Perioperative nutritional management system and method for patients undergoing hepatobiliary and pancreatic surgery

Abstract

The present application relates to the field of intelligent management, and specifically discloses a perioperative nutrition management system and method for hepatobiliary and pancreatic surgery patients, which uses data analysis and coding technology based on deep learning to organize vital sign data and multi-dimensional time series association, and semantically embed and associate the current test results, and then perform fine-grained cross-modal encoding of the multi-dimensional time series association features of vital signs and the semantic association features of test items, so as to intelligently generate a nutrition management plan based on the semantic cross-modal coding representation and the surgery type label. In this way, the physiological condition and test characteristics of each patient can be fully captured, and a unique nutrition treatment plan can be tailored for the patient. In addition, it also supports real-time monitoring and flexible adjustment of the patient's condition to ensure that appropriate intervention measures are taken at critical moments.

Description

Technical Field

[0001] The present application relates to the field of intelligent management, and more specifically, to a perioperative nutrition management system and method for hepatobiliary and pancreatic surgery patients. Background Art

[0002] Hepatobiliary and pancreatic surgery involves the liver, biliary system and pancreas, which play a key role in human metabolism and digestion. Therefore, nutritional management of patients undergoing such surgery during the perioperative period (before, during and after surgery) is crucial. Good nutritional support can not only promote postoperative recovery and reduce the risk of complications, but also enhance immune function and accelerate tissue repair.

[0003] Traditional perioperative nutritional management of hepatobiliary and pancreatic surgery patients usually relies on the physician's experience and conventional clinical guidelines. However, this approach often develops uniform standards based on group statistical data, and fails to fully consider the specific circumstances of each patient (such as underlying diseases, metabolic differences), resulting in the special needs of some patients not being met. In addition, due to the reliance on regular examination results to adjust treatment plans, it is difficult to respond to changes in the patient's physical condition in real time, especially in cases where the condition is complex or changes rapidly, which may lead to missing the best time for intervention. For example, patients with abnormal liver function may require stricter protein management, while diabetic patients need to pay special attention to blood sugar fluctuations. Traditional methods lack accurate grasp of these individualized needs, making nutritional support less sophisticated.

[0004] Therefore, an optimized perioperative nutritional management program for patients undergoing hepatobiliary and pancreatic surgery is desired. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of the present application, a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients is provided, comprising:

[0007] A vital sign data collection module is used to collect a time queue of vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data includes body temperature, heart rate, blood pressure, blood oxygen saturation and weight;

[0008] A test result acquisition module, used to acquire the current test result of the hepatobiliary and pancreatic surgery patient object;

[0009] A vital sign data analysis module, used for data sorting and multi-dimensional time series correlation of the time queue of the vital sign data to obtain multi-dimensional time series correlation characteristics of the vital sign;

[0010] A test result analysis module, used for performing embedded coding and test item semantic association coding on the current test result of the hepatobiliary and pancreatic surgery patient object to obtain test item semantic association features;

[0011] A cross-modal fine-grained encoding module, used for performing fine-grained cross-modal attention encoding on the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain a vital sign-test semantic cross-modal ablation encoding feature;

[0012] A surgery type label acquisition module, used to acquire the surgery type label of the hepatobiliary and pancreatic surgery patient object;

[0013] A nutrition management plan generation module is used to generate a nutrition management plan based on the vital sign-test semantic cross-modal ablation coding features and the surgery type label.

[0014] Furthermore, the vital signs data analysis module includes:

[0015] A vital sign data collating unit, configured to use a vital sign data collator to collate the time queue of the vital sign data according to the time dimension and the vital sign parameter sample dimension to obtain a multi-dimensional data time series matrix of the vital signs of the patient;

[0016] The vital sign multidimensional time series association unit is used to obtain a vital sign multidimensional time series association feature graph as the vital sign multidimensional time series association feature by passing the patient object's vital sign multidimensional data time series matrix through a vital sign multidimensional time series association feature extractor based on MCNN.

[0017] Furthermore, the test result analysis module includes:

[0018] A test result embedding coding unit, used for embedding coding the current test result of the hepatobiliary and pancreatic surgery patient object through a test result embedding matrix to obtain a sequence of test item embedding semantic coding vectors;

[0019] The test item semantic association unit is used to embed the sequence of the test item semantic coding vectors through a test item semantic association encoder based on an RNN-LSTM hybrid model to obtain a test item semantic association feature vector as the test item semantic association feature.

[0020] Furthermore, the cross-modal fine-grained encoding module includes:

[0021] A vital sign fine-grained decoupling unit, used for performing feature fine-grained decoupling on the multi-dimensional time series correlation feature graph of the vital sign to obtain a set of multi-dimensional time series correlation local feature matrices of the vital sign;

[0022] A vital sign fine-grained ablation unit, configured to perform fine-grained ablation optimization on each vital sign multi-dimensional temporal correlation local feature matrix in the set of the vital sign multi-dimensional temporal correlation local feature matrix based on the test item semantic correlation feature vector to obtain a set of cross-domain guided ablation vital sign multi-dimensional temporal correlation local feature matrices;

[0023] A vital sign feature aggregation unit is used to perform feature aggregation on the set of the cross-domain guided ablation vital sign multi-dimensional temporal correlation local feature matrices to obtain a vital sign-test semantic cross-modal ablation coding feature map as the vital sign-test semantic cross-modal ablation coding feature.

[0024] Furthermore, the vital sign fine-grained decoupling unit is used to perform fine-grained feature decoupling on the vital sign multidimensional time series association feature graph along the channel dimension of the vital sign multidimensional time series association feature graph to obtain a set of the vital sign multidimensional time series association local feature matrices.

[0025] Furthermore, the vital sign fine-grained ablation unit comprises:

[0026] A vital sign ablation weight factor calculation subunit, used to calculate the fine-grained ablation weight factor of each vital sign multidimensional time series association local feature matrix in the set of vital sign multidimensional time series association local feature matrices based on the test item semantic association feature vector to obtain a set of vital sign multidimensional time series association fine-grained ablation weight factors;

[0027] The vital sign fine-grained ablation modulation subunit is used to perform fine-grained ablation modulation on the set of vital sign multi-dimensional time series associated local feature matrices based on the set of vital sign multi-dimensional time series associated fine-grained ablation weight factors to obtain the set of cross-domain guided ablation vital sign multi-dimensional time series associated local feature matrices.

[0028] Furthermore, the vital sign ablation weight factor calculation subunit is used to:

[0029] Using each of the vital sign multidimensional time series association local feature matrices in the set of the vital sign multidimensional time series association local feature matrices as a key matrix, performing ablation factor encoding based on a simulated converter on the test item semantic association feature vector and the vital sign multidimensional time series association local feature matrix to obtain a set of vital sign multidimensional time series association fine-grained ablation factors;

[0030] The set of multi-dimensional time series associated fine-grained ablation factors of the vital signs is subjected to ablation effect encoding including normalization and masking to obtain the set of multi-dimensional time series associated fine-grained ablation weight factors of the vital signs.

[0031] Furthermore, the nutrition management plan generation module is used to: after expanding the vital signs-test semantic cross-modal ablation coding feature map into a vital signs-test semantic cross-modal ablation coding feature vector, add the surgery type label and generation prompt information to the end of the vital signs-test semantic cross-modal ablation coding feature vector, and input it into the AIGC-based perioperative nutrition manager to generate the nutrition management plan, and the generation prompt information is "based on the input corpus, generate a nutrition management plan for the hepatobiliary and pancreatic surgery patient object."

[0032] According to another aspect of the present application, a perioperative nutritional management method for hepatobiliary and pancreatic surgery patients is provided, comprising:

[0033] A time queue for collecting vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data includes body temperature, heart rate, blood pressure, blood oxygen saturation and weight;

[0034] Obtaining current test results of the hepatobiliary and pancreatic surgery patient object;

[0035] Performing data sorting and multi-dimensional time series correlation on the time queue of the vital sign data to obtain multi-dimensional time series correlation characteristics of the vital sign;

[0036] Performing embedded coding and test item semantic association coding on the current test result of the hepatobiliary and pancreatic surgery patient object to obtain test item semantic association features;

[0037] Performing fine-grained cross-modal attention encoding on the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain a vital sign-test semantic cross-modal ablation encoding feature;

[0038] Acquire a surgery type label of the hepatobiliary and pancreatic surgery patient object;

[0039] A nutritional management plan is generated based on the vital sign-test semantic cross-modal ablation coding features and the surgery type label.

[0040] Compared with the prior art, the present application provides a perioperative nutrition management system and method for hepatobiliary and pancreatic surgery patients, which collects the time queue of vital sign data (body temperature, heart rate, blood pressure, blood oxygen saturation and weight) of hepatobiliary and pancreatic surgery patients during the perioperative period, obtains the current test results and surgery type labels of the hepatobiliary and pancreatic surgery patients, uses data analysis and coding technology based on deep learning to sort the vital sign data and multi-dimensional time series association, and semantically embeds and associates the current test results, and then performs fine-grained cross-modal encoding of the multi-dimensional time series association features of the vital signs and the semantic association features of the test items, so as to intelligently generate a nutrition management plan based on the semantic cross-modal coding representation and the surgery type label. This technical solution can fully capture the physiological condition and test characteristics of each patient, and then tailor a unique nutrition treatment plan for the patient. In addition, it also supports real-time monitoring and flexible adjustment of the patient's condition, and can quickly optimize the treatment plan even in the face of complex and changing conditions, ensuring that appropriate intervention measures are taken at critical moments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0042] Figure 1 This is a block diagram of a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application.

[0043] Figure 2 This is a block diagram of a vital sign data analysis module in a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application.

[0044] Figure 3 It is a block diagram of the test result analysis module in the perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application.

[0045] Figure 4 This is a block diagram of a cross-modal fine-grained coding module in a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application.

[0046] Figure 5 This is a flow chart of a method for perioperative nutritional management of hepatobiliary and pancreatic surgery patients according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0048] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0049] It is worth noting that in this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0050] Therefore, in response to the problems in the above-mentioned background technology, the present application proposes a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients. Figure 1 FIG. 1 is a block diagram of a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application. Specifically, Figure 1 As shown, according to the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients of the embodiment of the present application, it includes: a vital sign data acquisition module 110, which is used to collect a time queue of vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data includes body temperature, heart rate, blood pressure, blood oxygen saturation and weight; a test result acquisition module 120, which is used to obtain the current test result of the hepatobiliary and pancreatic surgery patient; a vital sign data analysis module 130, which is used to sort the time queue of the vital sign data and perform multi-dimensional time series association to obtain multi-dimensional time series association characteristics of vital signs; a test result analysis module 140, A module 150 is used to perform embedding coding and semantic association coding of the current test results of the hepatobiliary and pancreatic surgery patient to obtain semantic association features of the test items; a cross-modal fine-grained coding module 150 is used to perform fine-grained cross-modal attention coding of the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain vital signs-test semantic cross-modal ablation coding features; a surgery type label acquisition module 160 is used to obtain the surgery type label of the hepatobiliary and pancreatic surgery patient; a nutrition management plan generation module 170 is used to generate a nutrition management plan based on the vital signs-test semantic cross-modal ablation coding features and the surgery type label.

[0051] In the embodiment of the present application, the vital sign data acquisition module 110 is used to collect the time queue of vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data includes body temperature, heart rate, blood pressure, blood oxygen saturation and weight. It should be understood that the normal body temperature of the human body is usually between 36.5 and 37.5 degrees Celsius. Changes in body temperature can reflect whether the body is in an infected state. Especially after surgery, fever may be an early sign of infection. For hepatobiliary and pancreatic surgery patients, monitoring body temperature helps to timely detect and deal with possible complications. The heart rate value is the number of heart beats per minute. The normal resting heart rate of an adult is about 60 to 100 beats per minute. Changes in heart rate can reflect the workload of the heart and the response to factors such as anesthetics, pain or anxiety. Perioperative heart rate monitoring is crucial for preventing cardiovascular events. Blood pressure values ​​are divided into systolic pressure (the highest pressure in the artery when the heart pumps blood) and diastolic pressure (the lowest pressure in the artery during the interval between two heartbeats). The normal blood pressure range for adults is 90-120 mmHg / 60-80 mmHg. Blood pressure levels are directly related to organ perfusion. Too low may lead to tissue ischemia, while too high may increase the risk of bleeding. Maintaining stable blood pressure is especially important for major surgeries such as hepatobiliary and pancreatic surgeries. Blood oxygen saturation indicates the ratio of oxygen in the blood to red blood cells, which should be maintained above 95% under normal circumstances. It is an important indicator for measuring respiratory function, especially under anesthesia, ensuring adequate oxygen supply is the key to ensuring patient safety. Although weight values ​​do not change as frequently in real time as the other four parameters, changes in weight can directly affect aspects such as drug dosage calculation and fluid balance management. For patients with hepatobiliary and pancreatic diseases, weight loss may be a manifestation of malnutrition, while rapid weight gain may indicate edema or other pathological conditions. Therefore, in the technical solution of the present application, the time queue of vital sign data of hepatobiliary and pancreatic surgical patients during the perioperative period is collected, which can comprehensively assess the health status of the patients and provide an important reference for perioperative nutritional management.

[0052] First, in order to ensure the effectiveness of data collection, hospitals usually establish a personalized perioperative monitoring plan for each patient who is about to undergo hepatobiliary and pancreatic surgery. This plan is based on the latest clinical guidelines and research findings, and is formulated in combination with the patient's specific condition. For example, for those patients with underlying diseases such as hypertension or diabetes, more frequent vital sign monitoring may be arranged to capture any abnormal changes in a timely manner. In addition, considering the changing characteristics of the patient's physical condition at different stages, the entire perioperative period is subdivided into three main parts: preoperative preparation, surgery, and postoperative recovery. Each part has its specific data collection frequency and focus.

[0053] During the preoperative preparation period, medical staff began to closely monitor the patient's vital signs. The main purpose of this stage is to understand the patient's basic health status, assess his tolerance to surgery, and adjust relevant treatment plans. Generally speaking, basic parameters such as body temperature, heart rate, and blood pressure are measured at least twice a day, while blood oxygen saturation may be tested more frequently depending on the patient's condition. At the same time, weight is also recorded regularly to observe whether there are significant fluctuations due to disease progression. All of this information is entered into the electronic medical record system to form a preliminary time series data set, which provides a basis for subsequent analysis.

[0054] During surgery, the patient's vital signs will change significantly due to the effects of anesthetics and other interventions, so data collection at this time is particularly important. The operating room is equipped with advanced monitoring equipment that can monitor and record the above-mentioned vital signs in real time. In particular, blood oxygen saturation is directly related to the patient's oxygen supply, which is even more important under general anesthesia. By continuously updating these key indicators, the medical team can quickly respond to possible problems, take necessary emergency measures, and ensure the smooth progress of the operation.

[0055] The postoperative recovery period is the longest stage of the entire perioperative management and is also a critical period that determines the success of the operation. During this period, as the patient gradually wakes up from anesthesia, their vital signs will undergo a series of complex adjustments. In order to accurately grasp these dynamic changes, medical staff will continue to perform strict monitoring tasks according to the established schedule. Generally speaking, in the early postoperative period (that is, within 24 hours after surgery), body temperature, heart rate, blood pressure and blood oxygen saturation will be checked every hour; then, as the patient's condition stabilizes, the measurement interval will be appropriately extended to every two to four hours. Weight measurement is relatively rare, but it is still maintained at least once a week, especially when fluid balance management and nutritional support are involved.

[0056] In the process of collecting vital signs data throughout the perioperative period, the application of information technology has greatly improved work efficiency and data quality. Modern hospitals generally use integrated information systems to achieve seamless data sharing between departments from wards to operating rooms to intensive care units. In this way, no matter which doctor views the patient's latest vital signs record, he can get the most complete information. Moreover, with the support of intelligent algorithms, the system can also automatically identify certain abnormal trends and send early warning signals to relevant personnel, making early intervention possible. For example, when an indicator exceeds the normal range, the system will immediately notify the nurse on duty or the attending physician, reminding them to further investigate the cause and consider appropriate treatment methods.

[0057] In an embodiment of the present application, the test result acquisition module 120 is used to obtain the current test results of the hepatobiliary and pancreatic surgery patient object. Accordingly, considering that the current test results refer to the data obtained from the patient's most recent one or a series of laboratory tests and imaging examinations, which include blood biochemical indicators, such as blood sugar levels, liver function tests, protein levels, etc., imaging tests, complete blood cell counts, inflammatory markers, etc. Therefore, in order to more accurately assess the patient's overall health status and thus develop a nutritional management plan that best suits individual needs, in the technical solution of the present application, the current test results of the hepatobiliary and pancreatic surgery patient object are obtained.

[0058] In practical terms, the clinician first issues a corresponding test application form based on the patient's specific condition. This application form lists in detail all the tests that need to be performed, and specifies the priority and urgency. Subsequently, the nurse will collect samples from the patient according to the doctor's instructions, usually drawing a certain amount of blood from the vein, and sometimes involving other body fluids or tissue samples. After the sampling is completed, the specimen will be immediately sent to the central laboratory or a special laboratory department within the hospital. In order to ensure the quality of the sample, the relevant standard operating procedures must be strictly followed during the entire transportation process, such as maintaining appropriate temperature conditions and preventing cross contamination.

[0059] Once the samples arrive at the laboratory, professional technicians begin to process and analyze them. Modern laboratories are equipped with advanced automated instruments and equipment, which can quickly and accurately complete most routine inspection tasks. For example, fully automatic biochemical analyzers can measure multiple biochemical parameters simultaneously in a short period of time; while flow cytometers can be used for complex immunological studies. For some special inspection items, more professional technologies and methods may be required, such as PCR (polymerase chain reaction) for genetic testing, ELISA (enzyme-linked immunosorbent assay) for antigen-antibody reaction determination, etc. In addition, the laboratory has established a strict quality control system to monitor the accuracy of the experiment through regular testing of standard products and quality control products to ensure that each report is highly credible.

[0060] With the development of information technology, many hospitals have realized the information management of the inspection process. This means that from sample reception, barcode identification, data entry to the final result release, the entire process can be automatically tracked and recorded through the computer system. This not only improves work efficiency and reduces the possibility of human errors, but more importantly, it promotes information sharing between different departments. When the test results are generated, they will be uploaded to the hospital information system (HIS) immediately and connected to the patient's electronic medical record. In this way, both the attending physician responsible for preoperative preparation and the team members involved in postoperative care can access the latest test data at any time and make more timely and effective judgments.

[0061] In the embodiment of the present application, the vital sign data analysis module 130 is used to perform data sorting and multi-dimensional time series association on the time queue of the vital sign data to obtain multi-dimensional time series association characteristics of the vital sign. Figure 2 : is a block diagram of a vital sign data analysis module in a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 2 As shown, the vital signs data analysis module 130 includes: a vital signs data sorting unit 131, which is used to use a vital signs data sorter to sort the time queue of the vital signs data according to the time dimension and the vital signs parameter sample dimension to obtain a multi-dimensional data time series matrix of the patient's vital signs; a vital signs multi-dimensional time series association unit 132, which is used to pass the multi-dimensional time series data time series matrix of the patient's vital signs through a vital signs multi-dimensional time series association feature extractor based on MCNN to obtain a vital signs multi-dimensional time series association feature graph as the vital signs multi-dimensional time series association feature.

[0062] It should be understood that, considering that each vital sign parameter in the perioperative vital sign data is continuously generated over time, and there is an inherent correlation between these different vital sign parameters. Therefore, in order to intuitively see how each vital sign parameter changes over time, and at the same time to dig out the temporal relationship between different parameters, in the technical solution of the present application, a vital sign data organizer is used to organize the time queue of the vital sign data according to the time dimension and the vital sign parameter sample dimension to integrate these data into an orderly whole, and obtain a multi-dimensional data time series matrix of the patient's vital signs. In this way, the changing trend of each vital sign parameter over a period of time can be clearly observed, which is very important for understanding the evolution of the patient's physiological state.

[0063] Accordingly, considering the inherent associations between the various parameters of vital signs, these associations are crucial to understanding the physiological state and disease changes of patients. For example, changes in blood pressure may be related to changes in heart rate, and this association may be complex, nonlinear, and dynamically change over time. Therefore, in order to capture and refine the correlation characteristics between the various parameters in the time dimension, in the technical solution of the present application, the multidimensional data time series matrix of the vital signs of the patient object is passed through a multidimensional time series correlation feature extractor of vital signs based on MCNN, so as to learn the interaction mode of different vital sign parameters at different time points by sliding the convolution kernel on the time series and different sign dimensions, and obtain a multidimensional time series correlation feature map of vital signs. It can be understood that MCNN is a multi-channel convolutional neural network, which is an extended form of convolutional neural network (CNN). CNN is mainly used to process data with a grid structure, such as a two-dimensional image (whose grid structure is a row and column distribution of pixels). On this basis, MCNN processes different types of data or different feature representations of data by setting multiple channels. Specifically, it consists of multiple convolutional layers, pooling layers, and fully connected layers. The convolution layer performs convolution operations by sliding the convolution kernel on the data to extract local features from the multidimensional data of vital signs. The pooling layer is used to downsample the data to reduce the dimension of the data while retaining the main feature information. The fully connected layer is used to integrate the previously extracted features and output the final classification result or feature representation. When processing vital sign data, each channel can correspond to a vital sign (such as one channel for body temperature data, one channel for heart rate data, etc.), and then through multiple layers of convolution and pooling operations, the correlation patterns between different vital sign parameters over time are mined.

[0064] In the embodiment of the present application, the test result analysis module 140 is used to perform embedded coding and test item semantic association coding on the current test result of the hepatobiliary and pancreatic surgery patient object to obtain test item semantic association features. Figure 3 : is a block diagram of a test result analysis module in a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3 As shown, the test result analysis module 140 includes: a test result embedding coding unit 141, which is used to embed the current test result of the hepatobiliary and pancreatic surgery patient object through a test result embedding matrix to obtain a sequence of test item embedded semantic coding vectors; a test item semantic association unit 142, which is used to pass the sequence of test item embedded semantic coding vectors through a test item semantic association encoder based on an RNN-LSTM hybrid model to obtain a test item semantic association feature vector as the test item semantic association feature.

[0065] It should be understood that, considering the complex semantic associations between each test item result in the current test results of the hepatobiliary and pancreatic surgery patient object, in order to convert it into a digital representation that can be understood and calculated by the computer, and to mine the potential semantic associations at the same time, in the technical solution of the present application, the current test results of the hepatobiliary and pancreatic surgery patient object are embedded and encoded through the test result embedding matrix to give each test item a specific digital vector representation, and the originally abstract and diverse semantic information is mapped to a unified vector space, and a sequence of test item embedded semantic coding vectors is obtained. In particular, the test result embedding matrix is ​​constructed using a large amount of case information historical data.

[0066] Next, considering that there are different degrees of semantic association information between different test item semantic information. Therefore, in order to accurately capture the semantic association between different test items, so as to provide a comprehensive and deeper understanding of the test information, in the technical solution of the present application, the sequence of the test items embedded in the semantic coding vector is passed through the test item semantic association encoder based on the RNN-LSTM hybrid model to obtain the test item semantic association feature vector. In particular, the RNN (recurrent neural network)-LSTM (long short-term memory network) hybrid model is a powerful tool specifically used to process sequence data. RNN is able to process semantic information in sequence data. It transfers information from the previous moment to the next moment through a loop structure, so that it can capture the semantic association in the sequence. However, traditional RNN has the problem of gradient vanishing or gradient exploding when processing data with a long span, and LSTM, as a variant of RNN, effectively solves this problem through a special gating mechanism (input gate, forget gate and output gate), which is particularly suitable for capturing dependencies within a long span. Therefore, the RNN-LSTM hybrid model can not only capture the close correlation between local semantic features, but also deeply explore the mutual influence of semantic information within the span of long sentences, so as to have a more comprehensive understanding of the patient's condition.

[0067] In an embodiment of the present application, the cross-modal fine-grained encoding module 150 is used to perform fine-grained cross-modal attention encoding on the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain vital sign-test semantic cross-modal ablation coding features. Furthermore, considering that the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items both contain crucial and complementary information. Vital signs reflect the real-time changes in the physiological state of the patient, and the test results can reveal the specific conditions inside the patient's body from the perspectives of biochemistry, pathology, etc. Therefore, the data of these two different modalities are effectively integrated to generate a unified feature representation containing richer information. In the technical solution of the present application, the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items are fused and encoded through a fine-grained cross-modal attention encoding mechanism to obtain vital sign-test semantic cross-modal ablation coding features. That is, it can use the attention mechanism to automatically focus on those features in different modal data that are more critical for judging the condition, etc. At the same time, through fine-grained ablation, it can remove relatively redundant or unimportant information, and accurately extract the key fine-grained semantic features after the interactive fusion of the two modal data, thereby dynamically capturing the semantic correlation between them, so as to more deeply reveal the disease clues hidden behind the data.

[0068] Figure 4 : is a block diagram of a cross-modal fine-grained encoding module in a perioperative nutrition management system for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application. Specifically, Figure 4 As shown, the cross-modal fine-grained encoding module 150 includes: a vital sign fine-grained decoupling unit 151, which is used to perform feature fine-grained decoupling of the vital sign multidimensional temporal association feature map to obtain a set of vital sign multidimensional temporal association local feature matrices; a vital sign fine-grained ablation unit 152, which is used to perform fine-grained ablation optimization on each vital sign multidimensional temporal association local feature matrix in the set of vital sign multidimensional temporal association local feature matrices based on the test item semantic association feature vector to obtain a set of cross-domain guided ablation vital sign multidimensional temporal association local feature matrices; a vital sign feature aggregation unit 153, which is used to perform feature aggregation on the set of cross-domain guided ablation vital sign multidimensional temporal association local feature matrices to obtain a vital sign-test semantic cross-modal ablation coding feature map as the vital sign-test semantic cross-modal ablation coding feature.

[0069] Specifically, in the embodiment of the present application, the vital sign fine-grained decoupling unit 151 is used to: perform feature fine-grained decoupling of the vital sign multidimensional time series association feature graph along the channel dimension of the vital sign multidimensional time series association feature graph to obtain a set of the vital sign multidimensional time series association local feature matrix. The process can be expressed by the formula:

[0070] ;

[0071] in, is the multi-dimensional time series correlation characteristic diagram of the vital signs, For feature decoupling operations, , , and They are the first, second, and third in the set of multi-dimensional temporal correlation local feature matrices of vital signs. and A multi-dimensional temporal correlation local feature matrix of vital signs.

[0072] It should be understood that the fine-grained feature decoupling of the multi-dimensional time series correlation feature graph of the vital signs is performed to decompose the original vital signs data into smaller and more specific feature parts, and a set of multi-dimensional time series correlation local feature matrices of vital signs is obtained. In this way, the model can capture the change pattern in the time series more carefully, which helps to reveal the structural information hidden in the complex data.

[0073] Specifically, in an embodiment of the present application, the vital sign fine-grained ablation unit 152 includes: a vital sign ablation weight factor calculation subunit, which is used to calculate the fine-grained ablation weight factor of each vital sign multi-dimensional temporal association local feature matrix in the set of vital sign multi-dimensional temporal association local feature matrices based on the semantic association feature vector of the inspection item to obtain a set of vital sign multi-dimensional temporal association fine-grained ablation weight factors; a vital sign fine-grained ablation modulation subunit, which is used to perform fine-grained ablation modulation on the set of vital sign multi-dimensional temporal association local feature matrices based on the set of vital sign multi-dimensional temporal association fine-grained ablation weight factors to obtain the set of cross-domain guided ablation vital sign multi-dimensional temporal association local feature matrices.

[0074] More specifically, in the embodiment of the present application, the vital sign ablation weight factor calculation subunit is used to:

[0075] Taking each of the multidimensional time series correlation local feature matrices of vital signs in the set of the multidimensional time series correlation local feature matrices of vital signs as a key matrix, the semantic correlation feature vector of the inspection item and the multidimensional time series correlation local feature matrix of vital signs are encoded with ablation factors based on an imitation transformer to obtain a set of multidimensional time series correlation fine-grained ablation factors of vital signs. The process can be expressed by the formula:

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] in, It is the first in the set of multi-dimensional temporal correlation local feature matrices of vital signs. A multi-dimensional temporal correlation local feature matrix of vital signs, is the semantic association feature vector of the test item, is matrix multiplication, and They are the test item semantic association feature query embedding matrix and the test item semantic association feature query bias vector, is the semantic association feature query vector of the test item, and are the embedding matrix of the semantic association eigenvalues ​​of the test items and the bias vector of the semantic association eigenvalues ​​of the test items, respectively. is the semantic association eigenvalue vector of the test item, and They are respectively the embedding matrix of the multi-dimensional time series associated local feature key of vital signs and the bias vector of the multi-dimensional time series associated local feature key of vital signs, yes The corresponding vital signs multi-dimensional time series associated local feature key matrix, yes The transposed matrix of yes The scale is the width of the matrix multiplied by the height of the matrix, yes function, It is the first in the set of fine-grained interaction vectors of vital signs and test semantics. Vital sign-test semantic fine-grained interaction vectors, To extract the maximum value, and They are The mean and variance of is a hyperparameter, For calculation The fine-grained ablation factor, is the first in the set of fine-grained ablation factors associated with multi-dimensional temporal series of vital signs. Multi-dimensional temporal association of vital signs with fine-grained ablation factors;

[0082] The set of the multi-dimensional temporal association fine-grained ablation factors of the vital signs is subjected to ablation effect encoding including normalization and masking to obtain the set of the multi-dimensional temporal association fine-grained ablation weight factors of the vital signs. The process can be expressed by the formula:

[0083] ;

[0084] in, is the first in the set of fine-grained ablation factors associated with multi-dimensional temporal series of vital signs. The multi-dimensional temporal correlation fine-grained ablation factor of each vital sign, For calculation The fine-grained ablation weight factor, represents the exponential function value with the natural constant e as the base, It is the first in the set of normalized vital sign multi-dimensional temporal association fine-grained ablation factors. Normalized vital sign multi-dimensional temporal correlation fine-grained ablation factor, is a mask operation, is the preset threshold, is the first in the set of fine-grained ablation weight factors of multi-dimensional temporal association of vital signs The multi-dimensional temporal association fine-grained ablation weight factor of each vital sign.

[0085] Accordingly, the multi-dimensional temporal correlation local feature matrix of each vital sign is used as the key matrix, and the semantic correlation feature vector of the test item is linearly transformed to obtain the query and value vectors, which are then encoded with the key matrix by the ablation factor based on the imitation transformer to obtain a set of fine-grained ablation factors. In particular, the imitation transformer draws on the idea of ​​the self-attention mechanism, which can effectively capture the relationship between the two modalities, and use it to dynamically measure the degree of correlation between each local feature matrix and the semantic correlation features of the test item, find those fine-grained factors that play a key role in the final semantic understanding and disease judgment in the cross-modal fusion process, and encode them into fine-grained ablation factors.

[0086] It should be understood that the ablation effect encoding including normalization and masking is performed on each of the fine-grained ablation factors to obtain a set of multi-dimensional temporal correlation fine-grained ablation weight factors of vital signs. That is, normalization ensures that all features are compared on the same scale to avoid some features dominating other features due to excessive numerical ranges. Through the masking operation, data points that are considered unimportant or abnormal can be selectively ignored, while emphasizing the most representative and decisive features, and the weight factor finally generated reflects the relative importance of each fine-grained local feature in the entire system, providing guidance for the next step.

[0087] More specifically, in the embodiment of the present application, the vital sign fine-grained ablation modulation subunit is used to perform fine-grained ablation modulation on the set of the vital sign multi-dimensional time series associated local feature matrices based on the set of the vital sign multi-dimensional time series associated fine-grained ablation weight factors to obtain the set of the cross-domain guided ablation vital sign multi-dimensional time series associated local feature matrices. The process can be expressed by the formula: ,in, It is the first in the set of multi-dimensional temporal correlation local feature matrices of vital signs. A multi-dimensional temporal correlation local feature matrix of vital signs, is the first in the set of fine-grained ablation weight factors of multi-dimensional temporal association of vital signs The multi-dimensional temporal association fine-grained ablation weight factor of each vital sign, It is the first in the set of multi-dimensional temporal correlation local feature matrices of cross-domain guided ablation vital signs. A multi-dimensional temporal correlation local feature matrix of vital signs in cross-domain guided ablation.

[0088] Accordingly, based on the set of fine-grained ablation weight factors of multi-dimensional temporal association of vital signs, the set of local feature matrices of multi-dimensional temporal association of vital signs is fine-grainedly ablated to guide the model to focus on those parts of information that are proven to be critical, and obtain the set of cross-domain guided ablation local feature matrices of multi-dimensional temporal association of vital signs. In this way, the modulated feature matrix not only retains the original time series characteristics, but also incorporates the understanding from the semantics of the test items, promoting the effective exchange of information between two different domains.

[0089] Specifically, the vital sign feature aggregation unit 153 is used to perform feature aggregation on the set of the cross-domain guided ablation vital sign multi-dimensional temporal correlation local feature matrix to obtain a vital sign-inspection semantic cross-modal ablation coding feature map as the vital sign-inspection semantic cross-modal ablation coding feature. The process can be expressed by the formula: ,in, , , and It is the first, second, and third in the set of multi-dimensional temporal correlation local feature matrices of cross-domain guided ablation vital signs. and A multi-dimensional temporal correlation local feature matrix of cross-domain guided ablation vital signs, It is the vital sign-test semantic cross-modal ablation encoding feature map.

[0090] Finally, the set of multi-dimensional temporal correlation local feature matrices of the cross-domain guided ablation vital signs are feature aggregated to form a complete and comprehensive vital sign-test semantic cross-modal ablation coding feature map, which combines the time series information of vital signs and the semantic information of test results, providing strong support for the generation of subsequent nutritional management plans.

[0091] In the embodiment of the present application, the surgery type label acquisition module 160 is used to obtain the surgery type label of the hepatobiliary and pancreatic surgery patient object. It should be understood that the surgery type label is a classification mark for the surgery performed by the hepatobiliary and pancreatic surgery patient, which is used to clearly define which category the surgery belongs to. For example, in the field of hepatobiliary surgery, common surgery type labels include partial liver resection, cholecystectomy, liver transplantation, etc.; in pancreatic surgery, there are pancreaticoduodenectomy, pancreatic body and tail resection, etc.; in biliary surgery, there are common bile duct exploration and lithotomy, radical resection of bile duct cancer, etc. It can concisely summarize the core operation of the surgery and the main anatomical parts involved. Based on this, in order to be more comprehensive and generate a reasonable nutrition management plan, in the technical solution of the present application, the surgery type label of the hepatobiliary and pancreatic surgery patient object is obtained. For example, patient A is diagnosed with gallstones and needs to undergo cholecystectomy. Based on this surgery type label, low-fat liquid food is given in the early postoperative period, and gradually transitions with recovery to prevent steatorrhea. Patient B has liver cancer and needs partial liver resection. After the operation, the nutritional supply is precisely adjusted according to the liver's recovery to ensure physical recovery.

[0092] First, the attending physician in charge of the patient will make a preliminary judgment on the possible applicable surgical method based on the patient's specific condition, imaging examination results, and other relevant test data. This step is usually completed at a multidisciplinary team (MDT) discussion meeting, with participants including surgeons, internists, radiologists, pathologists and other experts in various fields. By comprehensively analyzing the patient's clinical manifestations, tumor stage, liver reserve function and other factors, the team can reach a consensus and choose the most suitable surgical plan for the patient. This preliminary decision will be recorded and used as an important basis for the next step.

[0093] Next, in order to formally determine the surgery type label, a detailed surgery request form must be filled out by the surgeon or his designated assistant. This form is not only the basic document for surgical scheduling, but also the key carrier for obtaining the surgery type label. It contains all the necessary information about the surgery, such as the name of the surgery, the expected length of time, the anesthesia method, the expected risks and complications, etc. More importantly, there is a field on this form specifically for marking the type of surgery, where standard terms are required to describe the specific type of surgery to be performed. For example, "partial hepatectomy", "cholecystectomy" or "pancreaticoduodenectomy". These terms are widely recognized and standardized by the medical community, ensuring the consistency and accuracy of information exchange between different hospitals.

[0094] Once the surgery request form is ready, it will be submitted to the hospital's surgery scheduling department for review and arrangement. In this process, the information system plays an important role. Modern hospitals generally use integrated electronic medical record systems (EMR), which can automatically read the information in the surgery request form and convert it into a structured data format for storage. In this way, all medical staff related to the patient can access the latest surgery arrangement details through authorization. At the same time, the system will also generate a unique identifier, the surgery number, which is used to track the entire surgical process and its subsequent treatment progress. This number is also associated with the surgery type tag, so that any query can quickly locate the specific surgery information.

[0095] In an embodiment of the present application, the nutrition management plan generation module 170 is used to generate a nutrition management plan based on the vital signs-test semantic cross-modal ablation coding features and the surgery type label. Specifically, in an embodiment of the present application, the nutrition management plan generation module is used to: after expanding the vital signs-test semantic cross-modal ablation coding feature map into a vital signs-test semantic cross-modal ablation coding feature vector, add the surgery type label and generation prompt information to the tail of the vital signs-test semantic cross-modal ablation coding feature vector, and input it into the AIGC-based perioperative nutrition manager to generate the nutrition management plan, and the generation prompt information is "Based on the input corpus, generate the nutrition management plan for the hepatobiliary and pancreatic surgery patient object." In this way, after receiving clear instructions, the model can understand that it needs to generate a nutrition management plan specifically for the hepatobiliary and pancreatic surgery patient based on the comprehensive information such as the vital signs, test results and surgery type provided previously. This clear task prompt helps guide the model to reason and generate in the expected direction, avoids the model outputting content that is irrelevant to nutritional management, improves the accuracy and pertinence of the generated plan, enables it to better meet actual clinical needs, and provide scientific, reasonable, and personalized guidance and suggestions for patients' perioperative nutritional support.

[0096] In particular, after the vital sign-inspection semantic cross-modal ablation coding feature map is expanded into a vital sign-inspection semantic cross-modal ablation coding feature vector, the surgery type label and generation prompt information are added to the tail of the vital sign-inspection semantic cross-modal ablation coding feature vector, and it is input into the AIGC-based perioperative nutrition manager to generate a nutrition management plan. The specific processing process is as follows:

[0097] 1. Take the expanded vital sign-test semantic cross-modal ablation coding feature vector as the basic data, and add the surgery type label and generation prompt information to the tail of the feature vector in sequence to form a new combined data sequence. For example, if the feature vector is [X1, X2, X3, ..., Xn], the surgery type label is "partial liver resection", and the generation prompt information is "Based on the input corpus, generate the nutrition management plan for the hepatobiliary and pancreatic surgery patient object", then the new sequence after splicing is [X1, X2, X3, ..., Xn, "partial liver resection", "Based on the input corpus, generate the nutrition management plan for the hepatobiliary and pancreatic surgery patient object"].

[0098] 2. The AIGC-based perioperative nutrition manager first needs to parse the input vital signs-test semantic cross-modal ablation encoding feature vector, surgery type label, and generate prompt information. These data together provide the model with a comprehensive description of the patient's status and clear task instructions. By parsing the input data and initializing relevant parameters, the model can understand the patient's specific situation and prepare for the next treatment.

[0099] 3. After receiving and parsing the input, the AIGC model will combine the feature vectors of vital signs and test results with the surgery type label, and use attention mechanisms and other methods to achieve the fusion of multimodal features. At the same time, it generates a prompt message "Based on the input corpus, generate a nutrition management plan for the hepatobiliary and pancreatic surgery patient object" to help build an appropriate contextual environment and guide the model to generate content that meets the requirements. Next, the model activates submodules or paths specific to the type of surgery, accesses the built-in knowledge base or external database, and retrieves nutritional recommendations, dietary guidelines, and other relevant information related to the current patient condition.

[0100] 4. As the internal reasoning proceeds, the AIGC model not only relies on predefined text templates for content filling, but also applies natural language generation technology to automatically generate fluent and easy-to-understand text content, detailing the various nutrients needed daily and their sources (food or supplements), as well as special dietary guidance, etc. In this process, the model will consider the balance between the professionalism of medical terminology and the language style that is easy for ordinary patients to understand, ensuring that the generated plan is both scientific and easy to understand. In addition, based on individual differences in patients, such as age, gender, underlying diseases, etc., the model further optimizes the generated nutritional management plan, and proposes preventive measures or alternative plans considering the risks of adverse reactions or complications.

[0101] In a preferred example, considering that the multi-dimensional temporal association feature map of vital signs and the semantic association feature vector of test items respectively represent the sample-temporal cross-dimensional association features of the patient's vital signs data and the embedded coding features of the patient's current test results, when performing cross-domain attention joint encoding based on fine-grained ablation, due to the differences in cross-domain ablation mechanisms caused by the differences in heteromodal feature distribution of source data, the vital sign-test semantic cross-modal ablation coding feature map will also have probabilistic convergence and divergence based on different attention joint coding enhancements, thereby affecting the textual semantic quality of the nutrition management plan generated by the AIGC-based perioperative nutrition manager.

[0102] Therefore, the present application adds the surgery type label and generation prompt information to the tail of the vital sign-inspection semantic cross-modal ablation coding feature vector after the vital sign-inspection semantic cross-modal ablation coding feature map is expanded, and inputs it into the AIGC-based perioperative nutrition manager to generate a nutrition management plan. First, the vital sign-inspection semantic cross-modal ablation coding feature vector is optimized, including the following steps:

[0103] Determine the median eigenvalue, maximum eigenvalue and minimum eigenvalue in the feature set of the vital sign-test semantic cross-modal ablation coding feature vector, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the vital sign-test semantic cross-modal ablation coding distribution probability value, that is, ;in, , and are respectively the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the vital sign-test semantic cross-modal ablation encoding feature vector, is the probability value of the vital sign-test semantic cross-modal ablation coding distribution;

[0104] Each eigenvalue in the vital sign-test semantic cross-modal ablation coding feature vector is normalized to the maximum value to obtain the vital sign-test semantic cross-modal ablation coding probability eigenvalue ;

[0105] in, is the first in the vital sign-test semantic cross-modal ablation encoding feature vector Vital sign-test semantic cross-modal ablation encoding probability feature value after maximum normalization of feature values;

[0106] Subtract the vital sign-test semantic cross-modal ablation coding distribution probability value from one Compute each of the vital signs for the index - check semantic cross-modal ablation encoding probability feature value Power function to obtain the first vital sign - test semantic cross-modal ablation encoding probability convergence eigenvalue ;

[0107] in, is the first in the vital sign-test semantic cross-modal ablation encoding feature vector The first vital sign corresponding to the eigenvalue - check the semantic cross-modal ablation coding probability convergence eigenvalue;

[0108] Using the vital sign-test semantic cross-modal ablation coding distribution probability value Calculate a probability feature value for the index with each of the vital signs-test semantic cross-modal ablation encoding The power function of the difference of to obtain the second vital sign - to test the semantic cross-modal ablation encoding probability convergence eigenvalue ;

[0109] in, is the first in the vital sign-test semantic cross-modal ablation encoding feature vector The second vital sign corresponding to the eigenvalues ​​- test the semantic cross-modal ablation encoding probability convergence eigenvalues;

[0110] Each of the vital sign-test semantic cross-modal ablation encoding probability feature values The probability value of the distribution of the vital sign-test semantic cross-modal ablation coding is subtracted from one Multiply the difference to obtain the first vital sign - check the semantic cross-modal ablation encoding probability limit feature value ;

[0111] in, is the first in the vital sign-test semantic cross-modal ablation encoding feature vector The first vital sign corresponding to the feature value - the test semantic cross-modal ablation coding probability limit feature value;

[0112] Combine one with each of the vital signs-test semantic cross-modal ablation encoding probability feature value The difference between the vital signs-test semantic cross-modal ablation coding distribution probability value Multiply to get the second vital sign - test semantic cross-modal ablation encoding probability limit feature value ;

[0113] in, is the first in the vital sign-test semantic cross-modal ablation encoding feature vector The second vital sign corresponding to the feature value - check the semantic cross-modal ablation coding probability limit feature value;

[0114] After multiplying the first vital sign-test semantic cross-modal ablation coding probability convergence eigenvalue by the second vital sign-test semantic cross-modal ablation coding probability convergence eigenvalue, further add the first vital sign-test semantic cross-modal ablation coding probability limiting eigenvalue and the second vital sign-test semantic cross-modal ablation coding probability limiting eigenvalue to obtain the optimized eigenvalue of the vital sign-test semantic cross-modal ablation coding feature vector ;

[0115] in, is the first in the vital sign-test semantic cross-modal ablation encoding feature vector The optimized eigenvalues ​​of eigenvalues;

[0116] Based on this, by using the cross-entropy form power series of the regression probability of the vital sign-test semantic cross-modal ablation coding feature vector as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the vital sign-test semantic cross-modal ablation coding feature vector, so as to further guide the limited recovery strategy through the probability distribution cross-entropy as the objective function under the premise of avoiding the conservative convergence control of the fatal potential divergence of the vital sign-test semantic cross-modal ablation coding feature vector, so as to achieve the common agility of the convergence of the probability convergent and divergent feature set of the vital sign-test semantic cross-modal ablation coding feature vector to the probability density distribution space, and improve the text semantic quality of the generated results of the AIGC-based perioperative nutrition manager.

[0117] In summary, the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients based on the embodiment of the present application is explained, which collects the time queue of the vital signs data (body temperature, heart rate, blood pressure, blood oxygen saturation and weight) of the hepatobiliary and pancreatic surgery patients during the perioperative period, obtains the current test results and surgery type labels of the hepatobiliary and pancreatic surgery patients, uses data analysis and coding technology based on deep learning to sort the vital signs data and multi-dimensional time series association, and semantically embeds and associates the current test results, and then performs fine-grained cross-modal encoding of the multi-dimensional time series association features of the vital signs and the semantic association features of the test items, so as to intelligently generate a nutrition management plan based on the semantic cross-modal coding representation and the surgery type label. This technical solution can fully capture the physiological condition and test characteristics of each patient, and then tailor a unique nutrition treatment plan for the patient. In addition, it also supports real-time monitoring and flexible adjustment of the patient's condition, and can quickly optimize the treatment plan even in the face of complex and changing conditions, ensuring that appropriate intervention measures are taken at critical moments.

[0118] As described above, the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a perioperative nutrition management algorithm for hepatobiliary and pancreatic surgery patients. In one possible implementation, the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients can also be one of the many hardware modules of the wireless terminal.

[0119] Alternatively, in another example, the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients and the wireless terminal may also be separate devices, and the perioperative nutrition management system 100 for hepatobiliary and pancreatic surgery patients may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0120] Figure 5 Flow chart of the perioperative nutritional management method for hepatobiliary and pancreatic surgery patients according to an embodiment of the present application. Figure 5As shown, according to the perioperative nutritional management method for hepatobiliary and pancreatic surgery patients of the embodiment of the present application, the method comprises: S110, collecting a time queue of vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data comprises a body temperature value, a heart rate value, a blood pressure value, a blood oxygen saturation and a weight value; S120, obtaining the current test result of the hepatobiliary and pancreatic surgery patient; S130, sorting the time queue of the vital sign data and performing multi-dimensional time series association to obtain a multi-dimensional time series association feature of the vital sign; S140, The current test results of the hepatobiliary and pancreatic surgery patient object are embedded coded and the test item semantic association coded to obtain the test item semantic association features; S150, the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items are fine-grained cross-modal attention encoded to obtain the vital sign-test semantic cross-modal ablation coding features; S160, the surgery type label of the hepatobiliary and pancreatic surgery patient object is obtained; S170, a nutrition management plan is generated based on the vital sign-test semantic cross-modal ablation coding features and the surgery type label.

[0121] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned method for perioperative nutritional management of hepatobiliary and pancreatic surgery patients have been referenced above. Figures 1 to 4 The description of the perioperative nutritional management system for hepatobiliary and pancreatic surgical patients has been described in detail, and therefore, its repeated description will be omitted.

[0122] The above has described various implementations of the present disclosure, and the above description is exemplary and not exhaustive. It is not limited to the disclosed implementations, and many modifications and changes are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The choice of terms used in this article is intended to best explain the principles of each implementation, practical application or improvement of technology in the market, or to enable other ordinary technicians in this technical field to understand the various embodiments disclosed herein.

Claims

1. A perioperative nutrition management system for hepatobiliary and pancreatic surgery patients, characterized in that: include: A vital sign data collection module is used to collect a time queue of vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data includes body temperature, heart rate, blood pressure, blood oxygen saturation and weight; A test result acquisition module, used to acquire the current test result of the hepatobiliary and pancreatic surgery patient object; A vital sign data analysis module, used for data sorting and multi-dimensional time series correlation of the time queue of the vital sign data to obtain multi-dimensional time series correlation characteristics of the vital sign; A test result analysis module, used for performing embedded coding and test item semantic association coding on the current test result of the hepatobiliary and pancreatic surgery patient object to obtain test item semantic association features; A cross-modal fine-grained encoding module, used for performing fine-grained cross-modal attention encoding on the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain a vital sign-test semantic cross-modal ablation encoding feature; A surgery type label acquisition module, used to acquire the surgery type label of the hepatobiliary and pancreatic surgery patient object; A nutrition management plan generation module, used for generating a nutrition management plan based on the vital sign-test semantic cross-modal ablation coding features and the surgery type label; Wherein, the vital signs data analysis module includes: A vital sign data collating unit, configured to use a vital sign data collator to collate the time queue of the vital sign data according to the time dimension and the vital sign parameter sample dimension to obtain a multi-dimensional data time series matrix of the vital signs of the patient; A vital sign multidimensional time series association unit, used for passing the multidimensional data time series matrix of the patient's vital signs through a multidimensional time series association feature extractor of vital signs based on MCNN to obtain a multidimensional time series association feature graph of vital signs as the multidimensional time series association feature of vital signs; The cross-modal fine-grained encoding module includes: A vital sign fine-grained decoupling unit, used for performing feature fine-grained decoupling on the multi-dimensional time series correlation feature graph of the vital sign to obtain a set of multi-dimensional time series correlation local feature matrices of the vital sign; A vital sign fine-grained ablation unit, configured to perform fine-grained ablation optimization on each vital sign multi-dimensional temporal correlation local feature matrix in the set of the vital sign multi-dimensional temporal correlation local feature matrix based on the test item semantic correlation feature vector to obtain a set of cross-domain guided ablation vital sign multi-dimensional temporal correlation local feature matrices; A vital sign feature aggregation unit, used for performing feature aggregation on the set of the cross-domain guided ablation vital sign multi-dimensional temporal correlation local feature matrix to obtain a vital sign-test semantic cross-modal ablation coding feature map as the vital sign-test semantic cross-modal ablation coding feature; Among them, the nutrition management plan generation module is used to: after expanding the vital signs-test semantic cross-modal ablation coding feature map into a vital signs-test semantic cross-modal ablation coding feature vector, add the surgery type label and generation prompt information to the tail of the vital signs-test semantic cross-modal ablation coding feature vector, and input it into the AIGC-based perioperative nutrition manager to generate the nutrition management plan, and the generation prompt information is "Based on the input corpus, generate a nutrition management plan for the hepatobiliary and pancreatic surgery patient object." 2. The perioperative nutritional management system for hepatobiliary and pancreatic surgery patients according to claim 1, characterized in that: The test result analysis module comprises: A test result embedding coding unit, used for embedding coding the current test result of the hepatobiliary and pancreatic surgery patient object through a test result embedding matrix to obtain a sequence of test item embedding semantic coding vectors; The test item semantic association unit is used to embed the sequence of the test item semantic coding vectors through a test item semantic association encoder based on an RNN-LSTM hybrid model to obtain a test item semantic association feature vector as the test item semantic association feature.

3. The perioperative nutritional management system for hepatobiliary and pancreatic surgery patients according to claim 2, characterized in that: The vital sign fine-grained decoupling unit is used to perform feature fine-grained decoupling on the vital sign multidimensional time series association feature graph along the channel dimension of the vital sign multidimensional time series association feature graph to obtain a set of the vital sign multidimensional time series association local feature matrices.

4. The perioperative nutritional management system for hepatobiliary and pancreatic surgery patients according to claim 3, characterized in that: The vital sign fine-grained ablation unit comprises: A vital sign ablation weight factor calculation subunit, used to calculate the fine-grained ablation weight factor of each vital sign multidimensional time series association local feature matrix in the set of vital sign multidimensional time series association local feature matrices based on the test item semantic association feature vector to obtain a set of vital sign multidimensional time series association fine-grained ablation weight factors; The vital sign fine-grained ablation modulation subunit is used to perform fine-grained ablation modulation on the set of vital sign multi-dimensional time series associated local feature matrices based on the set of vital sign multi-dimensional time series associated fine-grained ablation weight factors to obtain the set of cross-domain guided ablation vital sign multi-dimensional time series associated local feature matrices.

5. The perioperative nutritional management system for hepatobiliary and pancreatic surgery patients according to claim 4, characterized in that: The vital sign ablation weight factor calculation subunit is used to: Using each of the vital sign multidimensional time series association local feature matrices in the set of the vital sign multidimensional time series association local feature matrices as a key matrix, performing ablation factor encoding based on a simulated converter on the test item semantic association feature vector and the vital sign multidimensional time series association local feature matrix to obtain a set of vital sign multidimensional time series association fine-grained ablation factors; The set of multi-dimensional time series associated fine-grained ablation factors of the vital signs is subjected to ablation effect encoding including normalization and masking to obtain the set of multi-dimensional time series associated fine-grained ablation weight factors of the vital signs.

6. A method for perioperative nutritional management of hepatobiliary and pancreatic surgery patients, characterized in that: include: A time queue for collecting vital sign data of hepatobiliary and pancreatic surgery patients during the perioperative period, wherein the vital sign data includes body temperature, heart rate, blood pressure, blood oxygen saturation and weight; Obtaining current test results of the hepatobiliary and pancreatic surgery patient object; Performing data sorting and multi-dimensional time series correlation on the time queue of the vital sign data to obtain multi-dimensional time series correlation characteristics of the vital sign; Performing embedded coding and test item semantic association coding on the current test result of the hepatobiliary and pancreatic surgery patient object to obtain test item semantic association features; Performing fine-grained cross-modal attention encoding on the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain a vital sign-test semantic cross-modal ablation encoding feature; Acquire a surgery type label of the hepatobiliary and pancreatic surgery patient object; generating a nutrition management plan based on the vital sign-test semantic cross-modal ablation coding features and the surgery type label; The time queue of the vital sign data is sorted and multi-dimensional time series is associated to obtain multi-dimensional time series association features of the vital sign, including: Using a vital sign data organizer to organize the time queue of the vital sign data according to the time dimension and the vital sign parameter sample dimension to obtain a multi-dimensional data time series matrix of the patient's vital signs; The multidimensional data time series matrix of the patient's vital signs is passed through a multidimensional time series correlation feature extractor of vital signs based on MCNN to obtain a multidimensional time series correlation feature graph of vital signs as the multidimensional time series correlation feature of vital signs; Among them, fine-grained cross-modal attention encoding is performed on the multi-dimensional temporal association features of the vital signs and the semantic association features of the test items to obtain the vital signs-test semantic cross-modal ablation encoding features, including: Performing feature fine-grained decoupling on the multi-dimensional time series correlation feature graph of the vital signs to obtain a set of multi-dimensional time series correlation local feature matrices of the vital signs; Based on the semantic association feature vector of the inspection item, fine-grained ablation optimization is performed on each of the multi-dimensional temporal association local feature matrices of the vital signs in the set of the multi-dimensional temporal association local feature matrices of the vital signs to obtain a set of cross-domain guided ablation multi-dimensional temporal association local feature matrices of the vital signs; Performing feature aggregation on the set of the cross-domain guided ablation vital sign multi-dimensional temporal correlation local feature matrices to obtain a vital sign-test semantic cross-modal ablation coding feature map as the vital sign-test semantic cross-modal ablation coding feature; Among them, based on the vital signs-test semantic cross-modal ablation coding features and the surgery type label, a nutrition management plan is generated, including: after expanding the vital signs-test semantic cross-modal ablation coding feature map into a vital signs-test semantic cross-modal ablation coding feature vector, adding the surgery type label and generation prompt information to the tail of the vital signs-test semantic cross-modal ablation coding feature vector, and inputting it into the AIGC-based perioperative nutrition manager to generate the nutrition management plan, and the generation prompt information is "based on the input corpus, generate a nutrition management plan for the hepatobiliary and pancreatic surgery patient object."

Images