Method and apparatus for generating clinical decision support information based on multi-modal parameters

Abstract

The application provides a kind of clinical auxiliary decision information generation method and device based on multi-modal parameters, method includes: the body parameter feature and physiological parameter set of first user are collected;Physiological parameter set is space-time alignment and feature extraction are carried out, and multidimensional feature vector is obtained;Through the risk prediction model of pre-construction, complication risk probability prediction is carried out according to multidimensional feature vector and body parameter feature, and the risk probability of multiple events in multiple time windows is obtained;According to the risk probability of multiple events in multiple time windows, multidimensional feature vector and body parameter feature, decision analysis is carried out, and decision support information is generated, can obtain the risk probability information of multiple time windows related to illness in advance, realize the dynamic tracking of user state;On this basis, risk probability and feature information are combined for decision analysis, and structured decision support information with explainability and operability is generated, forming an automated processing flow from data acquisition to information output.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, particularly to the field of artificial intelligence technology, and especially to a method and apparatus for generating clinical auxiliary decision-making information based on multimodal parameters. Background Technology

[0002] Patients with liver failure often experience severe impairment in their liver's synthetic, metabolic, and immune regulatory functions, resulting in an immune imbalance characterized by both "immune paralysis" and "excessive inflammation." This makes them a high-risk group for sepsis and multiple organ failure. Early access to disease-related information and timely intervention are crucial for improving patient prognosis.

[0003] Currently, obtaining information related to complications in patients with liver failure mainly relies on clinical symptom observation and traditional laboratory indicators (such as PCT and CRP). However, these methods suffer from difficulties in interpreting indicators and poor specificity in liver disease patients. Furthermore, these indicators typically only show significant changes in the middle and late stages of the disease, making it difficult to obtain early indications of the condition. Some existing technologies attempt to obtain infection-related information by detecting single or a few biomarkers, but these methods only focus on a single pathological link and cannot comprehensively reflect the multiple pathological mechanisms in liver failure patients, such as immune imbalance, intestinal barrier damage, and microcirculatory disturbances, resulting in incomplete information. With the development of multi-omics technologies, existing technologies have emerged that obtain multi-dimensional biomarkers through high-throughput detection. However, these methods only remain at the data detection level, with multi-omics data disconnected from the patient's continuous clinical information, lacking a systematic data integration and fusion analysis mechanism. In addition, a few existing technologies use traditional statistical models for information processing, but these models have limited ability to handle high-dimensional, nonlinear, and time-series data, and the processing results lack interpretability, making it difficult to translate into specific intervention guidance information.

[0004] In summary, existing technologies suffer from drawbacks such as delayed information acquisition, fragmented multi-source data, and a lack of interpretability in processing results, failing to achieve early, accurate, and dynamic acquisition of risk information related to complications in patients with liver failure. Therefore, there is an urgent need for a system that can integrate early biomarkers reflecting immune status, intestinal barrier function, endothelial function, and organ-specific damage, combined with continuous clinical data, and utilize artificial intelligence technology to achieve early and accurate acquisition of disease-related risk information, while also providing guidance on mechanistic interventions.

[0005] This section is intended to provide background or context for the embodiments of this application set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0006] One objective of this invention is to provide a method for generating clinical auxiliary decision-making information based on multimodal parameters. This method can acquire risk probability information related to the patient's condition over multiple time windows in advance, enabling dynamic tracking of the user's status. Based on this, it combines risk probability and feature information for decision analysis, generating structured decision support information that is interpretable and operable, forming an automated processing flow from data acquisition to information output. Another objective of this invention is to provide a device for generating clinical auxiliary decision-making information based on multimodal parameters. A further objective of this invention is to provide a computer-readable medium. A final objective of this invention is to provide a computer device.

[0007] To achieve the above objectives, this invention discloses a method for generating clinical auxiliary decision-making information based on multimodal parameters, comprising: Collect the body shape and physiological parameter set of the first user; Spatiotemporal alignment and feature extraction of physiological parameter sets are performed to obtain multi-dimensional feature vectors; By using a pre-built risk prediction model, the probability of complication risk is predicted based on multi-dimensional feature vectors and shape parameter features, and the probability of multiple events within multiple time windows is obtained. Decision analysis is performed based on the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features to generate decision support information.

[0008] Preferably, the set of physiological parameters includes physical signs, test parameters, and biomarker parameters; Collect the body shape features and physiological parameter set of the first user, including: Query the shape parameter features using the first user identifier; Vital signs are collected through the first data interface of the bedside monitoring device; Test parameters are collected through the second data interface of the hospital information system; The marker parameters are collected through the third data interface of the bedside monitoring device.

[0009] Preferably, the physiological parameter set is spatiotemporally aligned and features are extracted to obtain a multi-dimensional feature vector, including: The physiological parameter set is preprocessed to obtain the preprocessed physiological parameter set. Data preprocessing includes data cleaning and standardization. Using the admission time of the first user as the origin of the baseline time axis, the preprocessed physiological parameter set is spatiotemporally aligned to obtain a time series data matrix. The time series data matrix is ​​sliced ​​according to a preset time window and sliding step size to generate time series data segments. Multi-dimensional feature extraction is performed on time-series data segments to generate multi-dimensional feature vectors.

[0010] Preferably, multi-dimensional feature extraction is performed on time-series data segments to generate multi-dimensional feature vectors, including: Calculate the temporal characteristics of a time-series data segment; Frequency domain features are extracted from time-series data segments using wavelet transform. Calculate the correlation characteristics among various parameters in a time-series data segment; Multi-dimensional feature vectors are generated based on time-domain features, frequency-domain features, and correlation features.

[0011] Preferably, the method further includes: Obtain full multimodal parameters from multiple historical users; The model is trained on the pre-defined hybrid deep learning architecture using the weighted cross-entropy loss function based on all multimodal parameters to build an initial prediction model. The hybrid deep learning architecture consists of a coupled temporal network and a graph attention network. The initial prediction model is optimized through cross-validation to construct a risk prediction model.

[0012] Preferably, decision support information is generated by performing decision analysis based on the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features, including: Interpretability testing is performed based on the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features to extract key feature information; Risk level information is generated based on the risk probabilities of multiple events within multiple time windows; The system generates decision support information based on key feature information and risk level information through a pre-set clinical knowledge engine.

[0013] Preferably, interpretability detection is performed based on the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features to extract key feature information, including: By using an interpretability detection method, the contribution of multi-dimensional feature vectors and shape parameter features is detected based on the risk probability of multiple events within multiple time windows, and feature contribution information is obtained. Based on the feature contribution information, key features are selected and the corresponding key feature information is determined.

[0014] Preferably, risk level information is generated based on the risk probabilities of multiple events within multiple time windows, including: Based on a preset risk level threshold, the risk level information for each event within each time window is generated by comparing the risk probabilities of multiple events within multiple time windows.

[0015] Preferably, the method further includes: If the risk level information is high risk, the steps of predicting the risk probability of complications based on the pre-built risk prediction model and the multi-dimensional feature vector and shape parameter features are repeated a specified number of times to obtain the risk probability of multiple events in multiple time windows. Based on the risk probabilities of multiple events within multiple time windows, risk level information for each event within each time window is generated.

[0016] Preferably, the method further includes: Obtain all multimodal parameters of the first user, and behavioral and feedback data of the second user; Based on full multimodal parameters, behavioral data, and feedback data, the risk prediction model is incrementally learned to generate an updated risk prediction model.

[0017] The present invention also discloses a clinical auxiliary decision information generation system based on multimodal parameters, which is applied to the clinical auxiliary decision information generation method based on multimodal parameters as described above. The system includes: a computer device and a first user database and multiple data interfaces respectively connected to the computer device. The computer equipment is used to execute the clinical auxiliary decision-making information generation method based on multimodal parameters as described above; The first user database is used to store the multimodal data of the first user, which includes body parameter features and physiological parameter sets; The data interface is used for data transmission.

[0018] Preferably, the multiple data interfaces include: a first data interface, a second data interface, a third data interface, and a fourth data interface; the physiological parameter set includes vital sign parameters, test parameters, and biomarker parameters; The system also includes: bedside monitoring equipment, hospital information system, and bedside testing equipment; The first data interface is used to transmit the vital signs parameters detected by the bedside monitoring equipment to the computer equipment; The second data interface is used to transmit the test parameters recorded by the hospital information system to the computer equipment; The third data interface is used to transmit the marker parameters detected by the bedside monitoring equipment to the computer equipment; The fourth data interface is used to transmit vital signs, test parameters, and biomarker parameters from computer equipment to the first user database.

[0019] Preferably, the system also includes: a historical user database connected to a computer device; The historical user database is used to store the full multimodal data of historical users.

[0020] Preferably, the system also includes: a knowledge engine database connected to a computer device; The knowledge engine database is used to store clinical knowledge engines, which include the mapping relationship between decision support information, key feature information, and risk level information.

[0021] Preferably, the system also includes: a second user database; The second user database is used to store the behavioral and feedback data of the second users.

[0022] This invention also discloses a clinical auxiliary decision-making information generation device based on multimodal parameters, comprising: The acquisition unit is used to acquire the body parameter features and physiological parameter set of the first user; The data processing unit is used to perform spatiotemporal alignment and feature extraction on the physiological parameter set to obtain a multi-dimensional feature vector. The risk probability prediction unit is used to predict the risk probability of complications based on multi-dimensional feature vectors and shape parameter features through a pre-built risk prediction model, and to obtain the risk probability of multiple events within multiple time windows. The decision analysis unit is used to perform decision analysis based on the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features, and generate decision support information.

[0023] The present invention also discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0024] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, wherein the processor executes the program to implement the method described above.

[0025] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method described above.

[0026] This invention collects the body shape parameters and physiological parameter set of a first user; performs spatiotemporal alignment and feature extraction on the physiological parameter set to obtain a multi-dimensional feature vector; through a pre-constructed risk prediction model, it predicts the probability of complication risk based on the multi-dimensional feature vector and body shape parameters, obtaining the multi-event risk probability within multiple time windows; based on the multi-event risk probability within multiple time windows, the multi-dimensional feature vector, and the body shape parameters, it performs decision analysis to generate decision support information, enabling the early acquisition of multi-time window risk probability information related to the patient's condition and achieving dynamic tracking of the user's status; on this basis, it combines risk probability and feature information for decision analysis to generate structured decision support information that is interpretable and operable, forming an automated processing flow from data collection to information output. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A schematic diagram of the structure of a clinical auxiliary decision-making information generation system based on multimodal parameters provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a method for generating clinical auxiliary decision-making information based on multimodal parameters, provided in an embodiment of the present invention; Figure 3 A flowchart illustrating another method for generating clinical auxiliary decision-making information based on multimodal parameters provided in this embodiment of the invention; Figure 4 A flowchart for collecting body shape features and physiological parameter sets is provided in an embodiment of the present invention; Figure 5 A flowchart for extracting multi-dimensional features is provided in an embodiment of the present invention; Figure 6 A flowchart for extracting key feature information is provided in an embodiment of the present invention; Figure 7 A schematic diagram of an intervention recommendation report provided in an embodiment of the present invention; Figure 8 A schematic diagram of a clinical auxiliary decision-making information generation device based on multimodal parameters provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] It should be noted that the clinical auxiliary decision-making information generation method and device based on multimodal parameters disclosed in this application can be used in the field of artificial intelligence technology, or in any field other than artificial intelligence technology. The application field of the clinical auxiliary decision-making information generation method and device based on multimodal parameters disclosed in this application is not limited.

[0031] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution will be explained below. By collecting the body shape parameters and physiological parameter set including biomarker detection values, vital sign data, and blood test data of the first user (patient), multi-source heterogeneous data are integrated into a unified timeline using spatiotemporal alignment technology. Then, through feature extraction and machine learning model calculations, multi-time-window risk probability information reflecting the user's state change trend is output, ultimately generating structured auxiliary decision-making information. The entire processing flow is automatically executed by a computer system. The generated risk probability information is an intermediate result, used to provide a reference for the second user (clinical staff), rather than directly drawing a final conclusion.

[0032] Figure 1 This is a schematic diagram of the structure of a clinical auxiliary decision-making information generation system based on multimodal parameters provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the system includes: bedside monitoring equipment 100, hospital information system 200, bedside testing equipment 300, historical user database 400, knowledge engine database 500, computer equipment 600, first user database 700, multiple data interfaces, and second user database 900. The multiple data interfaces include: first data interface 801, second data interface 802, third data interface 803, and fourth data interface 804.

[0033] Computer device 600 is used to perform, for example Figure 2 or Figure 3The method for generating clinical decision support information based on multimodal parameters is shown. In the specific processing, the computer device 600 first performs spatiotemporal alignment and feature extraction on the collected physiological parameter set to generate a multi-dimensional feature vector. Using a pre-constructed risk prediction model, it calculates risk probabilities based on the multi-dimensional feature vector and body parameter features, outputting the risk probabilities of multiple events within multiple time windows. Based on the calculation results, it performs decision analysis to generate structured decision support information. After processing, the computer device 600 outputs the generated decision support information to a display device or clinical terminal for reference by a second user (clinical staff). The physiological parameter set includes vital sign parameters, laboratory test parameters, and biomarker parameters.

[0034] The first user database 700 is used to store the first user's multimodal data, which includes body parameter features and physiological parameter sets. During data processing, the computer device 600 stores the collected data and the intermediate results related to the first user (such as multi-dimensional feature vectors, risk probabilities, and decision support information) into the first user database 700, forming a structured data storage system for subsequent queries or model updates.

[0035] Multiple data interfaces establish communication connections with external data sources for data transmission.

[0036] In this embodiment of the invention, a first data interface 801 is disposed on the bedside monitoring device 100 and is used to transmit the vital signs parameters detected by the bedside monitoring device 100 to the computer device 600. The bedside monitoring device 100 is used to collect the vital signs parameters of the first user, including but not limited to time-series data of vital signs such as heart rate, blood pressure, respiratory rate, and blood oxygen saturation.

[0037] The second data interface 802 is installed on the hospital information system 200 and is used to transmit the test parameters recorded by the hospital information system 200 to the computer device 600. The hospital information system 200 is used to record the test parameters of the first user, including but not limited to routine laboratory test results such as complete blood count, liver function, and coagulation function.

[0038] The third data interface 803 is located in the bedside testing device 300 and is used to transmit the biomarker parameters detected by the bedside testing device 300 to the computer device 600. The bedside testing device 300 is used to detect biomarker parameters of the first user, including but not limited to intestinal barrier function biomarkers, immune status biomarkers, and endothelial injury biomarkers.

[0039] The fourth data interface 804 is located on the computer device 600 and is used to communicate with the first data interface 801, the second data interface 802 and the third data interface 803, and to transmit vital signs parameters, test parameters and marker parameters from the computer device 600 to the first user database 700 for storage.

[0040] In this embodiment of the invention, the historical user database 400 is used to store the full multimodal data of historical users, including the physical parameter features, physiological parameter sets, and corresponding clinical outcome information of historical users. When constructing a risk prediction model, the computer device 600 retrieves historical data from the historical user database 400 as training samples through the fourth data interface 804, and performs model training and parameter optimization through machine learning algorithms to generate a pre-constructed risk prediction model.

[0041] In this embodiment of the invention, the knowledge engine database 500 is used to store the clinical knowledge engine, which includes the mapping relationship between decision support information, key feature information, and risk level information. When generating decision support information, the computer device 600 uses the risk level information calculated from the risk probability and the key feature information extracted from the feature extraction to call the mapping relationship in the knowledge engine database 500 through the fourth data interface 804 for matching, thereby generating structured decision support information corresponding to the current user state.

[0042] In this embodiment of the invention, the second user database 900 is used to store the behavioral data and feedback data of the second user. The second user is a clinical staff member using this system, and their behavioral data includes, but is not limited to, records of viewing decision support information and response time, while the feedback data includes, but is not limited to, evaluation information on the accuracy of prediction results. The computer device 600 calls the feedback data accumulated in the second user database 900 through the fourth data interface 804 to perform incremental learning and continuous optimization of the risk prediction model, update model parameters, and improve the accuracy of subsequent processing.

[0043] The clinical decision support information generation system based on multimodal parameters provided by this invention forms a complete data flow path from data acquisition, storage, processing to information output. The components communicate with each other through data interfaces and buses to ensure real-time transmission and collaborative processing of multimodal data.

[0044] It is worth noting that, Figure 1 The clinical decision support information generation system based on multimodal parameters shown is also applicable to Figure 2 or Figure 3 The method for generating clinical auxiliary decision information based on multimodal parameters will not be elaborated here.

[0045] The following uses a clinical auxiliary decision-making information generation device based on multimodal parameters as an example to illustrate the implementation process of the clinical auxiliary decision-making information generation method based on multimodal parameters provided in this embodiment of the invention. It is understood that the execution entity of the clinical auxiliary decision-making information generation method based on multimodal parameters provided in this embodiment of the invention includes, but is not limited to, a clinical auxiliary decision-making information generation device based on multimodal parameters.

[0046] Figure 2 A flowchart of a method for generating clinical auxiliary decision-making information based on multimodal parameters provided in an embodiment of the present invention is shown below. Figure 2 As shown, the method includes: Step 101: Collect the body shape features and physiological parameter set of the first user.

[0047] In this embodiment of the invention, the first user is a patient.

[0048] In this embodiment of the invention, multiple data interfaces are connected to bedside monitoring equipment, hospital information system and bedside detection equipment respectively to collect multimodal data of the first user in real time.

[0049] In this embodiment of the invention, body shape parameters refer to the static attribute information of the first user, which is relatively stable over time and is used to characterize the user's basic state. Body shape parameters include the first user's static attribute information, such as age, etiology, and clinical score upon admission. The physiological parameter set refers to the first user's dynamic monitoring data, containing time-series variation characteristics, used to reflect real-time changes in the user's state. The physiological parameter set includes vital sign parameters, laboratory test parameters, and biomarker parameters.

[0050] Among them, vital signs parameters are collected through bedside monitoring equipment, including time-series data of vital signs such as heart rate, blood pressure, respiratory rate, and blood oxygen saturation; laboratory parameters are obtained through the hospital information system, including routine laboratory test results such as complete blood count, liver function, and coagulation function; biomarker parameters are detected through bedside testing equipment, including intestinal barrier function biomarkers, immune status biomarkers, and endothelial injury biomarkers.

[0051] Specifically, the process for detecting biomarker parameters using a specific biomarker panel is as follows: 1. Sample collection and processing procedure: The system collects biological samples from the first user via point-of-care testing equipment for biomarker parameter detection. For biomarkers requiring laboratory testing, 5 ml of peripheral venous blood is collected from the first user using a dedicated vacuum blood collection tube (containing EDTA anticoagulant). The collected blood is immediately placed in a 4°C cryogenic transport box, with the temperature monitored between 2°C and 8°C. The sample is delivered to the laboratory within 15 minutes of collection. Plasma separation employs a two-step centrifugation method: first, centrifugation at 2000g for 10 minutes to remove cellular components; second, centrifugation at 10000g for 10 minutes to remove particulate components. The separated plasma samples are aliquoted and stored in an ultra-low temperature freezer at -80°C, with temperature fluctuations controlled within ±2°C, until analysis. For rapid point-of-care testing (LPS, I-FABP), 100 μl of capillary blood is collected from the first user's fingertip using the manufacturer's matching fingertip capillary blood collection device. The collected blood is then directly analyzed using the point-of-care testing equipment.

[0052] 2. Biomarker detection methods and quality control: The system uses both bedside and laboratory testing equipment to detect the parameters of the three types of biomarkers.

[0053] Detection of intestinal barrier function markers: LPS was detected using the Limulus amebocyte lysate (LAL) assay, with a detection limit of 0.01 EU / ml; I-FABP was detected using the ELISA assay, with an inter-batch coefficient of variation of less than 8%.

[0054] Immune status marker detection: HLA-DR expression rate was detected by flow cytometry using a dual-labeling gating strategy of CD14 and HLA-DR, with gating criteria following EuroFlow standards; IL-6 and IL-10 were detected by electrochemiluminescence immunoassay, with a detection sensitivity of 0.5 pg / ml.

[0055] Endothelial injury marker detection: Ang-2 was detected using microfluidic chip technology, with a single detection time of less than 15 minutes; vWF antigen was detected using immunoturbidimetry, with a calibration curve determination coefficient R. 2 Greater than 0.99.

[0056] Before daily testing, the system performs three levels of quality control: Level 1 is testing of the manufacturer's quality control products, with the test results allowed within ±2 standard deviations of the nominal value of the quality control products; Level 2 is testing of laboratory-made quality control products; Level 3 is repeated testing of clinical samples, with 5% of clinical samples randomly selected for parallel retesting.

[0057] 3. Detection frequency and timing analysis: The system collects biomarker parameters according to the preset detection frequency. The routine monitoring plan is as follows: on the first day of the first patient's admission, a full set of biomarker panel tests are performed every 6 hours; after the first patient's condition stabilizes, the tests are adjusted to twice a day, at 08:00 and 20:00.

[0058] The system has preset critical threshold values ​​for biomarkers, including LPS greater than 5 EU / ml. When the detected value of any biomarker exceeds the preset critical threshold, the system automatically triggers an immediate retest command, requiring the point-of-care testing equipment to complete a repeat test of the biomarker within 30 minutes.

[0059] The system performs time-series analysis on the detected biomarker parameters and uses a mixed-effects model to calculate the following parameters: absolute concentration value of each biomarker, rate of change (change per unit time, calculated as Δ / hour), and diurnal fluctuation amplitude (amplitude parameter calculated using a cosine fitting method). All calculated parameters are incorporated into a multi-dimensional feature vector and used as input to the subsequent risk prediction model.

[0060] This invention enables specific monitoring of key physiological processes in the first user. Validation data show that the biomarker panel detection of this invention can detect abnormal changes 24 hours before significant changes occur in traditional indicators, with a sensitivity of 92% and a specificity of 88%. Among them, the combined elevation of LPS and I-FABP has the highest indicative value for gut microbiota translocation-related states, with an area under the curve of 0.91; a persistently low HLA-DR expression rate below 30% has a high indicative value for immune paralysis-related states, with a positive predictive value of 89%.

[0061] Step 102: Perform spatiotemporal alignment and feature extraction on the physiological parameter set to obtain a multi-dimensional feature vector.

[0062] In this embodiment of the invention, spatiotemporal alignment is a process of integrating multi-source data from different sources, with different frequencies, and different timestamp formats into a standardized time-series data matrix through time-axis unification and spatial dimension correlation. The multi-dimensional feature vector is a quantized representation set generated through feature extraction, containing time-domain features, frequency-domain features, trend features, and correlation features, used to describe the dynamic change pattern of the first user's state.

[0063] In this embodiment of the invention, since the data in the physiological parameter set come from different devices and have different sampling frequencies and timestamp formats, all data are unified onto the same time reference axis through spatiotemporal alignment technology to form a standardized time series data matrix. The sliding window method is used to extract features from the aligned time series data, calculate the statistics, trends, fluctuation characteristics, and correlation characteristics between parameters within the time window, and finally generate a multi-dimensional feature vector.

[0064] This invention solves the problem of fragmentation in the time dimension of multi-source heterogeneous data through spatiotemporal alignment processing, forming a unified and continuous time-series data view, which provides a standardized data foundation for subsequent feature extraction; through feature extraction processing, the original time-series data is transformed into mathematical features that can quantify and characterize the state change pattern, thereby improving the dimensional richness and representational ability of data processing.

[0065] Step 103: Using a pre-built risk prediction model, predict the probability of complications based on multi-dimensional feature vectors and shape parameter features to obtain the probability of multiple events within multiple time windows.

[0066] In this embodiment of the invention, multiple time windows refer to multiple different prediction time ranges used to characterize the timeliness of risk occurrence. Multi-event risk probability refers to the numerical value of the likelihood of occurrence calculated for each of the various preset event types, expressed in probabilistic form.

[0067] Specifically, multi-dimensional feature vectors and shape parameter features are combined and concatenated to form the model input features. A pre-built risk prediction model is then invoked, which is trained using machine learning algorithms based on full multimodal data from a historical user database. The risk prediction model calculates the probability of the first user experiencing various preset events within multiple future time windows based on the model input features. These multiple time windows include short-term, medium-term, and long-term periods, and the multiple events include infection-related events and organ function-related events.

[0068] This invention achieves automated processing of high-dimensional, nonlinear, and time-dependent data through a pre-constructed risk prediction model. It can output risk probability information for multiple time windows and multiple event types, forming a continuous and dynamic quantitative description of the first user's state. Compared with single-point static judgment, it has higher information richness and timeliness.

[0069] Step 104: Perform decision analysis based on the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features to generate decision support information.

[0070] In this embodiment of the invention, the decision support information is structured information generated after systematic analysis and available for reference by clinical staff, including but not limited to risk level, key feature descriptions, and graded guidance suggestions.

[0071] Specifically, risk level information is determined based on the numerical value of risk probability, and key feature information that contributes the most to the current risk probability is extracted from multi-dimensional feature vectors; the knowledge engine database is called to match the risk level information and key feature information with the preset mapping relationship to obtain the corresponding decision support information.

[0072] This invention comprehensively analyzes risk probability with feature information and shape parameter features, and combines the mapping relationship of a knowledge engine to generate decision support information that is interpretable and operable. It clarifies the key features related to risk and provides clearly defined guidance content, thereby enhancing the practical value of the output information.

[0073] In the technical solution provided by this invention, the body shape parameter features and physiological parameter set of the first user are collected; the physiological parameter set is spatiotemporally aligned and features are extracted to obtain a multi-dimensional feature vector; through a pre-constructed risk prediction model, the probability of complication risk is predicted based on the multi-dimensional feature vector and body shape parameter features to obtain the multi-event risk probability within multiple time windows; decision analysis is performed based on the multi-event risk probability within multiple time windows, the multi-dimensional feature vector, and the body shape parameter features to generate decision support information, which can obtain multi-time window risk probability information related to the condition in advance and realize dynamic tracking of the user's status; on this basis, decision analysis is performed by combining risk probability and feature information to generate structured decision support information with interpretability and operability, forming an automated processing flow from data collection to information output.

[0074] Figure 3 A flowchart of another clinical auxiliary decision-making information generation method based on multimodal parameters provided in this embodiment of the invention is shown below. Figure 3 As shown, the method includes: Step 201: Collect the body shape features and physiological parameter set of the first user.

[0075] In this embodiment of the invention, the physiological parameter set includes vital sign parameters, test parameters, and biomarker parameters.

[0076] Figure 4 A flowchart for collecting body shape features and physiological parameter sets is provided as an embodiment of the present invention, such as... Figure 4 As shown, step 201 specifically includes: Step 2011: Query the shape parameter features using the first user identifier.

[0077] In this embodiment of the invention, the physical parameter features of the user are queried from the first user database based on the first user identifier. The first user identifier is an information field that uniquely identifies the user, such as a hospital number or medical record number. The first user database stores the user's static attribute information, including age, cause of illness, and clinical score upon admission. As an optional approach, the query operation is performed using a structured query language, and the query results are used as physical parameter features.

[0078] Step 2012: Collect vital signs parameters through the first data interface of the bedside monitoring device.

[0079] In this embodiment of the invention, the vital signs parameters are time-series data of vital signs collected by bedside monitoring equipment, including heart rate, blood pressure, respiratory rate, and blood oxygen saturation.

[0080] Specifically, a communication connection is established with the bedside monitoring equipment through the first data interface. The bedside monitoring equipment includes an electrocardiogram monitor and a central monitoring station, used to collect the user's vital signs data. Data packets sent by the bedside monitoring equipment are continuously received at a preset sampling frequency (not less than once per minute). The data packets contain values ​​for heart rate, systolic blood pressure, diastolic blood pressure, respiratory rate, blood oxygen saturation, collection timestamp, and user identifier; the above information is parsed from the data packets as vital sign parameters.

[0081] Step 2013: Collect test parameters through the second data interface of the hospital information system.

[0082] In this embodiment of the invention, the test parameters are routine laboratory test results obtained through the hospital information system, including complete blood count, liver function, and coagulation function.

[0083] Specifically, a communication connection is established with the hospital information system through a second data interface. Data requests are sent to the hospital information system according to preset trigger conditions (such as after the test results are reported) or according to preset cycles (such as daily timed events). The hospital information system returns the user's test data, including the test item names, values, units, sampling time, report time, and user identifier for blood routine, liver function, and coagulation function tests. The above information is parsed from the returned data and used as test parameters.

[0084] Step 2014: Collect marker parameters through the third data interface of the bedside detection device.

[0085] The biomarker parameters are the specific biomarker detection values ​​detected by point-of-care testing equipment, including intestinal barrier function markers (LPS, I-FABP), immune status markers (HLA-DR expression rate, IL-6 / IL-10 ratio), and endothelial injury markers (Ang-2, vWF).

[0086] Specifically, a communication connection is established with the point-of-care testing device via a third data interface. The point-of-care testing device is a portable fluorescence immunoassay analyzer or a microfluidic chip analyzer, used to detect various indicators on the specific biomarker panel. Testing instructions are sent to the point-of-care testing device according to a preset testing frequency (once every 6 hours on the first day of admission, adjusted to twice daily after the patient's condition stabilizes), or the device automatically uploads test results. The test result data packet contains the biomarker name (LPS, I-FABP, HLA-DR expression rate, IL-6, IL-10, Ang-2, vWF), the test value, units, testing time, and user identifier; this information is parsed from the data packet and used as biomarker parameters.

[0087] Step 202: Determine whether the set of physiological parameters meets the triggering conditions. If yes, proceed to step 203; otherwise, proceed to step 201.

[0088] In this embodiment of the invention, trigger conditions are determined for the collected physiological parameter set. As an optional scheme, the trigger conditions include any one of the following three situations: First, the newly acquired test results in the biomarker parameters have been updated; second, a preset abnormal warning appears in the vital signs parameters, such as: systolic blood pressure is below 90 mmHg in three consecutive monitoring data; third, a preset important indicator change appears in the test parameters, such as: the creatinine value increases by more than 50% compared to the previous test value.

[0089] When the set of physiological parameters meets any of the above triggering conditions, the generation of clinical auxiliary decision-making information is triggered, and step 203 is executed; when the set of physiological parameters does not meet any triggering conditions, the process returns to step 201 and continues to collect the body parameter features and set of physiological parameters of the first user.

[0090] This invention establishes a trigger condition judgment mechanism to initiate subsequent data processing only when the physiological parameter set undergoes significant changes or abnormal conditions occur. This avoids redundant processing of unchanged data, reduces invalid calculations in the system, and improves data processing efficiency.

[0091] Step 203: Perform data preprocessing on the physiological parameter set to obtain the preprocessed physiological parameter set.

[0092] In this embodiment of the invention, data preprocessing includes data cleaning and standardization. Data cleaning includes range verification, logical verification, and missing value handling. Standardization refers to converting data from different devices and in different formats into a unified encoding format and unit. For example, blood pressure units output by devices from different manufacturers are unified to millimeters of mercury, temperature units are unified to degrees Celsius, all time fields are converted to a unified time zone reference, and the time precision is unified to the millisecond level.

[0093] Range verification refers to eliminating data points that exceed the preset range based on the clinically reasonable value range. For example, the preset range includes: systolic blood pressure 30 to 250 mmHg, diastolic blood pressure 10 to 150 mmHg, heart rate 30 to 220 beats per minute, respiratory rate 8 to 40 breaths per minute, and blood oxygen saturation 60% to 100%.

[0094] Logical verification refers to checking the logical consistency between physiological parameters, including checking the correlation between respiratory rate and blood oxygen saturation.

[0095] Missing value handling refers to filling in missing points in time series data using cubic spline interpolation; for missing values ​​in discrete detection data, they are marked as missing.

[0096] This invention eliminates abnormal and logically contradictory data that may occur during the data collection process through data cleaning, ensuring the accuracy and reliability of the input data. Through standardization, it solves the problem of inconsistent formats among multi-source data, providing a consistent data foundation with unified units for subsequent spatiotemporal alignment and feature extraction.

[0097] Step 204: Using the admission time of the first user as the origin of the time axis, perform spatiotemporal alignment on the preprocessed physiological parameter set to obtain a time series data matrix.

[0098] Specifically, the admission time of the first user is read from the first user database and defined as the origin of the time axis; each data record in the preprocessed physiological parameter set, including vital signs, test parameters and biomarker parameters, is traversed and the original timestamp in each data record is extracted; the time difference between the original timestamp and the admission time of each data record is calculated in milliseconds and used as the coordinate value of the data record on a unified time axis.

[0099] For vital signs parameters, due to their high collection frequency (no less than once per minute), each vital signs parameter data point is directly mapped to a unified time axis according to the calculated time coordinate value. For laboratory parameters and biomarker parameters, due to their low collection frequency (1 to 4 times per day), each laboratory parameter and biomarker parameter data point is mapped to a unified time axis according to the calculated time coordinate value, and the values ​​remain unchanged between two adjacent detection points until the next detection point updates.

[0100] All data records are sorted in ascending order according to their time coordinate values, and a time-series data matrix is ​​constructed using the time coordinate as the row index and the data item name as the column index. The time-series data matrix is ​​a two-dimensional data table constructed using time as the row index and data item as the column index, used to store standardized time-series data after spatiotemporal alignment. Each row of the matrix corresponds to a time point, and each column corresponds to a data item (such as heart rate, systolic blood pressure, LPS concentration, etc.). The matrix cells store the value of that data item at that time point.

[0101] This invention achieves spatiotemporal alignment by using the admission time as the origin of the time axis. It unifies vital signs, test parameters, and biomarker parameters from different devices with different sampling frequencies and timestamp formats onto the same time base, forming a standardized time-series data matrix. This solves the problem of fragmentation of multi-source heterogeneous data in the time dimension and provides a unified data view for subsequent data slicing and feature extraction.

[0102] Step 205: Slice the time series data matrix according to the preset time window and sliding step size to generate time series data segments.

[0103] In this embodiment of the invention, a time window refers to a fixed time length used for slicing. The sliding step size refers to the sliding interval between two adjacent time windows. A time-series data segment refers to a sub-matrix obtained by slicing the time-series data matrix according to time windows, with each segment corresponding to complete time-series data within a time window.

[0104] Specifically, using the time axis of the time series data matrix as a reference, starting from the earliest time point, all rows of data within the time window are selected to form the first time series data segment; the time window is then slid forward according to the sliding step size, and all rows of data within the next time window are selected to form the second time series data segment. This sliding slicing operation is repeated until the entire time range of the time series data matrix has been traversed.

[0105] As an optional approach, the preset time window length is 6 hours, and the preset sliding step size is 1 hour. Specifically, using the time axis of the time series data matrix as a reference, starting from the earliest time point, all rows of data within a 6-hour time interval are selected to form the first time series data segment; the time window is then slid forward by 1 hour, and all rows of data within the next 6-hour time interval are selected to form the second time series data segment. This sliding slicing operation is repeated until the entire time range of the time series data matrix has been traversed.

[0106] In this embodiment of the invention, each time series data segment is a sub-matrix containing the values ​​of each data item at all time points within the time window; the generated series of time series data segments are stored in chronological order for use in subsequent feature extraction steps.

[0107] This invention discretizes a continuous time-series data matrix into multiple time-series data segments of fixed length through time window slicing, enabling feature extraction for each segment independently. At the same time, the overlapping between segments is achieved by setting a sliding step size, ensuring the ability to continuously track changes in the patient's state.

[0108] Step 206: Extract multi-dimensional features from the time-series data segments to generate multi-dimensional feature vectors.

[0109] In this embodiment of the invention, multi-dimensional feature extraction includes time-domain feature extraction, frequency-domain feature extraction, and correlation feature extraction; all extracted features are combined to generate a multi-dimensional feature vector.

[0110] Figure 5 A flowchart for extracting multi-dimensional features is provided as an embodiment of the present invention, such as... Figure 5 As shown, step 206 specifically includes: Step 2061: Calculate the temporal characteristics of the time-series data segment.

[0111] In this embodiment of the invention, for each time-series data segment, the following time-domain features are calculated for each data item (such as heart rate, systolic blood pressure, LPS concentration, etc.): Calculate the arithmetic mean, standard deviation, maximum value, minimum value, median, and interquartile range of the data column within the time window; calculate the slope of the linear trend of the data column within the time window by fitting a linear regression equation using the least squares method; calculate the rate of change of the data column within the time window by dividing the window duration by the sum of the last value of the window and the first value of the window; calculate the coefficient of variation of the data column by dividing the standard deviation by the mean; calculate the approximate entropy of the data column by using the approximate entropy algorithm to calculate the sequence complexity.

[0112] In this embodiment of the invention, the above-mentioned time-domain features are calculated for all data items to obtain a set of time-domain features.

[0113] Step 2062: Extract frequency domain features from time-series data segments using wavelet transform.

[0114] In this embodiment of the invention, for each time-series data segment, frequency domain features are extracted using wavelet transform for each data item within it.

[0115] As an alternative approach, wavelet basis functions are used to perform three-level wavelet decomposition on the data sequence to obtain approximation coefficients and detail coefficients. The energy proportion of each frequency band is calculated based on the decomposition results, which is the ratio of the sum of squares of the energy of each level of detail coefficients to the total sum of squares of energy. The energy proportion of each frequency band is then used as a frequency domain feature.

[0116] In this embodiment of the invention, the above-mentioned frequency domain features are calculated for all data items to obtain a frequency domain feature set.

[0117] Step 2063: Calculate the correlation characteristics between various parameters in the time series data segment.

[0118] In this embodiment of the invention, for each time-series data segment, the correlation features between different data items are calculated.

[0119] As an optional approach, Pearson correlation coefficients are calculated among the vital signs parameters, including the correlation coefficients between heart rate and systolic blood pressure, and heart rate and respiratory rate; Pearson correlation coefficients are calculated among the biomarker parameters, including the correlation coefficients between LPS and I-FABP, and the correlation coefficients between HLA-DR expression rate and the IL-6 / IL-10 ratio; and mutual information values ​​are calculated between vital signs parameters and biomarker parameters to measure the nonlinear correlation between the two types of parameters.

[0120] In this embodiment of the invention, all the calculated correlation coefficients and mutual information values ​​are used as a set of associated features.

[0121] Step 2064: Generate a multi-dimensional feature vector based on time-domain features, frequency-domain features, and correlation features.

[0122] In this embodiment of the invention, the time-domain feature is a feature obtained by statistical calculation of the data sequence in the time domain; the frequency-domain feature is a feature extracted in the frequency domain through wavelet transform; the correlation feature is a quantitative indicator of the correlation between different data items; the multi-dimensional feature vector is a feature set composed of time-domain features, frequency-domain features and correlation features, represented in vector form, and is used to quantitatively describe the user state features corresponding to the time-series data segment.

[0123] Specifically, the time-domain features, frequency-domain features, and correlation features are concatenated to form a multi-dimensional feature vector for this time-series data segment. Each element in the multi-dimensional feature vector corresponds to a specific feature value.

[0124] This invention transforms raw time-series data fragments into mathematical feature vectors that quantify user state change patterns through multi-dimensional feature extraction. Time-domain features reflect the statistical distribution and trends of the data, frequency-domain features reflect the periodic fluctuations, and correlation features reflect the synergistic relationships between various parameters. The combination of these three features enables the multi-dimensional feature vectors to comprehensively describe the multifaceted characteristics of user states, providing rich input information for subsequent risk prediction models.

[0125] Step 207: Using a pre-built risk prediction model, predict the probability of complications based on multi-dimensional feature vectors and shape parameter features to obtain the probability of multiple events within multiple time windows.

[0126] Specifically, the risk prediction model is invoked, and multi-dimensional feature vectors and shape parameter features are concatenated to form a complete model input vector, which is then input into the risk prediction model. The risk prediction model calculates based on the input vector and outputs the probability of the first user experiencing various preset events within multiple future time windows.

[0127] As an optional approach, three time windows are preset: the next 6 hours, the next 24 hours, and the next 48 hours. Multiple event types are preset, including but not limited to sepsis, hepatorenal syndrome, and hepatic encephalopathy. The risk prediction model outputs a risk probability value for each combination of time window and event type, with a value ranging from 0 to 1. All risk probability values ​​constitute the multi-event risk probability within the multiple time windows.

[0128] In this embodiment of the invention, the process of constructing a risk prediction model is as follows: obtaining full multimodal parameters of multiple historical users; training a model on a preset hybrid deep learning architecture based on the full multimodal parameters using a weighted cross-entropy loss function to construct an initial prediction model, wherein the hybrid deep learning architecture is a coupled temporal network and a graph attention network; and performing cross-validation optimization on the initial prediction model to construct a risk prediction model.

[0129] The full set of multimodal parameters for historical users includes physical parameters, physiological parameter sets, and corresponding clinical outcome information. A labeling team composed of three associate chief physicians accurately labeled the event occurrence times in the clinical outcome information, and the labeling consistency was verified by the Kappa coefficient test (Kappa value greater than 0.85).

[0130] The temporal network used is the InceptionTime temporal network, with a depth of 15 layers, used to extract temporal dependencies between multi-dimensional feature vectors; the Graph Attention network is used to handle the associations between various biomarkers. In the weighted cross-entropy loss function, higher weight values ​​are set for event types with fewer samples in the training set. As an optional approach, the weight for sepsis events is set to 1.5, and the weight for hepatic encephalopathy events is set to 1.2. The training parameters are set as follows: initial learning rate 0.001, batch size 32, and early stopping patience value of 20 training epochs.

[0131] The initial prediction model was optimized using nested cross-validation, employing a nested cross-validation method with an outer 5-fold cross-validation layer and an inner 3-fold cross-validation layer. The area under the curve (AUC) was used as the primary evaluation metric, while sensitivity, specificity, and clinical usefulness index (CUI = sensitivity × PPV - (1 - specificity) × (1 - negative predictive value)) were used as auxiliary evaluation metrics. The final risk prediction model was selected based on the parameter configuration that yielded the highest AUC on the validation set, a sensitivity greater than 85%, a specificity greater than 80%, and a clinical usefulness index greater than a preset threshold.

[0132] Furthermore, when the AUC decreases by more than 3% for 30 consecutive days, the risk prediction model is retrained.

[0133] This invention achieves automated calculation and processing of multi-dimensional feature vectors and shape parameter features through a pre-constructed risk prediction model, and can output risk probability information for multiple time windows and multiple event types. The nested cross-validation optimization method ensures the generalization ability of the model parameter configuration, and the weighted cross-entropy loss function solves the problem of unbalanced sample distribution in the training data, improving the model's prediction accuracy for a few event types.

[0134] Step 208: Perform interpretability testing based on the multi-event risk probability, multi-dimensional feature vector, and shape parameter features within multiple time windows, and extract key feature information.

[0135] In this embodiment of the invention, the risk probabilities of multiple events within multiple time windows, multi-dimensional feature vectors, and shape parameter features are used as inputs for interpretability detection. The key feature information that contributes the most to the current risk probability is extracted from the multi-dimensional feature vectors and shape parameter features.

[0136] Figure 6 A flowchart for extracting key feature information is provided as an embodiment of the present invention, such as... Figure 6 As shown, step 208 specifically includes: Step 2081: Using the interpretability detection method, based on the risk probability of multiple events within multiple time windows, the contribution of multi-dimensional feature vectors and shape parameter features is detected to obtain feature contribution information.

[0137] In this embodiment of the invention, the SHAP interpretability detection method is employed to detect the contribution of multi-dimensional feature vectors and shape parameter features based on the risk probabilities of multiple events within multiple time windows. The SHAP method is based on the Shapley value theory in game theory, calculating the marginal contribution of each input feature to the model output.

[0138] Specifically, for each risk probability value (i.e., the combination of each time window and each event type), the SHAP value of each feature in the input vector is calculated. The input vector includes all elements of the multi-dimensional feature vector and all elements of the shape parameter features. A positive SHAP value indicates that the feature contributes positively to the risk probability, meaning that the presence of the feature increases the risk probability value; a negative SHAP value indicates that the feature contributes negatively to the risk probability, meaning that the presence of the feature decreases the risk probability value. The SHAP values ​​of all features are used as feature contribution information.

[0139] Step 2082: Based on the feature contribution information, filter out key features and determine the corresponding key feature information.

[0140] In this embodiment of the invention, interpretability testing refers to the process of performing attribution analysis on the output of the risk prediction model, calculating the contribution of each feature in the input vector to the output. Key feature information is the set of information corresponding to the feature with the highest contribution selected from the input vector, including the feature name, current value, trend, and SHAP contribution value.

[0141] Specifically, based on feature contribution information, all features in the input vector are sorted in descending order of their absolute SHAP values. The larger the absolute SHAP value, the greater the contribution of that feature to the current risk probability. The top K features in the sorting result are selected as key features; for example, K is preset to 5. For each selected key feature, its corresponding numerical value is extracted from the multi-dimensional feature vector or shape parameter features. Simultaneously, the trend information of this feature's change within a preset time window is extracted from the time-series data matrix, including the direction and magnitude of the change.

[0142] The key feature information is formed by combining the name, current value, trend of change, and SHAP contribution value of the key feature.

[0143] This invention uses an interpretability detection method to detect the contribution of the output of a risk prediction model, quantifying the contribution of each input feature to the prediction result and making the model output interpretable. By screening key features and extracting their corresponding values ​​and trends, the feature information with the highest correlation to risk probability is identified, providing an attribution basis for the subsequent generation of structured decision support information.

[0144] Step 209: Generate risk level information based on the risk probabilities of multiple events within multiple time windows.

[0145] Specifically, based on a preset risk level threshold, the risk level information for each event within each time window is generated by comparing the risk probabilities of multiple events within multiple time windows.

[0146] As an optional approach, the risk level thresholds include a first threshold and a second threshold, with the first threshold set to 0.7 and the second threshold set to 0.3. Each risk probability value is compared to a preset threshold: when the risk probability value is greater than or equal to the first threshold of 0.7, it is classified as high-risk; when the risk probability value is less than the first threshold of 0.7 but greater than or equal to the second threshold of 0.3, it is classified as medium-risk; and when the risk probability value is less than the second threshold of 0.3, it is classified as low-risk. For each combination of time window (next 6 hours, next 24 hours, next 48 hours) and event type (sepsis, hepatorenal syndrome, hepatic encephalopathy), corresponding risk level information is generated.

[0147] Furthermore, for risks deemed high-risk, the system performs stability verification. Specifically, if the risk level information is high-risk, the system repeatedly executes the steps of predicting the probability of complications based on multi-dimensional feature vectors and shape parameter features using a pre-built risk prediction model, obtaining the risk probabilities of multiple events within multiple time windows, and generating the risk level information for each event within each time window based on the risk probabilities of multiple events within multiple time windows.

[0148] The specified number of predictions is preset to 3. The multi-dimensional feature vectors corresponding to the three most recent time windows are used for repeated predictions to obtain the multi-event risk probabilities within multiple time windows. If the average of the multi-event risk probabilities within multiple time windows is greater than a first threshold, it is determined to be a high-risk level; if the average of the multi-event risk probabilities within multiple time windows is less than the first threshold but greater than or equal to a second threshold, it is determined to be a medium-risk level; if the average of the multi-event risk probabilities within multiple time windows is less than the second threshold, it is determined to be a low-risk level.

[0149] This invention classifies risk probabilities by setting a preset risk level threshold, transforming continuous probability values ​​into discrete risk level information, which facilitates matching and retrieval in subsequent decision analysis. By repeatedly predicting and verifying the stability of high-risk predictions, it can identify accidental high-risk results caused by data fluctuations or model uncertainties, thus improving the reliability of high-risk level information.

[0150] Step 210: Generate decision support information based on key feature information and risk level information through a preset clinical knowledge engine.

[0151] In this embodiment of the invention, the clinical knowledge engine is a set of rules stored in the knowledge engine database, which includes the mapping relationship between decision support information, key feature information combination and risk level information. The construction of the mapping relationship is based on the clinical guideline knowledge base.

[0152] In this embodiment of the invention, a clinical knowledge engine is invoked, with key feature information and risk level information as input. The mapping relationship in the clinical knowledge engine is used for matching to generate structured decision support information.

[0153] Specifically, key feature information (including feature name, current value, and trend) is combined with risk level information and matched and retrieved in the clinical knowledge engine. Upon successful matching, the corresponding decision support information is output.

[0154] In this embodiment of the invention, decision support information is organized hierarchically according to the level of evidence, including three levels: immediate action items, suggested consideration items, and evaluation options. Immediate action items correspond to recommendations at evidence level IA, such as collecting blood cultures or performing abdominal imaging examinations; suggested consideration items correspond to recommendations at evidence level IIB, such as requesting consultation with the intensive care unit and the infectious disease department; and evaluation options correspond to expert consensus recommendations, such as assessing the feasibility of using immunomodulatory therapy.

[0155] The following is a specific example of decision support information generation: When the risk level is high and the extracted key features include decreased HLA-DR expression in monocytes and increased plasma LPS concentration, the clinical knowledge engine will output the following decision support information: Warning information (risk level and predicted event): "High risk - sepsis"; Mechanism analysis: "Dominated by immune paralysis, accompanied by intestinal barrier damage"; The grading recommendations include: Immediate action: Blood culture and pathogen metagenomic sequencing; Recommended consideration: empirical use of antibiotics covering intestinal Gram-negative bacteria; Evaluation options: Request an infectious disease consultation to evaluate immunomodulatory therapy (such as GM-CSF).

[0156] The system presents the generated decision support information through a visual interface, including warning indicators, mechanism analysis content, and a list of graded recommendations, and pushes it in real time to the terminal devices used by clinical staff for their reference.

[0157] It is worth noting that when generating decision support information, the system automatically retrieves the first user's medication allergy history and filters or adds warning labels to suggestions containing contraindicated drugs. For example, β-lactam antibiotics are excluded for those allergic to penicillin.

[0158] Furthermore, the full set of multimodal parameters of the first user, the behavioral data and feedback data of the second user are obtained; based on the full set of multimodal parameters, behavioral data and feedback data, the risk prediction model is incrementally learned to generate an updated risk prediction model.

[0159] Specifically, the system acquires full multimodal parameters from the first user, including collected body shape features and physiological parameter sets, as well as all data generated during subsequent monitoring. It also acquires behavioral and feedback data from the second user, a clinical staff member using the system. Behavioral data includes records of viewing decision support information and response times. Feedback data includes evaluation information on the accuracy of the prediction results (e.g., true positive, false positive). A true positive indicates that the second user confirms the warning information matches the actual situation; a false positive indicates that the second user confirms the warning information does not match the actual situation.

[0160] Incremental learning of the risk prediction model is performed based on full multimodal parameters, behavioral data, and feedback data. When the accumulated new data reaches 100 complete case course data, the model update process is automatically triggered. The new data is merged with historical data in the historical user database to form an updated training set. Samples marked as true positives in the feedback data are weighted during training to strengthen the learning weight of high-value samples. An incremental learning framework is used to retrain the original risk prediction model and update the model parameters. The performance metrics of the new model are evaluated using a retained validation set. When the area under the curve of the new model on the validation set is not lower than that of the old model, the new model is deployed online, replacing the old model. The five most recent model versions are retained for rollback purposes.

[0161] This invention uses a pre-set clinical knowledge engine to match key feature information and risk level information with decision support information, realizing the transformation from risk probability output to structured intervention guidance information. The output information is graded according to the level of evidence, clarifying the priority of each recommendation. Decision support information is pushed in real time through a visual interface, facilitating quick access for clinical staff. Through an incremental learning mechanism, the risk prediction model can be continuously optimized based on newly added full-scale multimodal parameters, user behavior data, and feedback data, shortening the model update cycle and maintaining stable performance.

[0162] Furthermore, the generated decision support information is presented through a visual interface, providing the second user (clinical staff) with a risk warning dashboard, a patient details view, and intervention suggestion reports.

[0163] The risk warning dashboard uses color coding to indicate risk levels: green for low risk, yellow for medium risk, and red for high risk. The dashboard supports parallel monitoring views for multiple patients, allowing the simultaneous display of risk status for multiple primary users.

[0164] The patient details view includes a dynamic trend chart showing the changes in key feature information over 72 hours; a risk evolution timeline showing the changes in risk level in chronological order; and an outlier highlighting area, using box plots to annotate outliers that exceed a preset range.

[0165] The intervention recommendation report presents the generated decision support information in a structured text format, including warning indicators, mechanism analysis content, and a tiered recommendation list.

[0166] The system triggers different levels of early warning pushes based on risk level information. When the risk level information is high, the system pushes early warnings through three channels: the first channel is the central screen in the ICU, which displays the early warning information in a flashing red light; the second channel is the mobile terminal of a second user, which pushes the early warning information in the form of a pop-up window, which can only be closed after the second user clicks to confirm; the third channel is the hospital's early warning system, which sends an early warning notification in the form of an SMS message, which contains brief key information.

[0167] After the second user clicks on the alert information, they can view a complete view of the patient's details and an intervention recommendation report. The system automatically records the alert response time, which is the time difference between the time the alert information is pushed and the time the second user first views it. This time difference is stored in the second user's database as a quality evaluation indicator.

[0168] After clinical intervention, the second user must provide feedback on the accuracy of the early warning in the system. Feedback options include true positive and false positive. The system stores this feedback information in the second user's database as a basis for subsequent model optimization.

[0169] This invention employs a three-channel early warning push mechanism to ensure that high-risk information reaches clinical staff promptly. The combination of red flashing, pop-up confirmation, and SMS notification enhances the reliability and timeliness of information delivery. By recording the early warning response time, the system achieves a quantitative assessment of information delivery efficiency. By collecting feedback from clinical staff, the system can obtain true positive and false positive labeling information, which is used for sample-weighted training in subsequent model optimization. This allows the risk prediction model to continuously improve based on real clinical feedback, enhancing the model's accuracy in practical applications.

[0170] Figure 7 This is a schematic diagram of an intervention recommendation report provided in an embodiment of the present invention, such as... Figure 7As shown, this is an intervention recommendation report for patients at high risk of sepsis, presented in a structured table format. It is used to quickly communicate patient risk, mechanisms, and intervention plans to healthcare professionals. The medical alert form includes the patient (name of the primary user), user ID, risk level, predicted events with a time window, risk probability, immune typing, key characteristic information, and clinical intervention recommendations. For detailed implementation examples, please refer to [link to specific examples]. Figure 7 This will not be elaborated upon here.

[0171] It is worth noting that all data involved in this invention, including but not limited to body shape parameters and physiological parameter sets, are intermediate results generated during the monitoring process. Their direct purpose is solely to provide reference data for clinicians' decision-making. The aforementioned data itself does not constitute a direct judgment of a patient's condition or health status. For example, predicting an 85% probability that a patient will develop sepsis within the next 24 hours does not necessarily mean that the patient will develop sepsis. A comprehensive analysis and final diagnosis based on expert knowledge and clinical experience, combined with other test results, imaging reports, and the patient's clinical manifestations, is still required. This technical solution aims to provide an auxiliary monitoring and advisory tool, rather than an independent diagnostic device.

[0172] It is worth noting that the acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. The user information in the embodiments of this application was obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the client.

[0173] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0174] It is worth noting that the technical solution provided in this application provides users with a corresponding operation entry point, allowing users to choose to agree to or reject the automated decision-making result; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0175] The technical solution of the clinical auxiliary decision-making information generation method based on multimodal parameters provided in this invention involves collecting the body parameter features and physiological parameter set of a first user; performing spatiotemporal alignment and feature extraction on the physiological parameter set to obtain a multi-dimensional feature vector; using a pre-constructed risk prediction model, predicting the probability of complication risks based on the multi-dimensional feature vector and body parameter features to obtain the probability of multiple events within multiple time windows; performing decision analysis based on the probability of multiple events within multiple time windows, the multi-dimensional feature vector, and the body parameter features to generate decision support information, enabling the early acquisition of risk probability information related to the patient's condition within multiple time windows and achieving dynamic tracking of the user's status; and combining risk probability and feature information for decision analysis to generate structured decision support information that is interpretable and operable, forming an automated processing flow from data acquisition to information output.

[0176] Figure 8 This is a schematic diagram of a clinical auxiliary decision-making information generation device based on multimodal parameters provided in an embodiment of the present invention. This device is used to execute the aforementioned clinical auxiliary decision-making information generation method based on multimodal parameters, such as... Figure 8 As shown, the device includes: a data acquisition unit 11, a data processing unit 12, a risk probability prediction unit 13, and a decision analysis unit 14.

[0177] The acquisition unit 11 is used to acquire the body shape features and physiological parameter set of the first user.

[0178] The data processing unit 12 is used to perform spatiotemporal alignment and feature extraction on the physiological parameter set to obtain a multi-dimensional feature vector.

[0179] The risk probability prediction unit 13 is used to predict the risk probability of complications based on multi-dimensional feature vectors and shape parameter features through a pre-built risk prediction model, and obtain the risk probability of multiple events within multiple time windows.

[0180] The decision analysis unit 14 is used to perform decision analysis based on the risk probabilities of multiple events, multi-dimensional feature vectors, and shape parameter features within multiple time windows, and generate decision support information.

[0181] In this embodiment of the invention, the physiological parameter set includes vital sign parameters, test parameters, and biomarker parameters; the acquisition unit 11 is specifically used to query body parameter characteristics through a first user identifier; acquire vital sign parameters through a first data interface of the bedside monitoring device; acquire test parameters through a second data interface of the hospital information system; and acquire biomarker parameters through a third data interface of the bedside testing device.

[0182] In this embodiment of the invention, the data processing unit 12 is specifically used to preprocess the physiological parameter set to obtain a preprocessed physiological parameter set. The data preprocessing includes data cleaning and standardization; taking the admission time of the first user as the origin of the reference time axis, the preprocessed physiological parameter set is spatiotemporally aligned to obtain a time series data matrix; the time series data matrix is ​​sliced ​​according to a preset time window and sliding step size to generate time series data segments; and multi-dimensional feature extraction is performed on the time series data segments to generate multi-dimensional feature vectors.

[0183] In this embodiment of the invention, the data processing unit 12 is specifically used to calculate the time-domain features in the time-series data segment; extract the frequency-domain features in the time-series data segment through wavelet transform; calculate the correlation features between various parameters in the time-series data segment; and generate a multi-dimensional feature vector based on the time-domain features, frequency-domain features, and correlation features.

[0184] In this embodiment of the invention, the device further includes: a historical data acquisition unit 15, a model building unit 16, and a model optimization unit 17.

[0185] The historical data acquisition unit 15 is used to acquire the full multimodal parameters of multiple historical users.

[0186] The model building unit 16 is used to train the model on the preset hybrid deep learning architecture based on the full set of multimodal parameters using the weighted cross-entropy loss function, and to build the initial prediction model. The hybrid deep learning architecture consists of a coupled temporal network and a graph attention network.

[0187] The model optimization unit 17 is used to perform cross-validation optimization on the initial prediction model to construct a risk prediction model.

[0188] In this embodiment of the invention, the decision analysis unit 14 is specifically used to perform interpretability detection based on the multi-event risk probability, multi-dimensional feature vector, and shape parameter features within multiple time windows, and extract key feature information; generate risk level information based on the multi-event risk probability within multiple time windows; and generate decision support information based on the key feature information and risk level information through a preset clinical knowledge engine.

[0189] In this embodiment of the invention, the decision analysis unit 14 is specifically used to perform contribution detection on multi-dimensional feature vectors and shape parameter features based on the risk probability of multiple events within multiple time windows using an interpretability detection method, thereby obtaining feature contribution information; and to filter out key features and determine the corresponding key feature information based on the feature contribution information.

[0190] In this embodiment of the invention, the decision analysis unit 14 is specifically used to generate risk level information for each event in each time window by comparing the risk probabilities of multiple events within multiple time windows based on a preset risk level threshold.

[0191] In this embodiment of the invention, the device further includes a stability verification unit 18 and a risk level determination unit 19.

[0192] In this embodiment of the invention, the stability verification unit 18 is used to repeatedly execute the steps of predicting the probability of complications based on multi-dimensional feature vectors and shape parameter features by using a pre-built risk prediction model to predict the probability of multiple events within multiple time windows if the risk level information is high risk, thereby obtaining the probability of multiple events within multiple time windows.

[0193] The risk level determination unit 19 is used to generate risk level information for each event in each time window based on the risk probabilities of multiple events in multiple time windows.

[0194] In this embodiment of the invention, the device further includes an incremental data acquisition unit 20 and an incremental learning unit 21.

[0195] The incremental data acquisition unit 20 is used to acquire the full multimodal parameters of the first user, the behavioral data and feedback data of the second user.

[0196] The incremental learning unit 21 is used to perform incremental learning on the risk prediction model based on the full set of multimodal parameters, behavioral data, and feedback data, and generate an updated risk prediction model.

[0197] In this embodiment of the invention, the body shape parameters and physiological parameter set of the first user are collected; the physiological parameter set is spatiotemporally aligned and features are extracted to obtain a multi-dimensional feature vector; through a pre-constructed risk prediction model, the probability of complication risk is predicted based on the multi-dimensional feature vector and body shape parameters, resulting in the probability of multiple events within multiple time windows; decision analysis is performed based on the probability of multiple events within multiple time windows, the multi-dimensional feature vector, and the body shape parameters to generate decision support information, enabling the early acquisition of risk probability information related to the condition within multiple time windows and achieving dynamic tracking of the user's status; based on this, decision analysis is performed by combining risk probability and feature information to generate structured decision support information that is interpretable and operable, forming an automated processing flow from data collection to information output.

[0198] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device, specifically, a computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0199] This invention provides a computer device, including a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement the steps of the above-described embodiment of the clinical auxiliary decision-making information generation method based on multimodal parameters. For a detailed description, please refer to the above-described embodiment of the clinical auxiliary decision-making information generation method based on multimodal parameters.

[0200] The following is for reference. Figure 9 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing the embodiments of this application.

[0201] like Figure 9 As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer device 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0202] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal feedback (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed in storage section 608 as needed.

[0203] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611.

[0204] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0205] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0206] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0207] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0208] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0209] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0210] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0211] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0212] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0213] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0214] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0215] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for generating clinical auxiliary decision-making information based on multimodal parameters, characterized in that, The method includes: Collect the body shape and physiological parameter set of the first user; Spatiotemporal alignment and feature extraction are performed on the physiological parameter set to obtain a multi-dimensional feature vector; By using a pre-built risk prediction model, the probability of complication risk is predicted based on the multi-dimensional feature vector and body parameter features, and the probability of multiple events within multiple time windows is obtained. Decision analysis is performed based on the risk probabilities of multiple events, multi-dimensional feature vectors, and shape parameter features within the multiple time windows to generate decision support information.

2. The clinical auxiliary decision-making information generation method based on multimodal parameters according to claim 1, characterized in that, The set of physiological parameters includes vital signs, test parameters, and biomarker parameters; The set of body shape and physiological parameters collected from the first user includes: Query the shape parameter features using the first user identifier; The vital signs parameters are collected through the first data interface of the bedside monitoring device; The test parameters are collected through the second data interface of the hospital information system; The marker parameters are collected through the third data interface of the bedside monitoring device.

3. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 1, characterized in that, The process of spatiotemporal alignment and feature extraction of the physiological parameter set to obtain a multi-dimensional feature vector includes: The physiological parameter set is preprocessed to obtain a preprocessed physiological parameter set. The data preprocessing includes data cleaning and standardization. Using the admission time of the first user as the origin of the time axis, the preprocessed physiological parameter set is spatiotemporally aligned to obtain a time series data matrix. The time series data matrix is ​​sliced ​​according to a preset time window and sliding step size to generate time series data segments; Multi-dimensional feature extraction is performed on the time-series data segment to generate a multi-dimensional feature vector.

4. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 3, characterized in that, The step of extracting multi-dimensional features from the time-series data segment to generate a multi-dimensional feature vector includes: Calculate the temporal features in the time-series data segment; Frequency domain features are extracted from the time-series data segment using wavelet transform. Calculate the correlation characteristics between various parameters in the time-series data segment; Based on the time-domain features, frequency-domain features, and correlation features, a multi-dimensional feature vector is generated.

5. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 1, characterized in that, The method further includes: Obtain full multimodal parameters from multiple historical users; The model is trained on the preset hybrid deep learning architecture using the weighted cross-entropy loss function based on the full set of multimodal parameters to construct an initial prediction model. The hybrid deep learning architecture consists of a coupled temporal network and a graph attention network. The initial prediction model is optimized by cross-validation to construct the risk prediction model.

6. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 1, characterized in that, The step of performing decision analysis based on the multi-event risk probabilities, multi-dimensional feature vectors, and shape parameter features within the multi-time window to generate decision support information includes: Interpretability detection is performed based on the multi-event risk probability, multi-dimensional feature vector, and shape parameter features within the multi-time window to extract key feature information; Risk level information is generated based on the risk probabilities of multiple events within the multiple time windows; Decision support information is generated based on the key feature information and risk level information through a preset clinical knowledge engine.

7. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 6, characterized in that, The process involves interpretability detection based on the multi-event risk probabilities, multi-dimensional feature vectors, and shape parameter features within the multi-time window, extracting key feature information, including: By using an interpretability detection method, the contribution of multi-dimensional feature vectors and shape parameter features is detected based on the risk probabilities of multiple events within the multiple time windows, thereby obtaining feature contribution information. Based on the feature contribution information, key features are selected and the corresponding key feature information is determined.

8. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 6, characterized in that, The step of generating risk level information based on the risk probabilities of multiple events within the multiple time windows includes: Based on a preset risk level threshold, the risk level information for each event within each time window is generated by comparing the risk probabilities of multiple events within the multiple time windows.

9. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 8, characterized in that, The method further includes: If the risk level information is high risk, the steps of repeating the pre-built risk prediction model a specified number of times, predicting the probability of complication risk based on the multi-dimensional feature vector and body parameter features, and obtaining the probability of multiple events within multiple time windows are repeated to obtain the probability of multiple events within multiple time windows. Based on the risk probabilities of multiple events within multiple time windows, risk level information for each event within each time window is generated.

10. The method for generating clinical auxiliary decision-making information based on multimodal parameters according to claim 1, characterized in that, The method further includes: Obtain all multimodal parameters of the first user, and behavioral and feedback data of the second user; Based on the full set of multimodal parameters, behavioral data, and feedback data, the risk prediction model is incrementally learned to generate an updated risk prediction model.

11. A clinical decision support information generation system based on multimodal parameters, characterized in that, The system applied to the clinical auxiliary decision information generation method based on multimodal parameters according to any one of claims 1 to 10, the system comprising: a computer device and a first user database and a plurality of data interfaces respectively connected to the computer device; The computer device is used to execute the clinical auxiliary decision information generation method based on multimodal parameters as described in any one of claims 1 to 10; The first user database is used to store the multimodal data of the first user, the multimodal data including body parameter features and physiological parameter sets; The data interface is used for data transmission.

12. The clinical auxiliary decision-making information generation system based on multimodal parameters according to claim 11, characterized in that, The multiple data interfaces include: a first data interface, a second data interface, a third data interface, and a fourth data interface; the physiological parameter set includes vital sign parameters, test parameters, and biomarker parameters; The system also includes: bedside monitoring equipment, a hospital information system, and bedside testing equipment; The first data interface is used to transmit the vital signs parameters detected by the bedside monitoring device to the computer device; The second data interface is used to transmit the test parameters recorded by the hospital information system to the computer device; The third data interface is used to transmit the marker parameters detected by the bedside detection device to the computer device; The fourth data interface is used to transmit the vital signs parameters, test parameters, and biomarker parameters from the computer device to the first user database.

13. The clinical auxiliary decision-making information generation system based on multimodal parameters according to claim 11, characterized in that, The system also includes: a historical user database connected to the computer device; The historical user database is used to store the full multimodal data of historical users.

14. The clinical auxiliary decision-making information generation system based on multimodal parameters according to claim 11, characterized in that, The system also includes: a knowledge engine database connected to the computer device; The knowledge engine database is used to store clinical knowledge engines, which include the mapping relationship between decision support information, key feature information, and risk level information.

15. The clinical auxiliary decision-making information generation system based on multimodal parameters according to claim 11, characterized in that, The system also includes: a second user database; The second user database is used to store the behavioral data and feedback data of the second user.

16. A clinical auxiliary decision-making information generation device based on multimodal parameters, characterized in that, The device includes: The acquisition unit is used to acquire the body parameter features and physiological parameter set of the first user; The data processing unit is used to perform spatiotemporal alignment and feature extraction on the physiological parameter set to obtain a multi-dimensional feature vector; The risk probability prediction unit is used to predict the risk probability of complications based on the multi-dimensional feature vector and shape parameter features through a pre-built risk prediction model, and to obtain the risk probability of multiple events within multiple time windows. The decision analysis unit is used to perform decision analysis based on the multi-event risk probability, multi-dimensional feature vector, and shape parameter features within the multi-time window, and generate decision support information.

17. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the clinical auxiliary decision-making information generation method based on multimodal parameters as described in any one of claims 1 to 10.

18. A computer device comprising a memory and a processor, the memory for storing information including program instructions, and the processor for controlling the execution of the program instructions, characterized in that, When the program instructions are loaded and executed by the processor, they implement the clinical auxiliary decision information generation method based on multimodal parameters as described in any one of claims 1 to 10.

19. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the clinical auxiliary decision-making information generation method based on multimodal parameters as described in any one of claims 1 to 10.

Images