector, and the unified statistical feature response interaction based on the patient condition-screening rule query response encoding vector is used as the feature distribution multi-level fairness target bias to achieve a robust distribution fairness unified representation of the patient condition-screening rule query response encoding vector, form a fair collaboration paradigm under the feature distribution framework of the patient condition-screening rule query response encoding vector, and improve the accuracy of the analysis results obtained by inputting the classifier-based data analysis auxiliary decision-making module.

[0061] In summary, the data analysis system 100 for intraperitoneal hyperthermic perfusion chemotherapy based on the embodiment of the present application is explained, which uses deep learning-based data processing technology to embed and encode the patient's basic information and disease characteristics, then splices the embedded patient basic information embedding features and disease characteristics embedding features, and at the same time, semantic embedding encoding is performed on each indication screening rule, so as to intelligently analyze whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment during surgery based on the selection range ratio between the patient information-disease condition cascade coding feature and the semantic embedding feature of each indication screening rule according to the query response representation. In this way, it is conducive to providing more accurate and timely treatment strategy recommendations, thereby helping doctors make more scientific and reasonable decisions.

[0062] Figure 4 FIG. 4 is a flow chart of a data analysis method for hyperthermic intraperitoneal chemotherapy according to an embodiment of the present application. Figure 4 As shown, in the data analysis method for intraperitoneal hyperthermic perfusion chemotherapy, it includes: S110, obtaining the basic information and disease characteristics of the patient; S120, extracting a set of indication screening rules; S130, embedding and coding the basic information and disease characteristics of the patient to obtain a patient basic information embedding coding vector and a disease characteristic embedding coding vector, and splicing the patient basic information embedding coding vector and the disease characteristic embedding coding vector to obtain a patient information-disease condition cascade coding vector; S140, semantically embedding and coding each indication screening rule in the set of indication screening rules to obtain a set of indication screening rule semantic embedding coding vectors; S150, performing a dynamic semantic query response of a selection range ratio on the set of the patient information-disease condition cascade coding vector and the set of the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule query response coding vector; S160, based on the patient condition-screening rule query response coding vector, obtaining an analysis result for indicating whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment means during surgery.

[0063] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned data analysis method for intraperitoneal hyperthermic chemotherapy have been described in the above reference. Figures 1 to 3 The data analysis system for intraperitoneal hyperthermic chemotherapy has been described in detail, and therefore, its repeated description will be omitted.

[0064] In summary, the data analysis method for intraperitoneal hyperthermic perfusion chemotherapy based on the embodiment of the present application is explained, which uses deep learning-based data processing technology to embed and encode the patient's basic information and disease characteristics, then splices the embedded patient basic information embedding features and disease characteristics embedding features, and at the same time, semantic embedding encoding is performed on each indication screening rule, so as to intelligently analyze whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment during surgery based on the selection range ratio between the patient information-disease condition cascade coding feature and the semantic embedding feature of each indication screening rule according to the query response representation. In this way, it is conducive to providing more accurate and timely treatment strategy recommendations, thereby helping doctors make more scientific and reasonable decisions.

Claims

1. A data analysis system for intraperitoneal hyperthermic chemotherapy, characterized in that: include: Patient information acquisition module, used to obtain the patient's basic information and disease characteristics; A screening rule extraction module, used to extract a set of indication screening rules; A basic information-condition characteristic embedding fusion module is used to embed the basic information and condition characteristics of the patient to obtain a patient basic information embedding coding vector and a condition characteristic embedding coding vector, and concatenate the patient basic information embedding coding vector and the condition characteristic embedding coding vector to obtain a patient information-condition cascade coding vector; An indication screening rule semantic embedding module, used for performing semantic embedding coding on each indication screening rule in the set of indication screening rules to obtain a set of indication screening rule semantic embedding coding vectors; A patient condition-screening rule query response module, used for performing a dynamic semantic query response of a selection range ratio on the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule query response coding vector; The analysis result generation module is used to query the response coding vector based on the patient condition-screening rule to obtain an analysis result indicating whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an auxiliary treatment means during surgery.

2. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 1, characterized in that: The basic information includes ECOG score, nutritional assessment, whether there is underlying disease, whether chemotherapy, surgery or radiotherapy has been received, and whether there is drug resistance or chemotherapy insensitivity; the disease characteristics include cancer type, cancer stage, tumor volume burden and pathological grade, whether there are completely resectable macroscopic lesions and tumor markers.

3. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 2, characterized in that: The basic information-disease characteristics are embedded in the fusion module, including: A patient basic information and condition characteristics embedding coding unit, used for embedding the patient's basic information and condition characteristics using one-hot coding to obtain the patient basic information embedding coding vector and the condition characteristics embedding coding vector; The patient information-condition cascade coding feature generating unit is used to concatenate the patient basic information embedded coding vector and the condition feature embedded coding vector to obtain the patient information-condition cascade coding vector.

4. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 3, characterized in that: The indication screening rule semantic embedding module is used to: use a semantic embedding encoder based on the Bert model to perform semantic embedding encoding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding encoding vectors of the indication screening rules.

5. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 4, characterized in that: The patient condition-screening rule query response module includes: A patient condition-screening rule anchoring unit, used for dynamically anchoring the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule dynamic search anchoring vector; an indication screening rule selection range ratio determination unit, configured to determine the indication screening rule selection range ratio based on the characteristic distribution characteristics of the patient condition-screening rule dynamic search anchor vector, wherein the vector at the starting position of the indication screening rule selection range ratio is the patient condition-screening rule dynamic search anchor vector, and each indication screening rule semantic embedding coding vector in the indication screening rule selection range ratio is defined as an indication screening rule dynamic search optimization library feature vector; The dynamic semantic search unit is used to perform dynamic semantic search encoding on the patient information-condition cascade encoding vector and each indication screening rule dynamic search optimization library feature vector in the indication screening rule selection range ratio to obtain the patient condition-screening rule query response encoding vector.

6. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 5, characterized in that: The patient condition-screening rule anchoring unit is used to: Calculating the internal relationship score value of the patient information-condition cascade coding vector relative to each indication screening rule semantic embedding coding vector in the set of indication screening rule semantic embedding coding vectors to obtain a set of patient condition-screening rule internal relationship score values; The indication screening rule semantic embedding coding vector corresponding to the largest patient condition-screening rule internal relationship score value in the set of patient condition-screening rule internal relationship score values ​​is used as the patient condition-screening rule dynamic search anchor vector.

7. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 6, characterized in that: The dynamic semantic search unit is used to: Respectively calculating the cosine similarity between the patient information-condition cascade coding vector and the feature vector of each indication screening rule dynamic search optimization library in the indication screening rule selection range ratio to obtain a set of patient condition-screening rule semantic similarity; Taking the set of patient condition-screening rule semantic similarities as a weighted sequence, calculating the weighted sum of semantic differences between the patient information-condition cascade coding vector and the feature vectors of each indication screening rule dynamic search optimization library in the indication screening rule selection range ratio to obtain a patient condition-screening rule semantic measurement vector; The patient condition-screening rule semantic metric vector and the patient condition-screening rule dynamic search anchor vector are added point by point to obtain the patient condition-screening rule query response encoding vector.

8. The data analysis system for hyperthermic intraperitoneal chemotherapy according to claim 7, characterized in that: The analysis result generation module is used to: input the patient condition-screening rule query response encoding vector into a classifier-based data analysis auxiliary decision module to obtain the analysis result, and the analysis result is used to indicate whether intraperitoneal hyperthermic perfusion chemotherapy is suitable as a first-line treatment strategy or an auxiliary treatment method during surgery.

9. A data analysis method for hyperthermic intraperitoneal chemotherapy, characterized in that: include: Obtain the patient's basic information and disease characteristics; Extracting a set of indication screening rules; Embedding and coding the basic information and condition characteristics of the patient to obtain a patient basic information embedded coding vector and a condition characteristic embedded coding vector, and concatenating the patient basic information embedded coding vector and the condition characteristic embedded coding vector to obtain a patient information-condition cascade coding vector; Performing semantic embedding coding on each indication screening rule in the set of indication screening rules to obtain a set of indication screening rule semantic embedding coding vectors; Performing a dynamic semantic query response of a selection range ratio on the set of the patient information-condition cascade coding vector and the indication screening rule semantic embedding coding vector to obtain a patient condition-screening rule query response coding vector; Based on the patient condition-screening rule query response encoding vector, an analysis result is obtained to indicate whether intraperitoneal hyperthermic chemotherapy is suitable as a first-line treatment strategy or an adjuvant treatment means during surgery.

10. The data analysis method for hyperthermic intraperitoneal chemotherapy according to claim 9, characterized in that: The method comprises: performing semantic embedding coding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding coding vectors of the indication screening rules, including: using a semantic embedding encoder based on a Bert model to perform semantic embedding coding on each indication screening rule in the set of indication screening rules to obtain a set of semantic embedding coding vectors of the indication screening rules.

Images