Perioperative risk early warning method based on machine learning

Abstract

This invention provides a machine learning-based perioperative risk warning method. It involves acquiring assessment parameters associated with perioperative risk assessment for patients awaiting surgery; collecting multiple primary data points corresponding to these assessment parameters; processing these primary data points; inputting the processing results into a perioperative risk prediction model; and outputting a perioperative risk warning result for the patient. This method can automatically predict and warn of various potential perioperative risks based on the assessment parameters and corresponding data associated with perioperative risk assessment, using a perioperative risk prediction model. This approach does not rely on human experience and can detect adverse perioperative events early, thereby effectively reducing postoperative complications and ensuring surgical safety.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a perioperative risk warning method based on machine learning. Background Technology

[0002] Identifying and predicting / providing early warnings of perioperative risks in surgical patients is crucial for ensuring patient safety, creating favorable conditions for surgery, and supporting anesthesiologists' treatment decisions. It is also an important component of clinical anesthesia, playing a vital role in promoting rapid patient recovery and early reintegration into society. However, current interventions for perioperative vital sign fluctuations are primarily reactive and often delayed. Clinical methods for predicting significant fluctuations in vital signs are limited, relying heavily on anesthesiologists' experience based on the patient's condition, preoperative medications, surgical procedures, anesthetic drugs, and intraoperative monitoring of physiological indicators. Early circulatory instability is often difficult to detect clinically. While evidence suggests subtle dynamic relationships between different physiological variables during this period, experienced anesthesiologists can sometimes use this to predict adverse events. However, due to a lack of reproducibility and verification methods, this clinical experience is difficult to impart, hindering early detection of perioperative adverse events and thus limiting the effective reduction of postoperative complications and the assurance of surgical safety. Summary of the Invention

[0003] The purpose of this invention is to provide a perioperative risk warning method based on machine learning, so as to detect adverse perioperative events as early as possible, thereby effectively reducing postoperative complications and ensuring surgical safety.

[0004] This invention provides a perioperative risk warning method based on machine learning. The method includes: acquiring assessment parameters associated with perioperative risk assessment for patients undergoing surgery; collecting multiple first data points corresponding to the assessment parameters from a preset data source according to a preset acquisition method; processing the multiple first data points according to a preset method to obtain processing results; inputting the processing results into a pre-trained perioperative risk prediction model and outputting a perioperative risk warning result for the patient undergoing surgery; wherein the perioperative risk warning result includes at least one of the following: preoperative risk category and corresponding risk level, intraoperative risk category and corresponding risk level, postoperative risk category and corresponding risk level, and post-discharge risk category and corresponding risk level.

[0005] Furthermore, the steps for obtaining assessment parameters associated with perioperative risk assessment for patients awaiting surgery include: obtaining the initial case corresponding to the patient awaiting surgery; inputting the initial case into a pre-trained risk factor screening and assessment model, and outputting the assessment parameters associated with perioperative risk assessment for the patient awaiting surgery.

[0006] Furthermore, the steps for processing multiple primary data points according to a preset method to obtain processing results include: standardizing the multiple primary data points according to a preset standardization method to obtain processed primary data points; dividing the processed primary data points according to perioperative stages to obtain stage risk assessment data; wherein, the stage risk assessment data includes: preoperative risk assessment data; dividing the preoperative risk assessment data into preoperative static data and preoperative dynamic data according to the data fluctuation characteristics; processing the preoperative static data according to a first preset method to obtain preoperative static feature analysis results; and processing the preoperative dynamic data according to a second preset method to obtain preoperative dynamic feature analysis results.

[0007] Furthermore, the stage risk assessment data also includes: intraoperative risk assessment data; the method also includes: dividing the intraoperative risk assessment data into intraoperative static data and intraoperative dynamic data according to the fluctuation characteristics of the data; processing the intraoperative static data according to the first preset method to obtain the intraoperative static feature analysis results; and processing the intraoperative dynamic data according to the second preset method to obtain the intraoperative dynamic feature analysis results.

[0008] Furthermore, the stage risk assessment data also includes: postoperative risk assessment data; the method also includes: dividing the postoperative risk assessment data into postoperative static data and postoperative dynamic data according to the fluctuation characteristics of the data; processing the postoperative static data according to the first preset method to obtain the postoperative static feature analysis results; and processing the postoperative dynamic data according to the second preset method to obtain the postoperative dynamic feature analysis results.

[0009] Furthermore, the stage risk assessment data also includes: post-discharge risk assessment data; the method also includes: dividing the post-discharge risk assessment data into post-discharge static data and post-discharge dynamic data according to the fluctuation characteristics of the data; processing the post-discharge static data according to the first preset method to obtain the post-discharge static feature analysis results; and processing the post-discharge dynamic data according to the second preset method to obtain the post-discharge dynamic feature analysis results.

[0010] Furthermore, the steps for processing the preoperative static data according to the first preset method to obtain the preoperative static feature analysis results include: performing feature amplification on the preoperative static data to obtain amplified preoperative static data; performing feature extraction on the amplified preoperative static data to obtain preoperative static feature extraction results; and performing residual network analysis on the preoperative static feature extraction results to obtain preoperative static feature analysis results.

[0011] Furthermore, the steps for processing the preoperative dynamic data according to the second preset method to obtain the preoperative dynamic feature analysis results include: extracting the preoperative spatial domain features and preoperative temporal domain features from the preoperative dynamic data; performing feature extraction on the preoperative spatial domain features and preoperative temporal domain features respectively to obtain the preoperative spatial domain feature extraction results and the preoperative temporal domain feature extraction results; performing fully convolutional network analysis on the preoperative spatial domain feature extraction results to obtain the preoperative spatial domain feature analysis results; performing recurrent gate unit analysis on the preoperative temporal domain feature extraction results to obtain the preoperative temporal domain feature analysis results; and merging the preoperative spatial domain feature analysis results and the preoperative temporal domain feature analysis results to obtain the preoperative dynamic feature analysis results.

[0012] Furthermore, the steps for processing intraoperative static data according to the first preset method to obtain intraoperative static feature analysis results include: performing feature amplification on the intraoperative static data to obtain amplified intraoperative static data; performing feature extraction on the amplified intraoperative static data to obtain intraoperative static feature extraction results; and performing residual network analysis on the intraoperative static feature extraction results to obtain intraoperative static feature analysis results.

[0013] Furthermore, the steps for processing intraoperative dynamic data according to the second preset method to obtain intraoperative dynamic feature analysis results include: extracting intraoperative spatial and temporal features from the intraoperative dynamic data; performing feature extraction on the intraoperative spatial and temporal features respectively to obtain intraoperative spatial feature extraction results and intraoperative temporal feature extraction results; performing fully convolutional network analysis on the intraoperative spatial feature extraction results to obtain intraoperative spatial feature analysis results; performing recurrent gate unit analysis on the intraoperative temporal feature extraction results to obtain intraoperative temporal feature analysis results; and merging the intraoperative spatial feature analysis results and intraoperative temporal feature analysis results to obtain intraoperative dynamic feature analysis results.

[0014] Furthermore, the steps for processing postoperative static data according to the first preset method to obtain postoperative static feature analysis results include: performing feature amplification on the postoperative static data to obtain amplified postoperative static data; performing feature extraction on the amplified postoperative static data to obtain postoperative static feature extraction results; and performing residual network analysis on the postoperative static feature extraction results to obtain postoperative static feature analysis results.

[0015] Furthermore, the steps for processing postoperative dynamic data according to the second preset method to obtain postoperative dynamic feature analysis results include: extracting postoperative spatial and temporal features from the postoperative dynamic data; performing feature extraction on the postoperative spatial and temporal features respectively to obtain postoperative spatial feature extraction results and postoperative temporal feature extraction results; performing fully convolutional network analysis on the postoperative spatial feature extraction results to obtain postoperative spatial feature analysis results; performing recurrent gate unit analysis on the postoperative temporal feature extraction results to obtain postoperative temporal feature analysis results; and merging the postoperative spatial feature analysis results and postoperative temporal feature analysis results to obtain postoperative dynamic feature analysis results.

[0016] Furthermore, the steps for processing the post-discharge static data according to the first preset method to obtain the post-discharge static feature analysis results include: performing feature amplification on the post-discharge static data to obtain amplified post-discharge static data; performing feature extraction on the amplified post-discharge static data to obtain post-discharge static feature extraction results; and performing residual network analysis on the post-discharge static feature extraction results to obtain post-discharge static feature analysis results.

[0017] Furthermore, the steps for processing the post-discharge dynamic data according to the second preset method to obtain the post-discharge dynamic feature analysis results include: extracting the post-discharge spatial and temporal features of the post-discharge dynamic data; performing feature extraction on the post-discharge spatial and temporal features respectively to obtain the post-discharge spatial feature extraction results and the post-discharge temporal feature extraction results; performing fully convolutional network analysis on the post-discharge spatial feature extraction results to obtain the post-discharge spatial feature analysis results; performing cyclic gate unit analysis on the post-discharge temporal feature extraction results to obtain the post-discharge temporal feature analysis results; and merging the post-discharge spatial and temporal feature analysis results to obtain the post-discharge dynamic feature analysis results.

[0018] Furthermore, the steps of inputting the processing results into the pre-trained perioperative risk prediction model and outputting the perioperative risk warning results for the patients to be operated on include: inputting the preoperative static feature analysis results and the preoperative dynamic feature analysis results into the perioperative risk prediction model and outputting the preoperative risk category and the corresponding risk level.

[0019] Furthermore, the method also includes: inputting the results of intraoperative static feature analysis and intraoperative dynamic feature analysis into the perioperative risk prediction model, and outputting the intraoperative risk category and the corresponding risk level.

[0020] Furthermore, the method also includes: inputting the postoperative static feature analysis results and the postoperative dynamic feature analysis results into the perioperative risk prediction model, and outputting the postoperative risk category and the corresponding risk level.

[0021] Furthermore, the method also includes: inputting the post-discharge static feature analysis results and the post-discharge dynamic feature analysis results into the perioperative risk prediction model, and outputting the post-discharge risk category and the corresponding risk level.

[0022] Furthermore, the method also includes: inputting perioperative risk warning results into a pre-trained intervention plan model, and outputting a target intervention plan for patients to be operated on; wherein, the target intervention plan includes: preventive intervention plan and / or treatment intervention plan.

[0023] Furthermore, the method also includes: obtaining the anesthesia plan for the patient to be operated on; displaying the anesthesia plan, preoperative risk category and corresponding risk level through a first designated device; displaying the intraoperative risk category, and the corresponding risk level and intervention plan through a second designated device; displaying the postoperative risk category, and the corresponding risk level and intervention plan through a third designated device; and displaying the post-discharge risk category, and the corresponding risk level and intervention plan through a fourth designated device.

[0024] Furthermore, the method also includes: obtaining complete medical records of patients undergoing surgery after completing all treatments, new data related to adverse events, actual case report forms, and actual post-discharge follow-up information; optimizing the risk factor screening and assessment model based on the complete medical records, actual case report forms, and actual post-discharge follow-up information to obtain a new risk factor screening and assessment model; and optimizing the perioperative risk prediction model based on the new data to obtain a new perioperative risk prediction model.

[0025] Furthermore, the method also includes: obtaining the actual intervention plans used for patients undergoing surgery; and optimizing the intervention plan model based on the actual intervention plans used to obtain a new intervention plan model.

[0026] Furthermore, the method also includes: generating anesthesia clinical pathway information based on the target intervention plan; sending the anesthesia clinical pathway information to a designated workstation through a first designated interface; and sending the associated data of the patient to be operated on to a third-party platform through a second designated interface; wherein the associated data of the patient to be operated on includes at least: the perioperative risk warning results and the target intervention plan of the patient to be operated on.

[0027] This invention provides a perioperative risk warning device based on machine learning. The device includes: an acquisition module for acquiring assessment parameters associated with perioperative risk assessment of a patient to be operated on; a collection module for collecting multiple first data points corresponding to the assessment parameters of the patient to be operated on from a preset data source according to a preset collection method; a processing module for processing the multiple first data points according to a preset method to obtain processing results; and an output module for inputting the processing results into a pre-trained perioperative risk prediction model and outputting perioperative risk warning results corresponding to the patient to be operated on. The perioperative risk warning results include at least one of the following: preoperative risk category and corresponding risk level, intraoperative risk category and corresponding risk level, postoperative risk category and corresponding risk level, and post-discharge risk category and corresponding risk level.

[0028] The present invention provides an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the machine-based perioperative risk warning method described above.

[0029] The present invention provides a machine-readable storage medium storing machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions cause the processor to implement any of the above-mentioned machine learning-based perioperative risk warning methods.

[0030] This invention provides a machine learning-based perioperative risk warning method. The method acquires assessment parameters associated with perioperative risk assessment for patients awaiting surgery; collects multiple first data points corresponding to these assessment parameters; processes these first data points; inputs the processing results into a perioperative risk prediction model; and outputs a perioperative risk warning result for the patient. This approach can automatically predict and warn of various potential perioperative risks based on the assessment parameters and corresponding data associated with perioperative risk assessment, using a perioperative risk prediction model. This method does not rely on human experience and can detect adverse perioperative events early, thereby effectively reducing postoperative complications and ensuring surgical safety. Attached Figure Description

[0031] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0032] Figure 1A flowchart illustrating a perioperative risk warning method based on machine learning, provided as an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the workflow of a perioperative risk prediction model provided in an embodiment of the present invention;

[0034] Figure 3 A schematic diagram of the structure of a perioperative risk early warning system based on machine learning provided in an embodiment of the present invention;

[0035] Figure 4 A schematic diagram of a perioperative risk early warning device based on machine learning provided in an embodiment of the present invention;

[0036] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0037] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0038] Identifying surgical patients with potential perioperative risks and conducting preoperative, intraoperative, and postoperative risk assessments and predictions / early warnings is of great significance. This is especially true for patients with insufficient organ reserve or underlying diseases, who have a higher probability of experiencing adverse events during and after surgery compared to conventional treatment. Therefore, identifying and predicting / early warnings of perioperative risks in surgical patients is crucial for ensuring patient safety, creating favorable conditions for surgery, and supporting anesthesiologists' treatment decisions. It is also an important component of clinical anesthesia, playing a vital role in promoting rapid patient recovery and early reintegration into society.

[0039] With advancements in medical technology, an increasing number of monitoring methods are being used for preoperative, intraoperative, and postoperative monitoring, as well as for post-discharge patient monitoring. The increasing number of monitoring indicators and the growing volume of patient information and data provide both deeper and more comprehensive understanding of the patient's condition for anesthesiologists and, to some extent, an information overload.

[0040] Meanwhile, significant fluctuations in perioperative vital signs, such as intraoperative hypotension, hypertension, hypoxemia, and bradycardia, are independent risk factors for adverse perioperative outcomes and are closely related to many postoperative complications, such as a sharp increase in mortality, myocardial injury, acute kidney injury, postoperative delirium, and other neurological complications.

[0041] However, current interventions for perioperative vital sign fluctuations are primarily reactive and often delayed. Clinical methods for predicting significant fluctuations in vital signs are limited, relying mostly on the anesthesiologist's experience based on the patient's condition, preoperative medications, surgical procedures, anesthetic drugs, and intraoperative monitoring of physiological indicators. Early stages of circulatory instability are often difficult to detect clinically. Evidence suggests subtle dynamic relationships between different physiological variables during this period. Experienced anesthesiologists can sometimes use this to predict hypotensive events, but due to a lack of reproducibility and verification methods, this clinical experience is difficult to impart.

[0042] Furthermore, the perioperative period requires providing favorable conditions for successful surgery while simultaneously maintaining the functions of multiple systems, including the nervous, circulatory, respiratory, and urinary systems, and preventing various complications. This constitutes multi-objective task management. These multiple objectives may conflict; for example, simultaneously focusing on multiple management goals may lead to inconsistencies. For instance, using analgesics to relieve pain can result in respiratory depression, nausea, and vomiting if overdosed; accelerating intravenous fluid administration to maintain hemodynamic stability may lead to pulmonary edema; deepening anesthesia can suppress surgical stress, but excessively deep anesthesia is a high-risk factor for postoperative delirium. Anesthesiologists, especially those with limited experience, often struggle to coordinate these multiple objectives effectively, leading to difficulties in perioperative management. Therefore, this invention provides a machine learning-based perioperative risk warning method, applicable to scenarios requiring early warning of perioperative risks.

[0043] To facilitate understanding of this embodiment, a detailed description of a machine learning-based perioperative risk warning method disclosed in this invention will be provided first; for example... Figure 1 As shown, the method includes the following steps:

[0044] Step S102: Obtain assessment parameters associated with perioperative risk assessment for the patient to be operated on.

[0045] The assessment parameters mentioned above that are associated with perioperative risk assessment are typically the smallest granular parameters that contribute to perioperative risk assessment and prediction / early warning. Information on the elements that make up the case report form (CRF) can also usually be obtained. This CRF can be understood as the selected elements that can be used as clinical trial CRFs. This information serves as the primary basis for designing and defining the content and scoring criteria of the CRF. During the perioperative period, patients undergoing surgery typically undergo multiple examinations, which may contain a large amount of data. Some of this data is associated with perioperative risk assessment, while some is not. Therefore, when assessing the perioperative risk of patients undergoing surgery, it is necessary to first obtain the assessment parameters associated with perioperative risk assessment, and also obtain the information on the elements that make up the case report form.

[0046] Step S104: Collect multiple first data points corresponding to the evaluation parameters of the patient to be operated on from a preset data source according to a preset acquisition method.

[0047] The aforementioned preset data sources can be Electronic Medical Record (EMR) systems, Clinical Data Repository (CDR) systems, Hospital Information System (HIS) systems, surgical anesthesia information systems, medical record reports, real-time monitoring data of perioperative medical equipment, postoperative follow-up systems, etc. Different data sources may correspond to different data acquisition methods. For example, structured relevant data information can be obtained from third-party systems such as hospital EMR systems, clinical big data centers, laboratory testing systems, imaging examination systems, and surgical anesthesia information systems through data views, data interfaces, or hospital integration platforms. When monitoring vital signs and conducting health checks on patients during or after surgery, data monitored by medical equipment such as electrocardiogram monitors, anesthesia machines, ventilators, and bedside ultrasound can be collected in real time through dedicated data acquisition terminals and uploaded to the system's database. If case report form elements are also available, the case report form category and content can be defined based on these elements, and corresponding content can be filled in. Specifically, the corresponding data parameters can be extracted from the table. In addition, discharge follow-up information includes the distribution of follow-up information forms, records of telephone follow-up information, and information that can be directly extracted from the follow-up information forms.

[0048] In practice, multiple data sources can be pre-selected, and each data source can be used to collect multiple first data points corresponding to the assessment parameters associated with perioperative risk assessment for the patient to be operated on from multiple data sources. If the element information of the case report form is also obtained, multiple second data points corresponding to the element information that makes up the case report form can also be collected.

[0049] Step S106: Process the multiple first data according to a preset method to obtain the processing result.

[0050] Once the aforementioned multiple sets of primary data are acquired, these data are typically processed, such as standardization, normalization, consistency processing, correction of erroneous data, and appropriate processing of static and dynamic data to obtain corresponding processing results. If multiple sets of secondary data are also collected, they are also processed accordingly to obtain corresponding processing results.

[0051] Step S108: Input the processing results into the pre-trained perioperative risk prediction model and output the perioperative risk warning results for the patient to be operated on; wherein, the perioperative risk warning results include at least one of the following: preoperative risk category and corresponding risk level, intraoperative risk category and corresponding risk level, postoperative risk category and corresponding risk level, and post-discharge risk category and corresponding risk level.

[0052] The aforementioned perioperative risk prediction model can be implemented based on a deep learning algorithm that integrates multi-view features. The risk categories can be adverse events that may occur in patients undergoing surgery within a preset timeframe, before, during, or after surgery, or after discharge. For example, intraoperative adverse events may include: risks of hypotension, hypertension, hypoxemia, malignant arrhythmias, cardiac arrest, complications, and death. The risk levels can be specifically divided into three categories: high risk, medium risk, and low risk. High risk indicates a serious threat to the patient's life or other serious consequences, requiring immediate intervention; or a high probability of occurrence, requiring timely intervention to reduce the occurrence of adverse events. Medium risk indicates certain consequences for the patient, and without timely intervention, it may develop into a high-risk condition. Low risk indicates the patient is currently in a relatively safe state, with an extremely low probability of adverse events occurring in the short term. Of course, more risk levels can be divided according to actual needs. Usually, each risk category can have its corresponding risk level. In actual implementation, after obtaining the above processing results, the processing results can be input into the pre-trained perioperative risk prediction model. The perioperative risk prediction model outputs the risk categories of the patient before surgery, during surgery, after surgery, or after discharge, as well as the risk level corresponding to each risk category.

[0053] The aforementioned machine learning-based perioperative risk warning method acquires assessment parameters associated with perioperative risk assessment for patients awaiting surgery; collects multiple primary data points corresponding to these assessment parameters; processes these primary data points; inputs the processed results into a perioperative risk prediction model; and outputs a perioperative risk warning result for the patient. This method can automatically predict and warn of various potential perioperative risks based on the assessment parameters and corresponding data associated with perioperative risk assessment, through a perioperative risk prediction model. This approach does not rely on human experience and can detect adverse perioperative events as early as possible, thereby effectively reducing postoperative complications and ensuring surgical safety.

[0054] This invention also provides another perioperative risk warning method based on machine learning, which is implemented based on the method in the above embodiments and includes the following steps:

[0055] Step 1: Obtain the initial medical records of the patients to be operated on.

[0056] Step two involves inputting the initial case data into a pre-trained risk factor screening and assessment model, which outputs assessment parameters related to perioperative risk assessment for patients awaiting surgery.

[0057] The initial case may include basic examinations performed on the patient before surgery, such as height, weight, and blood pressure. This initial case is input into a pre-trained risk factor screening and assessment model. The model outputs results to confirm whether the patient is suitable for perioperative risk assessment. If the patient is confirmed as suitable, they are marked, and the assessment parameters associated with perioperative risk assessment are output. The model can also output information about the elements that make up the case report form. The assessment parameters associated with perioperative risk assessment are generally selected from a large number of medical records, identifying 64 categories of data that significantly contribute to perioperative risk warning / prediction. These mainly include: basic patient information (age, gender, height, weight, etc.), lifestyle habits (drinking history, smoking history, etc.), underlying diseases (hypertension, diabetes, hyperlipidemia, etc.), past medical history, past surgical history, laboratory test information (hemoglobin, albumin, serum creatinine, etc.), and imaging examination information (ultrasound, CT, etc., where CT stands for Computed Tomography). Tomography (computed tomography), intraoperative monitoring (respiration, blood oxygen, blood pressure, pulse, etc.), postoperative in-hospital monitoring information (vital signs, complications in various systems, etc.), post-discharge follow-up information (survival status, activities of daily living score, health status score, pain score, cognitive function, etc.), surgical information (surgical name, surgical grade, patient position, etc.), and anesthesia-related information (ASA rating, anesthesia method, intraoperative / postoperative sedation, analgesia, and other adjuvant drugs, etc. ASA stands for American Society of Anesthesiologists (ASA) Physical Status Classification System.

[0058] The case report form can include information on the following elements: activities of daily living (eating independently, walking independently, toileting, etc.); severity of insomnia (difficulty falling asleep, early awakening, impact on quality of life, etc.); nutrition (appetite, weight changes, activity level, etc.); anxiety (understanding of surgery, anesthesia, signed informed consent form, etc.); depression (feelings of emptiness, boredom, worry, memory, self-esteem, etc.); respiratory system; cardiovascular system; endocrine system; liver and kidney function; laboratory tests (complete blood count, blood glucose, thyroid function, electrolytes, arterial blood gases, etc.); nervous system; surgery; anesthesia; intraoperative monitoring data; intraoperative anesthetic information; postoperative analgesia information; intraoperative complications; postoperative complications; recovery status; perioperative adverse event management plan; post-discharge follow-up (survival status, activities of daily living score, health status score, pain score, cognitive function, etc.). The results obtained through the risk factor screening and assessment model can reduce the pressure of subsequent data collection while ensuring the efficiency and effectiveness of the learning algorithm for the subsequent perioperative risk prediction model.

[0059] Step 3: Collect multiple first data points corresponding to the evaluation parameters from the preset data source of the patient to be operated on, according to the preset collection method.

[0060] Step 4: Standardize the multiple first data points according to the preset standardization method to obtain the processed multiple first data points.

[0061] Due to differences in information collection methods and channels, data quality issues such as duplication, inconsistency, and errors may exist. Therefore, it is necessary to standardize, normalize, and ensure consistency of the data according to pre-defined standardization methods to correct erroneous data and ensure the quality of input parameters used in perioperative risk assessment and predictive early warning calculations. For example, duplicate data can be retained in isolation; obviously erroneous data (e.g., a body temperature of 25 degrees Celsius) can be deleted directly; data from different systems can be combined according to patient ID (Identity Document) to expand the parameter items under the same patient ID; and padding can be performed when the length of a parameter field from different systems does not meet the standard length. These are all methods for standardizing, normalizing, and ensuring consistency of the data. If the collected data includes multiple primary data points and multiple secondary data points corresponding to element information, then both primary and secondary data points need to be standardized separately to obtain their respective processed primary and secondary data points.

[0062] Step 5: Divide the processed primary data into perioperative stages to obtain stage risk assessment data; the stage risk assessment data includes: preoperative risk assessment data, intraoperative risk assessment data, postoperative risk assessment data, and post-discharge risk assessment data.

[0063] Step 6: Based on the fluctuation characteristics of the data, the preoperative risk assessment data is divided into preoperative static data and preoperative dynamic data.

[0064] The processed first data can be categorized. If multiple processed second data are also obtained, they can be categorized in the same way. The information categorization can be done in two steps: First, the perioperative stage can be divided to obtain preoperative risk assessment data, intraoperative risk assessment data, postoperative risk assessment data, and post-discharge risk assessment data. Specifically, the risk assessment data can be obtained according to the patient's actual perioperative stage. For example, if the patient is in the preoperative stage, only preoperative risk assessment data is usually obtained. If the patient is in the postoperative stage, the data can include preoperative risk assessment data, intraoperative risk assessment data, and postoperative risk assessment data. The specific categorization results can be obtained according to the actual situation.

[0065] Data from each stage can be repeated; for example, preoperative test data needs to be stored in data storage devices for three stages. Secondly, the data information at each stage may correspond to different data sources, methods, acquisition timing, and uses. It can be categorized into dynamic and static data based on their fluctuation characteristics, and corresponding algorithms can be applied to each type of data. Static data mainly refers to patient age, gender, surgical procedure, preoperative test results, preoperative medication, and preoperative assessment scale data, etc. These parameters do not fluctuate significantly during the perioperative period. Dynamic data mainly refers to vital sign parameters, EEG monitoring parameters, and bedside ultrasound monitoring parameters that fluctuate considerably during the perioperative period.

[0066] Step 7: Process the preoperative static data according to the first preset method to obtain the preoperative static feature analysis results;

[0067] This step seven can be achieved through the following steps 70 to 72:

[0068] Step 70: Perform feature amplification on the preoperative static data to obtain amplified preoperative static data;

[0069] In practice, once the preoperative static data is obtained, feature amplification can be performed on the preoperative static data to obtain amplified preoperative static data. Since the amplified preoperative static data has a larger data volume and richer data information, analysis based on the amplified preoperative static data helps to obtain more accurate data analysis results.

[0070] Step 71: Extract features from the amplified preoperative static data to obtain the preoperative static feature extraction results;

[0071] Feature extraction is performed on the amplified preoperative static data, that is, the most effective features are found from the amplified preoperative static data to transform the amplified preoperative static data into a set of features with obvious physical or statistical significance, and the preoperative static feature extraction results are obtained.

[0072] Step 72: Perform residual network analysis on the preoperative static feature extraction results to obtain the preoperative static feature analysis results.

[0073] The preoperative static feature extraction results are input into a preset residual network for residual network analysis to obtain the preoperative static feature analysis results.

[0074] For example, by utilizing multiple views, static data features such as preoperative patient basic information (age, gender, height, weight, etc.), lifestyle habits (drinking history, smoking history, etc.), underlying diseases (hypertension, diabetes, hyperlipidemia, etc.), past medical history, past surgical history, laboratory test information (hemoglobin, albumin, serum creatinine, etc.), and imaging examination information are fully extracted from different angles. Intraoperative temporal features are then selected according to the observation window, and the mean, maximum, minimum, and median of features such as blood pressure and heart rate are extracted to form temporal statistical features. These features are then combined with the selected preoperative static features to construct a static feature view. To align features, a set of identical preoperative static features is combined for the temporal statistical features of each patient for each time window. ResNet learning can output multidimensional integrated static features.

[0075] Step 8: Process the preoperative dynamic data according to the second preset method to obtain the preoperative dynamic feature analysis results.

[0076] This step eight can be achieved through the following steps 80 to 82:

[0077] Step 80: Extract the preoperative spatial domain features and preoperative temporal domain features from the preoperative dynamic data. Perform feature extraction on the preoperative spatial domain features and preoperative temporal domain features respectively to obtain the preoperative spatial domain feature extraction results and the preoperative temporal domain feature extraction results.

[0078] Step 81: Perform a fully convolutional network analysis on the preoperative spatial feature extraction results to obtain the preoperative spatial feature analysis results;

[0079] Step 82: Perform cyclic gate unit analysis on the preoperative time-domain feature extraction results to obtain the preoperative time-domain feature analysis results;

[0080] Step 83: Combine the preoperative spatial domain feature analysis results and the preoperative temporal domain feature analysis results to obtain the preoperative dynamic feature analysis results.

[0081] In practical implementation, after obtaining preoperative dynamic data, the preoperative dynamic data can be preprocessed. Specifically, preoperative spatial and temporal features of the preoperative dynamic data can be extracted. Feature extraction is performed on the preoperative spatial features, and the extracted results are input into a pre-defined fully convolutional network to obtain the preoperative spatial feature analysis results. Feature extraction is performed on the preoperative temporal features, and the extracted results are input into a pre-defined fully convolutional network to obtain the preoperative temporal feature analysis results. The preoperative spatial and temporal feature analysis results are then merged to obtain the preoperative dynamic feature analysis results.

[0082] For example, intraoperative temporal features such as mean arterial pressure (MAP), respiratory rate (RR), heart rate (HR), body temperature (BT), end-tidal carbon dioxide (ETCO2), and oxygen saturation (SpO2) are recorded according to the observation window and then duplicated. One copy is used for extracting time-dependent features to form a temporal feature view; the other copy is used for extracting time-invariant features from multivariate time-series data to construct a spatial feature view. For the temporal feature view, the temporal dependencies between temporal features are key to identifying early signs of physiological deterioration. To effectively uncover temporal dependencies in the temporal feature view, a GRU model (a type of recurrent neural network model) that can effectively capture potential dependencies in time-series data can be used. It uses the hidden states of preceding steps as additional inputs to subsequent steps to capture temporal dependencies. Multi-layer GRU units can be used. For the spatial feature view, the movement of FCN convolutional kernels can be used to capture time-invariant features in multivariate time-series data. Spatial features are extracted from local features and are not dependent on all data; they can appear at any time point. An FCN network with multiple convolutional layers, global pooling layers, and fully connected layers can be used. To prevent overfitting, a Dropout layer can be added after each GRU unit and the FCN global pooling layer. After learning through multiple layers of GRU and FCN fully connected layers, multidimensional temporal and spatial features can be output.

[0083] Step 9: Based on the fluctuation characteristics of the data, divide the intraoperative risk assessment data into intraoperative static data and intraoperative dynamic data;

[0084] Step 10: Process the intraoperative static data according to the first preset method to obtain the intraoperative static feature analysis results;

[0085] This step ten can be achieved through the following steps 100 to 102:

[0086] Step 100: Perform feature amplification on the intraoperative static data to obtain amplified intraoperative static data;

[0087] Step 101: Extract features from the amplified intraoperative static data to obtain the intraoperative static feature extraction results;

[0088] Step 102: Perform residual network analysis on the intraoperative static feature extraction results to obtain the intraoperative static feature analysis results.

[0089] Step 11: Process the intraoperative dynamic data according to the second preset method to obtain the intraoperative dynamic feature analysis results.

[0090] This eleventh step can be specifically achieved through steps 110 to 113:

[0091] Step 110: Extract the intraoperative spatial and temporal features of the intraoperative dynamic data. Perform feature extraction on the intraoperative spatial and temporal features respectively to obtain the intraoperative spatial feature extraction results and the intraoperative temporal feature extraction results.

[0092] Step 111: Perform fully convolutional network analysis on the intraoperative spatial feature extraction results to obtain the intraoperative spatial feature analysis results;

[0093] Step 112: Perform cyclic gate unit analysis on the intraoperative time-domain feature extraction results to obtain the intraoperative time-domain feature analysis results;

[0094] Step 113: Combine the results of intraoperative spatial feature analysis and intraoperative temporal feature analysis to obtain the results of intraoperative dynamic feature analysis.

[0095] The processing methods for intraoperative static and intraoperative dynamic data can refer to the above methods for processing preoperative static and intraoperative dynamic data, and will not be repeated here.

[0096] Step 12: Based on the fluctuation characteristics of the data, the postoperative risk assessment data is divided into postoperative static data and postoperative dynamic data.

[0097] Step thirteen: Process the postoperative static data according to the first preset method to obtain the postoperative static feature analysis results;

[0098] This step thirteen can be specifically achieved through steps 130 to 132:

[0099] Step 130: Perform feature amplification on the postoperative static data to obtain amplified postoperative static data;

[0100] Step 131: Extract features from the amplified postoperative static data to obtain the postoperative static feature extraction results;

[0101] Step 132: Perform residual network analysis on the postoperative static feature extraction results to obtain the postoperative static feature analysis results.

[0102] Step fourteen: Process the postoperative dynamic data according to the second preset method to obtain the postoperative dynamic feature analysis results.

[0103] Step fourteen can be specifically achieved through the following steps 140 to 143:

[0104] Step 140: Extract the postoperative spatial and temporal features of the postoperative dynamic data. Perform feature extraction on the postoperative spatial and temporal features respectively to obtain the postoperative spatial feature extraction results and the postoperative temporal feature extraction results.

[0105] Step 141: Perform fully convolutional network analysis on the postoperative spatial feature extraction results to obtain the postoperative spatial feature analysis results;

[0106] Step 142: Perform cyclic gate unit analysis on the postoperative time-domain feature extraction results to obtain the postoperative time-domain feature analysis results;

[0107] Step 143: Combine the postoperative spatial feature analysis results and the postoperative temporal feature analysis results to obtain the postoperative dynamic feature analysis results.

[0108] The processing methods for postoperative static and dynamic data can refer to the above methods for processing preoperative static and intraoperative dynamic data, and will not be repeated here.

[0109] Step 15: Based on the fluctuation characteristics of the data, divide the post-discharge risk assessment data into post-discharge static data and post-discharge dynamic data;

[0110] Step 16: Process the post-discharge static data according to the first preset method to obtain the post-discharge static feature analysis results;

[0111] This step sixteen can be specifically achieved through steps 160 to 162:

[0112] Step 160: Perform feature amplification on the post-discharge static data to obtain amplified post-discharge static data;

[0113] Step 161: Extract features from the amplified post-discharge static data to obtain the post-discharge static feature extraction results;

[0114] Step 162: Perform residual network analysis on the static feature extraction results after discharge to obtain the static feature analysis results after discharge.

[0115] Step 17: Process the post-discharge dynamic data according to the second preset method to obtain the post-discharge dynamic feature analysis results.

[0116] This step seventeen can be specifically achieved through steps 170 to 173:

[0117] Step 170: Extract the spatial and temporal features of the post-discharge dynamic data. Perform feature extraction on the spatial and temporal features of the post-discharge data to obtain the spatial feature extraction results and temporal feature extraction results of the post-discharge data.

[0118] Step 171: Perform fully convolutional network analysis on the spatial domain feature extraction results after discharge to obtain the spatial domain feature analysis results after discharge.

[0119] Step 172: Perform cyclic gate cell analysis on the time-domain feature extraction results after discharge to obtain the time-domain feature analysis results after discharge;

[0120] Step 173: Combine the spatial domain feature analysis results and the temporal domain feature analysis results after discharge to obtain the dynamic feature analysis results after discharge.

[0121] The processing methods for post-discharge static and post-discharge dynamic data can refer to the above methods for processing preoperative static and intraoperative dynamic data, and will not be repeated here.

[0122] Step 18: Input the preoperative static feature analysis results and the preoperative dynamic feature analysis results into the perioperative risk prediction model, and output the preoperative risk category and the corresponding risk level.

[0123] Step 19: Input the intraoperative static feature analysis results and intraoperative dynamic feature analysis results into the perioperative risk prediction model, and output the intraoperative risk category and corresponding risk level.

[0124] Step 20: Input the postoperative static feature analysis results and the postoperative dynamic feature analysis results into the perioperative risk prediction model, and output the postoperative risk category and the corresponding risk level.

[0125] Step 21: Input the post-discharge static feature analysis results and the post-discharge dynamic feature analysis results into the perioperative risk prediction model, and output the post-discharge risk category and the corresponding risk level.

[0126] The static and dynamic feature analysis results at each stage are input into the perioperative risk prediction model to predict perioperative risks, obtaining preoperative risk categories and corresponding risk levels, intraoperative risk categories and corresponding risk levels, postoperative risk categories and corresponding risk levels, and post-discharge risk categories and corresponding risk levels. For example, possible adverse events during surgery include: hypotension risk, hypertension risk, hypoxemia risk, malignant arrhythmia risk, cardiac arrest risk, complication risk, and death risk. Possible adverse events after surgery include: postoperative pain, hypothermia, nausea and vomiting, hypotension, hypertension, hypoxemia, postoperative delirium, deep vein thrombosis, pulmonary complications, major cardiovascular events, stroke, acute kidney injury, and in-hospital death. Possible adverse events within a preset time after discharge include: cardiovascular and cerebrovascular events within 1 year, pulmonary complications, malnutrition, weakness, cognitive decline, and death. The model also outputs high, medium, and low risk warnings for adverse events. The preoperative risk prediction results are usually not displayed on the main intraoperative interface, but can be viewed through the intraoperative interface. At the same time, the preoperative risk prediction results are also one of the input parameters of the intraoperative risk warning algorithm; the intraoperative risk prediction results can be displayed in real time during the operation.

[0127] See Figure 2 The diagram illustrates the workflow of a perioperative risk prediction model. Specifically, this model is a perioperative risk prediction / early warning model based on deep learning and multi-view feature fusion. Static features (corresponding to the static data mentioned above) include: age, gender, surgeon, and laboratory and examination results. After feature amplification, feature extraction, and residual network analysis (ResNet) of the static features, they are input into the deep learning result fusion module (corresponding to the perioperative risk prediction model mentioned above) for preoperative risk prediction. Dynamic features (corresponding to the dynamic data mentioned above) include: blood pressure, heart rate, blood oxygen saturation, end-tidal carbon dioxide partial pressure, EEG monitoring, bedside ultrasound detection, etc. The dynamic features are processed by extracting spatial and temporal features, and then performing fully convolutional network analysis (FCN) and gated recurrent unit analysis (GRU) respectively. These dynamic features are then input into the deep learning result fusion module for intraoperative / postoperative / post-discharge risk prediction, such as low-risk and high-risk adverse events warnings.

[0128] Step 22: Input the perioperative risk warning results into the pre-trained intervention plan model and output the target intervention plan for the patient to be operated on; wherein, the target intervention plan includes: preventive intervention plan and / or treatment intervention plan.

[0129] The intervention plan in this embodiment is also based on machine learning. The input information for the machine learning algorithm model of the intervention plan includes: effective experience of clinical anesthesiologists, management plans for previous cases, output parameters of perioperative risk assessment and prediction and early warning models, etc. The learning results of the intervention plan algorithm model include the following categories:

[0130] 1. Develop preventative intervention plans to address avoidable adverse events and prevent their occurrence. For example, an intraoperative hypotension risk prediction model can provide an early warning of hypotension risk 15-30 minutes in advance. Based on the warning, anesthesiologists can intervene in advance by accelerating intravenous infusion, administering vasoactive drugs, changing the patient's position, and adjusting anesthetic drugs to avoid intraoperative hypotension.

[0131] 2. For unavoidable adverse events, develop management intervention plans to ensure adequate preparation before the event occurs and timely intervention when it does, thus avoiding adverse consequences. For example, postoperative nausea and vomiting (PONV) has many influencing factors and is often unavoidable. However, predictive models can stratify patients into risk groups and provide different interventions. For instance, patients without PONV risk factors do not require prophylactic medication. Low- and intermediate-risk patients can use one or two of the aforementioned medications for prophylaxis. High-risk patients can use a combination of two to three medications for prophylaxis.

[0132] 3. Develop corresponding preventative and treatment intervention plans for various adverse events during the perioperative period at different stages. For example, for pulmonary complications in perioperative patients, different anesthesia methods, induction methods, airway management methods, and mechanical ventilation parameter settings are selected based on risk stratification to specifically avoid or intervene in risk factors at different stages.

[0133] 4. Based on the prevention and intervention plan for perioperative adverse events, and combined with the patient's vital signs and organ status at the time of the occurrence or prediction of adverse events, a personalized intervention plan is formed for each patient. For example, based on the basic plan for preventing postoperative delirium, postoperative cardiovascular and cerebrovascular complications, pulmonary complications, acute kidney injury, etc., and combined with the patient's cognitive function, baseline blood pressure, degree of cardiovascular and cerebrovascular lesions, renal function, etc., individualized management targets for parameters such as depth of anesthesia, blood pressure, heart rate, body temperature, and urine output are given, forming a personalized multi-objective management plan.

[0134] In practice, the perioperative risk warning results of the patient to be operated on can be input into a pre-trained intervention plan model, which will output preventive intervention plans and / or treatment intervention plans for the patient to be operated on.

[0135] Step 23: Obtain the anesthesia plan for the patient to be operated on.

[0136] Step 24: Display the anesthesia plan, preoperative risk category, and corresponding risk level through the first designated device.

[0137] The aforementioned anesthesia plan can be an anesthesia plan formulated by an anesthesiologist for the patient to be operated on; the aforementioned first designated device may include: an anesthesiologist's mobile terminal, on-call workstation, dedicated workstation, etc.; in actual implementation, the aforementioned anesthesia plan, preoperative adverse event risk categories and corresponding risk levels can be displayed through the first designated device.

[0138] Step 25: Display the intraoperative risk category, the corresponding risk level, and the intervention plan through the second designated device.

[0139] The aforementioned second designated device may include intraoperative monitoring terminals, etc. In actual implementation, real-time intraoperative early warning can be performed through the second designated device, specifically including: intraoperative adverse event categories and corresponding risk levels and intervention plans. Risk reminders can be based on the definition of "high, medium, and low" risk. For information on possible medium- or high-risk or adverse events, prominent reminders should be given, including through color, sound, flashing, etc. When the predicted risk is low, it should be presented normally, without prominent color, sound, flashing, etc.

[0140] Step 26: Display the postoperative risk category, the corresponding risk level, and the intervention plan through the third designated device.

[0141] The aforementioned third designated device may include: anesthesiologist mobile terminal, on-call workstation, dedicated workstation, etc.; in actual implementation, the third designated device can be used to display the postoperative adverse event risk category and the corresponding risk level and intervention plan.

[0142] Step 27: Display the post-discharge risk category, the corresponding risk level, and the intervention plan through the fourth designated device.

[0143] The aforementioned fourth designated device may include: anesthesiologist mobile terminal, on-call workstation, patient terminal, etc. In actual implementation, the fourth designated device can be used to display the adverse event risk categories and corresponding risk levels and intervention plans within a preset time after the patient's discharge. The preset time can be one year after the patient's discharge, etc., and can be set according to actual needs.

[0144] The aforementioned perioperative risk assessment and prediction early warning system covers the preoperative, intraoperative, postoperative in-hospital, and post-discharge preset time periods, fully combining clinical needs, and designs the output methods and content for each stage and implements them through the system.

[0145] Step 28: Obtain complete medical records of patients undergoing surgery after all treatments have been completed, new data related to adverse events, actual case report forms created, and actual follow-up information after discharge.

[0146] Step 29: Based on complete cases, actual established case report forms, and actual post-discharge follow-up information, optimize the risk factor screening and assessment model to obtain a new risk factor screening and assessment model.

[0147] Step 30: Based on the new data, optimize the perioperative risk prediction model to obtain a new perioperative risk prediction model.

[0148] Step 31: Obtain the actual intervention plan used for the surgical patient;

[0149] Step 32: Based on the actual intervention plan adopted, optimize the intervention plan model to obtain a new intervention plan model.

[0150] The accuracy of the perioperative risk prediction model can be determined by comparing its output with that of actual surgical patients. When inaccurate risk prediction events are found, targeted analysis is required. Inaccurate events mainly fall into the following categories.

[0151] 1. The perioperative risk prediction model outputs results predicting adverse events, but no intervention can be implemented, and the events do not actually occur;

[0152] 2. An adverse event occurred in the real world, but the perioperative risk prediction model predicted low risk or no risk warning.

[0153] 3. The adverse event that occurred in the real world is event 1, but the adverse event predicted by the perioperative risk prediction model is event 2;

[0154] 4. The severity of adverse events in the real world is high, but the severity of adverse events predicted by the perioperative risk prediction model is medium.

[0155] By applying the results of the intervention plan model to the real world, the effectiveness or efficiency of the intervention plan can be verified. Further optimization of the model is needed when: the intervention plan, corresponding to the risk level, fails to prevent adverse events (i.e., adverse events still occur); or the intervention plan fails to fully mitigate the adverse event, requiring additional measures to improve it; or the plan to prevent or mitigate adverse events is overly complex and can be further optimized; or when actions taken in practice are completely inconsistent with the intervention plan.

[0156] To address the above issues, this invention employs the following method for model optimization.

[0157] 1. Feedback the medical records of patients who have completed all treatments to the risk factor screening and assessment model, iteratively optimize the screening results of the risk factor screening and assessment model, and provide more valuable input parameters for the perioperative risk prediction model.

[0158] 2. Based on clinical observation and analysis, parameters that have a real impact on the occurrence of adverse events but were previously overlooked are fed back into the input of the perioperative risk prediction model to optimize the algorithm's feature vector. New data related to adverse events usually requires manual screening. For example, intraoperative urine output was not previously used as an input parameter for the early warning model, but after observing multiple cases, it was found that most patients with high risk (intraoperative hypertension) had increased urine output, so urine output will be used as an input parameter for the early warning model.

[0159] 3. Based on clinical observation and analysis, the actual measures taken to prevent or manage adverse events are fed back to the input of the intervention plan model to optimize the feature vector of the algorithm model.

[0160] 4. In conjunction with scientific research, the content, scores, and ratings of the established CRF scales for enrolled patients, which are currently used in anesthesiology, are fed back into the risk factor screening and assessment model. The screening elements of the CRF scale in the risk factor screening and assessment model are iteratively optimized to make the content of the CRF more scientific and more in line with the actual situation of patients.

[0161] 5. Combined with post-discharge follow-up information, the newly added follow-up information is fed back to the risk factor screening and assessment model to iteratively optimize the input parameters of the risk factor screening and assessment model for potential perioperative risk populations and perioperative risk prediction models, as well as the elements of post-discharge follow-up information.

[0162] The above method feeds the learning results of the machine learning algorithm model back to each machine learning model, adjusts the learning parameters, optimizes the model, improves the learning effect of the risk factor screening and assessment model, the perioperative risk prediction model, and the intervention plan model, and continuously optimizes the model.

[0163] Step 33: After all the information for each element of the case report form has been filled in with the corresponding data, the case report form is scored to generate a complete case report form for the patient awaiting surgery.

[0164] Step 34: Save the complete case report form to the preset database through the third designated interface.

[0165] The process data, risk assessment results, and adopted solutions generated during the application of this solution can be processed through the data conversion module of this solution, and then interfaced with research, clinical, and remote teaching systems to output corresponding application data. Specifically, for the research system, this embodiment outputs scale elements related to the Case Report Form (CRF) during the risk factor screening and assessment model algorithm stage. Clinicians design CRF-related scale content based on this and store the completed table format in the database. Through the data conversion module, each CRF corresponding to a patient is retrieved from the database on a patient-by-patient basis. Combined with the automation of the information collection module, the CRF scale content can be automatically filled in. Data that cannot be collected by the information collection module is manually supplemented by medical staff. After the system determines that the CRF scale content is complete, the system will automatically score the CRF scale content to form a complete CRF report for the patient, i.e., the aforementioned complete case report form, and store it in the corresponding database of the research system through the interface (corresponding to the third specified interface mentioned above).

[0166] Step 35: Generate anesthesia clinical pathway information based on the target intervention plan.

[0167] Step 36: Send the anesthesia clinical pathway information to the designated workstation through the first designated interface.

[0168] For the clinical system, this embodiment of the solution can use the data conversion and processing module to extract key patient vital signs, past medical history, allergy history, planned surgery, anesthesia method, and detailed anesthesia events at each stage from the intervention plan output by the intervention plan model. These elements are then combined to form anesthesia clinical pathway information. By connecting to the clinical system interface (corresponding to the first designated interface mentioned above), the anesthesia clinical pathway information can be imported into the anesthesia clinical workstation (corresponding to the designated workstation mentioned above), forming a basis for anesthesiologists to carry out anesthesia-related work in a standardized manner, and ensuring the quality and safety of anesthesia.

[0169] Step 37: Send the associated data of the patient to be operated on to the third-party platform through the second designated interface; wherein, the associated data of the patient to be operated on includes at least: the perioperative risk warning results and the target intervention plan of the patient to be operated on.

[0170] For the remote teaching system, this embodiment of the solution can, through the data conversion and processing module, guide lower-level hospitals in conducting perioperative risk assessment and prediction / early warning. On the other hand, it can also teach lower-level hospitals the methods and content of perioperative risk assessment and prediction / early warning. Specifically, ① it guides lower-level hospitals: Surgical patient information from lower-level hospitals is remotely transmitted to higher-level hospitals via the network. The data conversion and processing module standardizes and converts this information into data that meets the format requirements of this solution. The risk assessment and prediction / early warning results, intervention plans, and other information output by this system are then remotely transmitted to lower-level hospitals via the network. By connecting to the remote teaching system interface, it accesses the lower-level hospital's teaching platform, providing real-time guidance on perioperative risk assessment and prediction / early warning. ② When teaching perioperative risk assessment and prediction and early warning to lower-level hospitals, this solution, according to teaching requirements, uses a data conversion module to extract necessary input and output parameters, patient conditions, assessment results, intervention plans, intervention effects, and other information from the system for each stage of perioperative risk assessment and prediction and early warning. This system can recombine this information according to pre-set rules, connect to the remote teaching system interface, store it in the remote teaching system database, and then the remote teaching system can use it as needed.

[0171] The aforementioned machine learning-based perioperative risk warning method, through the collection of a large amount of historical data and in-depth mining of the correlations between data, combined with artificial intelligence, proactively identifies potential perioperative risk populations and conducts risk assessments and predictions / warnings at preset time points before, during, and after surgery and discharge. It also intelligently generates personalized treatment plans, transforming risk prediction that previously relied solely on experience into more scientific prediction and warning. This allows more experiential results to be transformed into knowledge and technological achievements, which can be applied to clinical practice, research, and teaching. It can significantly help anesthesiologists detect adverse events earlier, reduce postoperative complications, ensure surgical safety, and powerfully promote the development of the discipline.

[0172] To further understand the above embodiments, the following provides... Figure 3 The diagram illustrates the structure of a machine learning-based perioperative risk early warning system. This system organically combines information filtering, information collection, machine learning algorithm application (risk factor screening and assessment model, perioperative risk prediction model, intervention plan model), information output and presentation, secondary data processing, a feedback module, and a model optimization module to form an applicable information system. The system uses a front-end / back-end separation approach. The front-end application only includes the output results of the "information presentation" module; the back-end application is responsible for information integration, management, and complex calculations, as detailed below:

[0173] The information filtering algorithm application module (corresponding to the risk factor screening and assessment model mentioned above) learns from massive amounts of basic cases to screen out patients to be assessed, basic parameters related to perioperative risk assessment and prediction, and elements of the case report form. Based on the screened information, information is collected. Specifically, static data collection can be performed from data sources such as clinical data centers, electronic medical record systems, laboratory testing systems, surgical anesthesia information systems, surgical nursing information systems, and post-discharge follow-up systems. Dynamic data collection can be performed from data sources such as electrocardiogram monitors, anesthesia machines, ventilators, bedside ultrasound, infusion pumps, and wearable vital sign acquisition devices outside the hospital. The obtained static and dynamic data undergo information preprocessing. The processing results are input into the risk prediction algorithm application module (corresponding to the perioperative risk prediction model mentioned above). The module outputs preoperative risk prediction results, intraoperative / postoperative / post-discharge risk prediction results, and the corresponding risk level, such as high risk, medium risk, or low risk. The output risk prediction results and corresponding risk levels can be displayed on the display terminal through the information output module. Simultaneously, the intervention plan algorithm application module (corresponding to the intervention plan model mentioned above) can output an intervention plan, which can also be displayed on the display terminal through the information output module. The output results of the machine learning algorithm model can be fed back to the information filtering algorithm application module, risk prediction algorithm application module, and intervention plan algorithm application module through the feedback module and model optimization module to optimize each module and improve their learning effectiveness.

[0174] This system can predict and provide early warnings about potential adverse events that surgical patients may face before, during, and after surgery, as well as after discharge, and offers corresponding management methods and treatment plans. Based on comprehensive information about surgical patients, including their basic physical condition, preoperative diagnosis, laboratory tests and examinations, planned surgery, preoperative assessment, intraoperative monitoring, anesthesia management, postoperative treatment, and post-discharge follow-up, the system uses artificial intelligence technology to predict and provide early warnings about various risks and adverse events that may occur during the perioperative period. This supports anesthesiologists, clinicians, and other medical staff in taking proactive risk management measures, thereby ensuring perioperative surgical safety.

[0175] This system incorporates artificial intelligence machine learning algorithms, fully leveraging their ability to interpret a wide range of features and analyze extremely complex feature sets with varying depths of interaction. This enables the system to achieve human-like perception and simultaneous thinking, allowing perioperative risk warning work, previously only achievable by a few anesthesiologists with extensive clinical experience, to be more widely applied in clinical practice. This promotes the downward flow of high-quality medical resources and addresses the increasingly numerous and complex surgical treatments in my country. Specifically, this manifests in the following aspects.

[0176] (1) With the increasing aging population and the continuous improvement of medical standards in my country, the number of surgeries and the difficulty of surgeries are constantly rising, and the age range of surgical patients is also expanding. Consequently, the inherent risks of surgery and postoperative complications are also increasing. Using artificial intelligence to assist anesthesiologists in predicting and making treatment decisions based on preoperative and intraoperative monitoring indicators can help address the unpredictable perioperative risks caused by frailty, comorbidities, and decline in physical function, thus providing safety guarantees for surgical patients.

[0177] (2) In conjunction with clinical anesthesia work, this invention is applied to guide the control of anesthesia depth and the maintenance of organ function, and is embedded into the closed loop of assessment-diagnosis-intervention-reassessment to create a complete perioperative assessment system and provide technical support for precision medicine.

[0178] (3) This solution uses an information management module to comprehensively structure perioperative information, which directly impacts scientific research and provides big data support that meets the needs of scientific research.

[0179] Corresponding to the above method embodiments, a perioperative risk early warning device based on machine learning is provided below, such as... Figure 4 As shown, the device includes: an acquisition module 40, used to acquire assessment parameters associated with perioperative risk assessment of the patient to be operated on; a collection module 41, used to collect multiple first data points corresponding to the assessment parameters of the patient to be operated on from a preset data source according to a preset collection method; a processing module 42, used to process the multiple first data points according to a preset method to obtain processing results; and an output module 43, used to input the processing results into a pre-trained perioperative risk prediction model and output perioperative risk warning results corresponding to the patient to be operated on; wherein, the perioperative risk warning results include at least one of the following: preoperative risk category and corresponding risk level, intraoperative risk category and corresponding risk level, postoperative risk category and corresponding risk level, and post-discharge risk category and corresponding risk level.

[0180] The aforementioned machine learning-based perioperative risk warning device acquires assessment parameters associated with perioperative risk assessment for patients awaiting surgery; collects multiple primary data points corresponding to these assessment parameters; processes these primary data points; inputs the processing results into a perioperative risk prediction model; and outputs a perioperative risk warning result for the patient. This device can automatically predict and warn of various potential perioperative risks based on the assessment parameters and corresponding data associated with perioperative risk assessment, using the perioperative risk prediction model. This method does not rely on human experience and can detect adverse perioperative events as early as possible, thereby effectively reducing postoperative complications and ensuring surgical safety.

[0181] Furthermore, the acquisition module is used to: acquire the initial case corresponding to the patient to be operated on; input the initial case into the pre-trained risk factor screening and assessment model, and output the assessment parameters of the patient to be operated on that are associated with perioperative risk assessment.

[0182] Furthermore, the processing module is also used to: standardize multiple first data according to a preset standardization method to obtain multiple processed first data; divide the multiple processed first data according to the perioperative stage to obtain stage risk assessment data; wherein, the stage risk assessment data includes: preoperative risk assessment data; divide the preoperative risk assessment data into preoperative static data and preoperative dynamic data according to the fluctuation characteristics of the data; process the preoperative static data according to a first preset method to obtain preoperative static feature analysis results; process the preoperative dynamic data according to a second preset method to obtain preoperative dynamic feature analysis results.

[0183] Furthermore, the stage risk assessment data also includes: intraoperative risk assessment data; the processing module is also used to: divide the intraoperative risk assessment data into intraoperative static data and intraoperative dynamic data according to the fluctuation characteristics of the data; process the intraoperative static data according to the first preset method to obtain the intraoperative static feature analysis results; process the intraoperative dynamic data according to the second preset method to obtain the intraoperative dynamic feature analysis results.

[0184] Furthermore, the stage risk assessment data also includes: postoperative risk assessment data; the processing module is also used to: divide the postoperative risk assessment data into postoperative static data and postoperative dynamic data according to the fluctuation characteristics of the data; process the postoperative static data according to the first preset method to obtain the postoperative static feature analysis results; process the postoperative dynamic data according to the second preset method to obtain the postoperative dynamic feature analysis results.

[0185] Furthermore, the stage risk assessment data also includes: post-discharge risk assessment data; the processing module is also used to: divide the post-discharge risk assessment data into post-discharge static data and post-discharge dynamic data according to the fluctuation characteristics of the data; process the post-discharge static data according to the first preset method to obtain the post-discharge static feature analysis results; process the post-discharge dynamic data according to the second preset method to obtain the post-discharge dynamic feature analysis results.

[0186] Furthermore, the processing module is also used to: perform feature amplification on the preoperative static data to obtain amplified preoperative static data; perform feature extraction on the amplified preoperative static data to obtain preoperative static feature extraction results; and perform residual network analysis on the preoperative static feature extraction results to obtain preoperative static feature analysis results.

[0187] Furthermore, the processing module is also used to: extract preoperative spatial and temporal features from preoperative dynamic data; perform feature extraction on the preoperative spatial and temporal features respectively to obtain preoperative spatial feature extraction results and preoperative temporal feature extraction results; perform fully convolutional network analysis on the preoperative spatial feature extraction results to obtain preoperative spatial feature analysis results; perform recurrent gate unit analysis on the preoperative temporal feature extraction results to obtain preoperative temporal feature analysis results; and merge the preoperative spatial feature analysis results and preoperative temporal feature analysis results to obtain preoperative dynamic feature analysis results.

[0188] Furthermore, the processing module is also used to: perform feature amplification on intraoperative static data to obtain amplified intraoperative static data; perform feature extraction on the amplified intraoperative static data to obtain intraoperative static feature extraction results; and perform residual network analysis on the intraoperative static feature extraction results to obtain intraoperative static feature analysis results.

[0189] Furthermore, the processing module is also used to: extract intraoperative spatial and temporal features from intraoperative dynamic data; perform feature extraction on the intraoperative spatial and temporal features respectively to obtain intraoperative spatial feature extraction results and intraoperative temporal feature extraction results; perform fully convolutional network analysis on the intraoperative spatial feature extraction results to obtain intraoperative spatial feature analysis results; perform recurrent gate unit analysis on the intraoperative temporal feature extraction results to obtain intraoperative temporal feature analysis results; and merge the intraoperative spatial feature analysis results and intraoperative temporal feature analysis results to obtain intraoperative dynamic feature analysis results.

[0190] Furthermore, the processing module is also used to: perform feature amplification on postoperative static data to obtain amplified postoperative static data; perform feature extraction on the amplified postoperative static data to obtain postoperative static feature extraction results; and perform residual network analysis on the postoperative static feature extraction results to obtain postoperative static feature analysis results.

[0191] Furthermore, the processing module is also used to: extract postoperative spatial and temporal features from postoperative dynamic data; perform feature extraction on the postoperative spatial and temporal features respectively to obtain postoperative spatial feature extraction results and postoperative temporal feature extraction results; perform fully convolutional network analysis on the postoperative spatial feature extraction results to obtain postoperative spatial feature analysis results; perform recurrent gate unit analysis on the postoperative temporal feature extraction results to obtain postoperative temporal feature analysis results; and merge the postoperative spatial feature analysis results and postoperative temporal feature analysis results to obtain postoperative dynamic feature analysis results.

[0192] Furthermore, the processing module is also used to: perform feature amplification on the post-discharge static data to obtain amplified post-discharge static data; perform feature extraction on the amplified post-discharge static data to obtain post-discharge static feature extraction results; and perform residual network analysis on the post-discharge static feature extraction results to obtain post-discharge static feature analysis results.

[0193] Furthermore, the processing module is also used to: extract the spatial and temporal features of the post-discharge dynamic data; perform feature extraction on the spatial and temporal features respectively to obtain the spatial and temporal feature extraction results; perform fully convolutional network analysis on the spatial feature extraction results to obtain the spatial feature analysis results; perform cyclic gate unit analysis on the temporal feature extraction results to obtain the temporal feature analysis results; and merge the spatial and temporal feature analysis results to obtain the dynamic feature analysis results.

[0194] Furthermore, the output module is also used to: input the preoperative static feature analysis results and the preoperative dynamic feature analysis results into the perioperative risk prediction model, and output the preoperative risk category and the corresponding risk level.

[0195] Furthermore, the output module is also used to: input the intraoperative static feature analysis results and the intraoperative dynamic feature analysis results into the perioperative risk prediction model, and output the intraoperative risk category and the corresponding risk level.

[0196] Furthermore, the output module is also used to input the postoperative static feature analysis results and the postoperative dynamic feature analysis results into the perioperative risk prediction model, and output the postoperative risk category and the corresponding risk level.

[0197] Furthermore, the output module is also used to input the post-discharge static feature analysis results and the post-discharge dynamic feature analysis results into the perioperative risk prediction model, and output the post-discharge risk category and the corresponding risk level.

[0198] Furthermore, the device is also used to: input perioperative risk warning results into a pre-trained intervention plan model and output a target intervention plan for patients to be operated on; wherein the target intervention plan includes: preventive intervention plan and / or treatment intervention plan.

[0199] Furthermore, the device is also used to: acquire anesthesia plans for patients undergoing surgery; display the anesthesia plan, preoperative risk categories, and corresponding risk levels via a first designated device; display the intraoperative risk categories, corresponding risk levels, and intervention plans via a second designated device; display the postoperative risk categories, corresponding risk levels, and intervention plans via a third designated device; and display the post-discharge risk categories, corresponding risk levels, and intervention plans via a fourth designated device.

[0200] Furthermore, the device is also used to: acquire complete medical records of patients undergoing surgery after completing all treatments, new data related to adverse events, actual case report forms, and actual post-discharge follow-up information; optimize the risk factor screening and assessment model based on the complete medical records, actual case report forms, and actual post-discharge follow-up information to obtain a new risk factor screening and assessment model; and optimize the perioperative risk prediction model based on the new data to obtain a new perioperative risk prediction model.

[0201] Furthermore, the device is also used to: obtain the actual intervention plan used for patients undergoing surgery; and optimize the intervention plan model based on the actual intervention plan used to obtain a new intervention plan model.

[0202] Furthermore, the device is also used to: generate anesthesia clinical pathway information according to the target intervention plan; send the anesthesia clinical pathway information to a designated workstation through a first designated interface; and send the associated data of the patient to be operated on to a third-party platform through a second designated interface; wherein the associated data of the patient to be operated on includes at least: the perioperative risk warning results and the target intervention plan of the patient to be operated on.

[0203] The perioperative risk warning device based on machine learning provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned perioperative risk warning method based on machine learning. For the sake of brevity, any parts not mentioned in the embodiment of the perioperative risk warning device based on machine learning can be referred to the corresponding content in the aforementioned perioperative risk warning method based on machine learning.

[0204] This invention also provides an electronic device, see [link to relevant documentation]. Figure 5 As shown, the electronic device includes a processor 130 and a memory 131. The memory 131 stores machine-executable instructions that can be executed by the processor 130. The processor 130 executes the machine-executable instructions to implement the aforementioned perioperative risk warning method based on machine learning.

[0205] Furthermore, Figure 5The electronic device shown also includes a bus 132 and a communication interface 133, with the processor 130, the communication interface 133 and the memory 131 connected via the bus 132.

[0206] The memory 131 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 133 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network. The bus 132 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0207] Processor 130 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 130 or by instructions in software form. Processor 130 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 131, and processor 130 reads the information in memory 131 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0208] This invention also provides a machine-readable storage medium storing machine-executable instructions. When these machine-executable instructions are called and executed by a processor, they cause the processor to implement the aforementioned perioperative risk warning method based on machine learning. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0209] The computer program product of the perioperative risk warning method based on machine learning provided in this embodiment of the invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0210] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0211] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A perioperative risk early warning device based on machine learning, characterized in that, The device includes: The acquisition module is used to acquire assessment parameters associated with perioperative risk assessment for patients awaiting surgery, including: acquiring the initial case corresponding to the patient awaiting surgery; wherein, the initial case includes basic examinations performed on the patient before surgery; the basic examinations include: height, weight, and blood pressure; inputting the initial case into a pre-trained risk factor screening and assessment model, and outputting the assessment parameters associated with perioperative risk assessment for the patient awaiting surgery. The data acquisition module is used to acquire multiple first data points corresponding to the evaluation parameters of the patient to be operated on from a preset data source according to a preset acquisition method; the preset data source includes: electronic medical record system, clinical data center, hospital information system, surgical anesthesia information system, medical record report form, perioperative medical equipment real-time monitoring data, and postoperative follow-up system; The processing module is used to standardize multiple sets of first data according to a preset standardization method to obtain multiple sets of first data after processing; divide the multiple sets of first data according to the perioperative stage to obtain stage risk assessment data; wherein, the stage risk assessment data includes: preoperative risk assessment data, intraoperative risk assessment data, postoperative risk assessment data, or post-discharge risk assessment data; divide the stage risk assessment data into static data and dynamic data according to the fluctuation characteristics of the data; perform feature amplification on the static data to obtain amplified static data; perform feature extraction on the amplified static data to obtain static feature extraction results; perform residual network analysis on the static feature extraction results to obtain static feature analysis results; extract the spatial and temporal features of the dynamic data, and perform feature extraction on the spatial and temporal features respectively to obtain spatial feature extraction results and temporal feature extraction results; perform fully convolutional network analysis on the spatial feature extraction results to obtain spatial feature analysis results; perform recurrent gate unit analysis on the temporal feature extraction results to obtain temporal feature analysis results; and merge the spatial feature analysis results and the temporal feature analysis results to obtain dynamic feature analysis results. The output module is used to input the static feature analysis results and the dynamic feature analysis results into the pre-trained perioperative risk prediction model and output the perioperative risk warning results corresponding to the patient to be operated on; wherein, the perioperative risk warning results include at least one of the following: preoperative risk category and corresponding risk level, intraoperative risk category and corresponding risk level, postoperative risk category and corresponding risk level, and post-discharge risk category and corresponding risk level.

2. The apparatus according to claim 1, characterized in that, The device further includes: The intervention module is used to input the perioperative risk warning results into a pre-trained intervention plan model and output a target intervention plan for the patient to be operated on; wherein, the target intervention plan includes: preventive intervention plan and / or treatment intervention plan.

3. The apparatus according to claim 2, characterized in that, The device further includes: The display module is used to acquire the anesthesia plan for the patient to be operated on; to display the anesthesia plan, the preoperative risk category and the corresponding risk level through a first designated device; to display the intraoperative risk category, the corresponding risk level and intervention plan through a second designated device; to display the postoperative risk category, the corresponding risk level and intervention plan through a third designated device; and to display the post-discharge risk category, the corresponding risk level and intervention plan through a fourth designated device.

4. The apparatus according to claim 1, characterized in that, The device further includes: The model optimization module is used to acquire complete medical records of the patient undergoing surgery after all treatments have been completed, new data related to adverse events, actual case report forms, and actual follow-up information after discharge. Based on the complete medical records, the actual case report forms, and the actual follow-up information after discharge, the risk factor screening and assessment model is optimized to obtain a new risk factor screening and assessment model. Based on the new data, the perioperative risk prediction model is optimized to obtain a new perioperative risk prediction model.

5. The apparatus according to claim 2, characterized in that, The intervention module is also used to obtain the actual intervention plan used for the patient to be operated on; based on the actual intervention plan used, the intervention plan model is optimized to obtain a new intervention plan model.

6. The apparatus according to claim 2, characterized in that, The device further includes: The information sending module is used to generate anesthesia clinical pathway information according to the target intervention plan; send the anesthesia clinical pathway information to a designated workstation through a first designated interface; and send the associated data of the patient to be operated on to a third-party platform through a second designated interface; wherein the associated data of the patient to be operated on includes at least: the perioperative risk warning result of the patient to be operated on and the target intervention plan.

7. An electronic device, characterized in that, It includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to achieve the function of the perioperative risk warning device based on machine learning as described in any one of claims 1-6.

8. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores machine-executable instructions that, when invoked and executed by a processor, cause the processor to perform the functions of the perioperative risk warning device based on machine learning as described in any one of claims 1-6.

Images