A perioperative risk grading model construction method and a perioperative risk grading method

Abstract

The present application relates to the field of clinical medicine technology, and provides a perioperative risk grading model construction method and a perioperative risk grading method, wherein the model construction method includes: obtaining multiple groups of historical grading data, the historical grading data including the collection information, grading result information and perioperative real state information of the measured person; extracting the consistency degree of the grading result information and the perioperative real state information; dividing the multiple groups of historical grading data into positive sample sets and negative sample sets based on the consistency degree and inputting them into the initial grading model, calculating the loss function value according to the positive sample grading result and the negative sample grading result, and adjusting the model parameters through the optimization algorithm to minimize the loss function value, and obtain a trained grading model. The grading model learns the correspondence between the collection information and the correct grading result in the positive sample during the training process, and can distinguish the correspondence between the collection information and the grading result with large deviation, thereby achieving accurate grading.

Description

Technical Field

[0001] The present application relates to the field of clinical medical technology, and in particular to a perioperative risk grading model construction method and a perioperative risk grading method. Background Art

[0002] More than 200 million patients worldwide undergo surgery every year, and the medical issues facing this population are becoming increasingly complex. Many adverse events in hospitalized patients are related to surgery and operations, and nearly half of perioperative adverse complications are completely preventable. Therefore, it is crucial to reduce the risk of perioperative adverse complications as much as possible. Preoperative risk assessment of patients and implementation of graded management are effective measures to control risks and improve anesthesia safety. The American Society of Anesthesiologists Physical Status (ASA-PS) grading system is one of the most commonly used preoperative assessment tools.

[0003] However, the ASA-PS classification is too subjective and simple. Therefore, in studies on adults, different physicians have poor consistency in grading the same patient. It lacks precise definitions and is not quantifiable enough: the lack of precise definitions for each grade category leads to ambiguity in the actual application of the ASA-PS classification.

[0004] In addition, perioperative risks are multifaceted. For example, the patient's airway must be unobstructed and effectively ventilated during the perioperative period, and endotracheal intubation is one of the important means for anesthesiologists to manage the airway, and difficult endotracheal intubation is life-threatening. Therefore, it is crucial to conduct a preoperative airway assessment before anesthesia. For another example, patients with cardiovascular disease are very common during routine anesthesia visits. Patients with cardiovascular disease have a 25% to 50% higher incidence of cardiovascular complications after non-cardiac surgery. Therefore, for patients with concomitant cardiovascular disease, in order to reduce the occurrence of postoperative adverse events (including death), the patient's cardiac function rating is also an important part of preoperative evaluation and risk prediction.

[0005] However, the above-mentioned risk grading process is often based on some relatively vague grading tables and relies on the physician's experience to perform manual risk grading, which has the problems of strong subjectivity and poor objectivity, and is not effective in predicting perioperative risks.

[0006] Therefore, how to objectively and accurately perform perioperative risk stratification has become a technical problem that needs to be solved urgently. Summary of the invention

[0007] In view of this, the embodiments of the present disclosure provide a perioperative risk grading model construction method and a perioperative risk grading method to solve the technical problem of how to perform perioperative risk grading more objectively and accurately in the prior art.

[0008] According to a first aspect of an embodiment of the present application, a method for constructing a perioperative risk grading model is provided, comprising: obtaining multiple groups of historical grading data, wherein the historical grading data include collection information, grading result information and perioperative real status information of the measured person; wherein the collection information includes basic vital signs information, combined disease information and organ function information; extracting the degree of consistency between the grading result information and the perioperative real status information; based on the degree of consistency, dividing the multiple groups of historical grading data into a positive sample set and a negative sample set, wherein the positive sample set is a sample set constructed from historical grading data whose degree of consistency between the grading result information and the perioperative real status information is greater than a preset value, and the negative sample set is a sample set constructed from historical grading data whose degree of consistency between the grading result information and the perioperative real status information is less than a preset value; inputting the positive sample set and the negative sample set into an initial grading model, calculating the loss function value according to the positive sample grading results and the negative sample grading results, and adjusting the model parameters through an optimization algorithm to minimize the loss function value to obtain a trained grading model.

[0009] Optionally, the grading result information includes ASA-PS grade, cardiac function grade and airway grade; the extraction of the consistency between the grading result information and the perioperative real status information includes: obtaining the preoperative ASA-PS grade result, cardiac function grade result and airway grade result of the measured person; determining the perioperative assessment status information based on the ASA-PS grade result, cardiac function grade result and airway grade result, the perioperative assessment status information including preoperative anesthesia assessment status information, surgical process assessment status information and postoperative assessment status information; obtaining the perioperative real status information of the same measured person, the perioperative real status information including preoperative anesthesia real status information, surgical process real status information and postoperative real status information; comparing the perioperative assessment status information with the perioperative real status information to obtain the consistency between the grading result information and the perioperative real status information.

[0010] Optionally, comparing the perioperative assessment status information with the perioperative real status information to obtain the degree of consistency between the grading result information and the perioperative real status information includes: respectively extracting the first key indicator from the preoperative anesthesia assessment status information, the surgical process assessment status information and the postoperative assessment status information; scoring the first key indicator and constructing a first key indicator scoring matrix; respectively extracting the second key indicator from the preoperative anesthesia real status information, the surgical process real status information and the postoperative real status information; scoring the second key indicator and constructing a second key indicator scoring matrix; calculating the similarity between the first key indicator scoring matrix and the second key indicator scoring matrix as the degree of consistency.

[0011] Optionally, the grading model includes: a cascaded feature extraction module and a multi-head attention mechanism layer, and an airway grading module, a cardiac function grading module and an ASA-PS grading module cascaded in parallel after the multi-head attention mechanism layer; the inputting of the positive sample set and the negative sample set into the initial grading model includes: respectively inputting the positive sample set and the negative sample set into the feature extraction module, respectively extracting the dynamic features and the static features of the basic vital sign information, the combined disease information and the organ function information in each set of historical grading data; in the multi-head attention mechanism layer, the dynamic features and the static features are adaptively weighted and matched to the airway grading module, the cardiac function grading module and the ASA-PS grading module input layer.

[0012] Optionally, the feature extraction module includes a physical sign entity feature extraction unit, an image feature extraction unit and a time series feature extraction unit; the step of respectively inputting the positive sample set and the negative sample set into the feature extraction module includes: inputting the acquisition information into the feature extraction module, extracting physical sign entity features using the physical sign entity feature extraction unit, extracting airway image features using the image feature extraction unit, and using the physical sign entity features and the airway image features as the static features; respectively inputting the physical sign entity features and the airway image features into the time series feature extraction unit, and extracting the time series feature sequence and time series change features of the static features as the dynamic features.

[0013] Optionally, the multi-head attention mechanism layer adaptively weights and matches the dynamic features and the static features to the airway grading module, cardiac function grading module and ASA-PS grading module input layer, including: constructing one-to-one corresponding attention matrices based on the airway grading module, cardiac function grading module and ASA-PS grading module respectively; calculating the dot product of the static features and the dynamic features with the attention matrices respectively, to obtain the attention weights of the static features and the dynamic features in the airway grading module, cardiac function grading module and ASA-PS grading module respectively; and combining the static features and the dynamic features based on the attention weights to construct the input features of the airway grading module, cardiac function grading module and ASA-PS grading module respectively.

[0014] Optionally, before extracting the consistency degree between the grading result information and the perioperative real status information, it also includes: extracting the corresponding personnel type of each historical grading data respectively, and the personnel types include: pregnant women, children and the elderly; marking the historical grading data with personnel types based on the personnel types; and adjusting the model parameters of the grading model using the personnel type marking to learn the correspondence between the personnel type and the grading results.

[0015] According to the second aspect, an embodiment of the present application provides a perioperative risk grading method, comprising: obtaining collected information of the person to be tested, the collected information including basic vital sign information, combined disease information and organ function information; inputting the collected information into a grading model constructed by the perioperative risk grading model construction method described in any one of the first aspects above, to obtain a grading result of the person to be tested.

[0016] Optionally, the grading results include the ASA-PS grading result, the cardiac function grading result and the airway grading result of the person to be tested.

[0017] According to the third aspect, an embodiment of the present application also provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, and the memory is used to store computer programs; the processor is used to execute the perioperative risk grading model construction method described in any one of the first aspect and / or the perioperative risk grading method described in the second aspect by running the computer program stored in the memory.

[0018] According to the fourth aspect, an embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the perioperative risk grading model construction method described in any one of the first aspect and / or the perioperative risk grading method described in the second aspect when run.

[0019] Compared with the prior art, the embodiments of the present invention have at least the following beneficial effects:

[0020] In the present application, by collecting the historical grading data of each patient who has completed the operation during the perioperative period, according to the consistency between the grading results in the grading data and the actual state of the perioperative period, by quantifying the grading results and the actual state of the perioperative period, the historical grading data is divided into positive samples representing more accurate grading and negative samples with larger grading deviations. By dividing the positive samples and negative samples, the grading model can learn the corresponding relationship between the collected information in the positive samples and the correct grading results during the training process. At the same time, through the learning of negative samples, the grading model can distinguish the corresponding relationship between the collected information and the grading results with larger deviations, especially for the positive samples and negative samples corresponding to similar collected data in different groups of historical grading data, that is, through the learning of positive samples and negative samples, the grading model can learn the logical relationship between similar collected data and different grading results, and then when grading again, it can perform more accurate grading on the same or similar collected data. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 It is a flowchart of a method for constructing a perioperative risk grading model provided in an embodiment of the present application;

[0023] Figure 2 It is a structural schematic diagram of a hierarchical model provided in an embodiment of the present application;

[0024] Figure 3 It is a flowchart of a perioperative risk grading method provided in an embodiment of the present application;

[0025] Figure 4 It is a structural schematic diagram of a perioperative risk grading model construction device provided in an embodiment of the present application;

[0026] Figure 5 It is a structural schematic diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.

[0028] See also Figure 1 , the present application provides a method for constructing a perioperative risk grading model, which specifically includes the following steps:

[0029] S10. Obtain multiple groups of historical grading data, the historical grading data including the collected information, grading result information and perioperative real state information of the measured person; wherein the collected information includes basic physical sign information, combined disease information and organ function information. In one embodiment, the grading data of each patient who has completed the operation in the perioperative period can be collected, and the grading data can include the collected information of the measured person, for example, the basic physical sign information, combined disease information and organ function information before the operation. And the corresponding grading results and perioperative real state of the measured person, often the grading results and perioperative real state have a corresponding relationship. The perioperative real state information can include various indicator information of the measured person during the anesthesia process, for example, basic physiological indicators such as electrocardiogram, blood pressure, blood oxygen saturation, body temperature, and deep physiological indicators such as central venous pressure, depth of anesthesia, coagulation function, neuromuscular conduction function, and neuroelectrophysiology. The perioperative real status information may also include various index information during the postoperative anesthesia recovery process, such as basic physiological indicators such as electrocardiogram, blood pressure, blood oxygen saturation, body temperature, recovery time, and the physiological status of the person being tested after recovery, such as postoperative nausea and vomiting, postoperative chills, postoperative respiratory obstruction, etc. The perioperative real status information may also include the status of complications after anesthesia or surgery.

[0030] In one embodiment, the basic physical sign information before the operation may include health status, smoking status, drinking status, body mass index, airway examination information, respiratory status, vital sign parameters, consciousness status, pregnancy status, heart function related examination information, etc. The combined disease information may include the oral medical history and case of the person being tested, the historical medical history and case information extracted from the historical medical records of the person being tested, and the medical history and case information judged by the physician, etc. The organ function information may be an extraction of the laboratory test report of the person being tested in the hospital, and the key indicators include: heart valve orifice area, regurgitation area, cardiac ejection fraction, glomerular filtration rate, etc.

[0031] In historical grading data, doctors often grade the subjects based on their collected information and customized grading forms. At the same time, after the subjects enter the operation, their real conditions before, during and after the operation are collected. The collected information of the subjects before the operation, the grading results based on the collected information, and the real conditions of the subjects before, during and after the operation are used as a set of historical grading data.

[0032] S20. Extract the consistency between the grading result information and the perioperative true status information. In one embodiment, since the historical grading data is often graded by subjective judgment, there may be some groups where the grading results are positively correlated with the perioperative true status, and there may also be some groups where the grading results are greatly different from the perioperative true status. In this embodiment, the accuracy of the grading in multiple groups of historical grading data is used to construct training samples for the grading model. Therefore, the consistency of each historical grading data is evaluated separately through the correlation between the existing grading result information and the perioperative true status corresponding to the grading result information.

[0033] In one embodiment, the preoperative ASA-PS grading result, cardiac function grading result and airway grading result of the measured person can be obtained respectively; based on the ASA-PS grading result, cardiac function grading result and airway grading result, perioperative assessment status information is determined, and the perioperative assessment status information includes preoperative anesthesia assessment status information, surgical process assessment status information and postoperative assessment status information; the perioperative real status information of the same measured person is obtained, and the perioperative real status information includes preoperative anesthesia real status information, surgical process real status information and postoperative real status information; the perioperative assessment status information and the perioperative real status information are compared to obtain the consistency degree between the grading result information and the perioperative real status information.

[0034] ASA-PS grading results are usually divided into six levels, among which:

[0035] Level 1: Healthy people without organic, biochemical or psychological diseases. Perioperative mortality rate 0.0.6~0.08%;

[0036] Level 2: Mild systemic diseases, such as mild asthma or well-controlled hypertension. No serious impact on daily life. No impact on anesthesia and surgery. Perioperative mortality rate 0.27~0.40%;

[0037] Grade 3: Severe systemic disease restricts normal activities, such as renal failure requiring dialysis or grade 2 congestive heart failure. Significantly affects daily life. Anesthesia and surgery are likely to be affected. Perioperative mortality rate is 1.82~4.30%;

[0038] Level 4: Severe illness, life-threatening or requiring intensive treatment, such as acute myocardial infarction, respiratory failure requiring mechanical ventilation. Severe limitation of daily activities. Significant impact on anesthesia and surgery. Perioperative mortality rate 7.80~23.0%;

[0039] Level 5: Critically ill patients who will die within 24 hours regardless of surgery. Perioperative mortality rate: 9.40-50.7%;

[0040] Level 6: Confirmed to be brain dead, and the organs are intended for organ transplantation.

[0041] For each grading result, there is corresponding status information that may occur in the perioperative period of the measured person. For example, there is a certain risk for grade 3 patients to receive anesthesia. They need to be fully prepared before anesthesia, and the selection of anesthetic drugs should be very cautious. Relevant monitoring measures need to be taken during anesthesia, and relevant measures should be taken to actively prevent complications that may occur during and after anesthesia. The anesthesia risk of grade 4 and grade 5 patients is extremely high, their tolerance to anesthesia is extremely poor, and they are threatened with death at any time. Anesthesia and surgery are extremely dangerous. In this embodiment, the preoperative anesthesia assessment status information, surgical process assessment status information, and postoperative assessment status information corresponding to ASA-PS can be listed in advance through the grading results.

[0042] In addition, the cardiac function classification results are usually divided into four levels, which can be NYHA cardiac function classification or Killip cardiac function classification for heart failure patients. Similarly, the preoperative anesthesia assessment status information, surgical process assessment status information and postoperative assessment status information corresponding to the cardiac function classification are listed for different classification results.

[0043] The airway grading may be a modified Mallampati grading, for example, may include 1st level: visible soft palate, pharyngeal cavity, uvula, and pharyngeal palate arch; 2nd level: visible soft palate, pharyngeal cavity, and uvula; 3rd level: only soft palate and uvula base are visible; 4th level: no soft palate is visible. For different airway grading results, corresponding preoperative anesthesia assessment status information, surgical process assessment status information, and postoperative assessment status information are also listed.

[0044] In one embodiment, the perioperative assessment status information corresponding to the ASA-PS grading results, the cardiac function grading results, and the airway grading results can be fused to obtain fused perioperative assessment status information; or, based on the perioperative assessment status information corresponding to the ASA-PS grading results, the cardiac function grading results, and the airway grading results, the key indicators in the preoperative anesthesia assessment status information, the surgical process assessment status information, and the postoperative assessment status information corresponding to each grading result can be obtained in the above embodiment to construct an estimated indicator table. In one embodiment, each estimated indicator can be coded or scored according to its severity or the corresponding grading result level as the perioperative assessment status quantitative data.

[0045] At the same time, actual status indicators are extracted from the perioperative result status, for example, various indicator information of the person being tested during the anesthesia process, such as basic physiological indicators such as electrocardiogram, blood pressure, blood oxygen saturation, body temperature, and deep physiological indicators such as central venous pressure, depth of anesthesia, coagulation function, neuromuscular conduction function, and neuroelectrophysiology. Various indicator information during the anesthesia recovery process after surgery, such as basic physiological indicators such as electrocardiogram, blood pressure, blood oxygen saturation, body temperature, recovery time, physiological state of the person being tested after recovery, such as postoperative nausea and vomiting, postoperative chills, postoperative respiratory obstruction, and complications during or after anesthesia, are used as actual indicators extracted from the perioperative result status, and an actual indicator table is constructed based on the actual indicators. In one embodiment, the encoding or scoring is performed with the same standard as the estimated indicators as the quantitative data of the true state of the perioperative period.

[0046] When calculating the consistency degree, the similarity between the two indicator tables can be calculated as the consistency degree by the number of indicators in the estimated indicator table and the actual indicator table and / or the score of each indicator. In this embodiment, an indicator matrix can be constructed based on the two indicator tables respectively, and the similarity between the two indicator matrices can be calculated.

[0047] S30. Based on the consistency level, multiple groups of historical grading data are divided into positive sample sets and negative sample sets. The positive sample set is a sample set constructed from historical grading data whose consistency level between the grading result and the perioperative true status information is greater than a preset value, and the negative sample set is a sample set constructed from historical grading data whose consistency level between the grading result information and the perioperative true status information is less than a preset value. In one embodiment, since grading is often performed by subjective judgment of physicians, there are some relatively accurate grading results in the historical grading data, and there are also some inaccurate grading results. Therefore, in this embodiment, by separating the accurate and inaccurate grading results, positive samples and negative samples are constructed respectively. Through the constructed positive and negative samples, the model training process can simultaneously learn the correspondence between the collected information in the positive sample and the correct grading result and the correspondence between the negative sample and the grading result with a large deviation. Therefore, the model can more accurately perform automatic grading based on the collected information.

[0048] S40. Input the positive sample set and the negative sample set into the initial classification model, calculate the loss function value according to the positive sample classification results and the negative sample classification results, and adjust the model parameters through the optimization algorithm to minimize the loss function value to obtain a trained classification model. In this embodiment, during the training process, the classification model will be exposed to the data of positive samples and negative samples at the same time.

[0049] The classification model learns the correspondence between the collected information and the correct classification results in the positive samples. At the same time, through the learning of negative samples, the classification model can distinguish the correspondence between the collected information and the classification results with large deviations, especially for the positive and negative samples corresponding to similar collected data in different groups of historical classification data. That is, through the learning of positive and negative samples, the classification model can learn the logical relationship between similar collected data and different classification results. Then, when classifying, it can classify the same or similar collected data more accurately, thereby reducing misjudgment. During the training process, the value of the loss function is calculated based on the prediction results of the positive and negative samples. The loss function measures the difference between the model prediction results and the true label (positive sample or negative sample). Through the optimization algorithm, such as stochastic gradient descent, the parameters of the model are adjusted to minimize the loss function value, thereby improving the classification accuracy of the model for positive and negative samples. Afterwards, an independent test set containing positive and negative samples can be used. Indicators such as accuracy, recall, and F1 value are calculated to evaluate the performance of the classification model in distinguishing positive and negative samples.

[0050] In the present application, by collecting the historical grading data of each patient who has completed the operation during the perioperative period, according to the consistency between the grading results in the grading data and the actual state of the perioperative period, by quantifying the grading results and the actual state of the perioperative period, the historical grading data is divided into positive samples representing more accurate grading and negative samples with larger grading deviations. By dividing the positive samples and negative samples, the grading model can learn the corresponding relationship between the collected information in the positive samples and the correct grading results during the training process. At the same time, through the learning of negative samples, the grading model can distinguish the corresponding relationship between the collected information and the grading results with larger deviations, especially for the positive samples and negative samples corresponding to similar collected data in different groups of historical grading data, that is, through the learning of positive samples and negative samples, the grading model can learn the logical relationship between similar collected data and different grading results, and then when grading again, it can perform more accurate grading on the same or similar collected data.

[0051] In one embodiment, in order to more accurately quantify the consistency between the grading result information and the perioperative true status information, in this embodiment, the key indicators in the perioperative assessment status and the perioperative true status can be extracted respectively, and the key indicators can be scored according to the same standard to obtain scoring data. The consistency between the grading result information and the perioperative true status information is calculated through the scoring data to obtain a more objective positive and negative sample division result.

[0052] In this embodiment, when comparing the perioperative assessment status information with the perioperative real status information, obtaining the consistency degree between the grading result information and the perioperative real status information may include:

[0053] The first key indicators are extracted from the preoperative anesthesia evaluation status information, the surgical process evaluation status information, and the postoperative evaluation status information respectively; the first key indicators may include the basic physiological indicators, deep physiological indicators, and complications of the patient during the anesthesia process, the surgical process, and the postoperative period in the above-mentioned embodiment. In this embodiment, the first key indicators that may appear and the evaluation status of the first key indicators are listed through the scoring results to obtain a set of the first key indicators.

[0054] The first key indicators are scored and a first key indicator scoring matrix is ​​constructed. In this embodiment, the first key indicators can be scored according to the degree to which they deviate from the corresponding normal values. For example, some basic physiological indicators can be scored by analyzing the proportion or value of the estimated values ​​of the normal sorting indicators that may deviate from the normal values ​​through the grading results as the scores of the corresponding indicators. All first key indicators are traversed to obtain the first key indicator scoring set and construct the first key indicator scoring matrix.

[0055] The second key indicator is extracted from the preoperative anesthesia real state information, the surgical process real state information and the postoperative real state information respectively. The second key indicator is extracted according to the actual collected perioperative real state. In this embodiment, the second key indicator can correspond to the first key indicator one by one.

[0056] The second key indicator is scored and a second key indicator scoring matrix is ​​constructed. The scoring method of the second key indicator can be unified with the method of the first key indicator to obtain the second key indicator scoring matrix.

[0057] The similarity between the first key indicator scoring matrix and the second key indicator scoring matrix is ​​calculated as the consistency degree. In this embodiment, the first key indicator and the second key indicator can be compared to obtain a comparison result of the first key indicator and the second key indicator. In this embodiment, the comparison method can be the difference between the scores of the two, the sum of the scores of the two, or the deviation of the scores of the two from the standard score. After obtaining the comparison result, a comparison result judgment matrix can be constructed to determine the similarity between the first key indicator scoring matrix and the second key indicator scoring matrix. For example, the similarity A can be calculated using the following judgment matrix:

[0058] ,

[0059] Among them, a ij represents the result of comparing the i-th first key indicator score with the j-th second key indicator score, and A is the similarity between the first key indicator score matrix and the second key indicator score matrix.

[0060] In one embodiment, if Figure 2As shown, the grading model includes: a cascaded feature extraction module and a multi-head attention mechanism layer, and an airway grading module, a cardiac function grading module and an ASA-PS grading module cascaded in parallel after the multi-head attention mechanism layer. In this embodiment, the features of the collected data in the positive samples and negative samples are extracted by the feature extraction model. Since the types of features concerned by the airway grading module, the cardiac function grading module and the ASA-PS grading module are different, and the input features required by the airway grading module, the cardiac function grading module and the ASA-PS grading module have a certain influence on other grading modules or recognition results, for example, the organ function information required for the cardiac function grading plays a relatively important role in the ASA-PS grading process. Similarly, the throat detection information required for the airway grading also plays a relatively important role in the ASA-PS grading process. The basic signs required for the ASA-PS grading, such as health data and comorbidities, also have a certain influence on the airway grading module and the cardiac function grading module. Therefore, in this embodiment, it is necessary to fuse all the features extracted from the collected information in order to improve the grading accuracy of each airway grading module, the cardiac function grading module and the ASA-PS grading module. At the same time, since the features concerned by each grading module are different, in this embodiment, the extracted features are adaptively weighted and matched to the airway grading module, the cardiac function grading module and the ASA-PS grading module input layer through the multi-head attention mechanism. By enabling each grading module to learn the features of all collected information at the same time, it can focus on the more important features for each grading type and improve the accuracy of grading results of all grading types.

[0061] In one embodiment, the positive sample set and the negative sample set are respectively input into the feature extraction module, and the dynamic features and static features of the basic sign information, the combined disease information and the organ function information in each set of historical hierarchical data are respectively extracted. Since the collected information includes multiple types of data, such as case data, oral condition data, health status data, test data, and image data such as airway image data and organ medical images. At the same time, the collected data may also include static data, such as oral data, test data, and image data such as airway image data and organ medical images during evaluation, and may also include some dynamic data, such as historical case data and historical case data combined with static data collected before evaluation, which constitute dynamic data in time series to characterize the changing characteristics of the condition, symptoms, health, etc. of the tested person over time.

[0062] Based on this, in this embodiment, static data and dynamic data in the collected information can be extracted separately, wherein the feature extraction module includes a physical sign entity feature extraction unit, an image feature extraction unit and a time series feature extraction unit. The collected information is input into the feature extraction module, the physical sign entity feature is extracted by the physical sign entity feature extraction unit, and the airway image feature is extracted by the image feature extraction unit, and the physical sign entity feature and the airway image feature are used as the static feature; the physical sign entity feature and the airway image feature are respectively input into the time series feature extraction unit, and the time series feature sequence and time series change feature of the static feature are extracted as the dynamic feature. In this embodiment, the collected information can be divided into three categories, the first is text type collected data, such as case data, oral data, inspection data, etc., the second is image data, such as airway image data, organ medical images, etc., and the third is the change data of text type data and image data in time series. In this embodiment, for text feature extraction, text feature extraction modules such as Word2Vec, Glove, Fast text, etc. can be used, and for image feature extraction, convolutional neural network can be used for extraction. After obtaining static features such as text features and image features, the time series change features of static features with time series changes can be extracted through a recurrent neural network, especially for features related to cardiac function, which often have strong time series correlation and time series accumulation. Therefore, by extracting the corresponding case features, influence features, test and examination features, and symptom features of the time series change characteristics, the cardiac function of the current patient being tested can be evaluated more accurately.

[0063] In an exemplary embodiment, the temporal feature extraction unit may use a gated recurrent unit (GRU) for extraction. During the learning process, the temporal cumulative change features of features related to cardiac function are extracted through the GRU layer, which can not only better capture the correlation in the sequence, but also capture various dependencies within each feature sequence.

[0064] In this embodiment, the GRU has an update gate and a reset gate. The gate control unit can control the flow of information, determine which information should be passed and which information should be discarded, and can regulate the interaction between input data and previous hidden states. By using the gate control unit, the GRU can effectively transfer and update information between different time steps, and can better capture the cumulative effect of multiple types of static feature sequence data in time, that is, capture the cumulative change characteristics of multiple static features coupled in time. In turn, it can better capture the correlation in the sequence and various dependencies within each feature sequence.

[0065] In one embodiment, the multi-head attention mechanism layer adaptively weights and matches the dynamic features and the static features to the airway grading module, the cardiac function grading module and the ASA-PS grading module input layer, including: constructing a one-to-one corresponding attention matrix based on the airway grading module, the cardiac function grading module and the ASA-PS grading module respectively; calculating the dot product of the static features and the dynamic features with the attention matrix respectively, and obtaining the attention weights of the static features and the dynamic features in the airway grading module, the cardiac function grading module and the ASA-PS grading module respectively; combining the static features and the dynamic features based on the attention weights to construct the input features of the airway grading module, the cardiac function grading module and the ASA-PS grading module respectively.

[0066] In this embodiment, the airway grading module, the cardiac function grading module and the ASA-PS grading module are respectively used as the heads of the multi-head attention mechanism. An independent linear transformation matrix of query (Q), key (K) and value (V) can be created for each head. For each head, the input features are multiplied by the corresponding Q and K matrices, and the dot product is calculated and scaled. The scaled result is normalized by the softmax function to obtain the attention score. The attention score is multiplied by the corresponding V value and summed to obtain the output features of each head. The output features of the three heads can be connected or combined in some way. The combined features are respectively assigned to the airway grading module, the cardiac function grading module and the ASA-PS grading module for subsequent processing and learning.

[0067] In one embodiment, the airway grading module can use the yolov4-tiny network model for real-time recognition. For example, the collected information may include front, side and maximum mouth opening images of the face, as well as airway-related images such as Mallampati grading images. The collected images are preprocessed, for example, face detection and alignment, feature point detection, spatial rotation angle compensation, etc., and a target recognition algorithm is used to predict dental abnormalities and Mallampati grading, and the yolov4-tiny network model is used for real-time recognition.

[0068] The cardiac function grading module and the ASA-PS grading module can be graded using grading models constructed using convolutional neural networks and recurrent neural networks, respectively.

[0069] In one embodiment, current ratings are often performed on ordinary adults. However, the physical conditions, tolerance, etc. of special groups such as children, pregnant women, and the elderly are significantly different from those of ordinary adults. Therefore, in this implementation, the corresponding person type of each of the historical grading data is extracted separately, and the person types include: pregnant women, children, and the elderly; the historical grading data is labeled with the person type based on the person type; the model parameters of the grading model are adjusted using the person type label to learn the correspondence between the person type and the grading result.

[0070] In this implementation, for pregnant women, children and the elderly, the corresponding content can be adaptively collected, and when the model is trained, the classification results are adjusted based on the pregnant women, children and the elderly. For example, when the tested person is pregnant women, children and the elderly, the classification result can be automatically increased by one level. This achieves advantages such as more comprehensive information collection, more intelligent data analysis and processing, and more intuitive and easy-to-understand output results.

[0071] This application also provides a perioperative risk grading method, such as Figure 3 As shown, including:

[0072] S100. Acquire the collected information of the person to be tested, the collected information including basic physical sign information, combined disease information and organ function information. The collected information can refer to the collected information in the historical classification data in the above implementation. In one embodiment, when collecting information, the corresponding person can also be collected adaptively according to the type of person.

[0073] S200. Input the collected information into a grading model constructed by a perioperative risk grading model construction method to obtain a grading result of the person to be tested.

[0074] The grading results include the ASA-PS grading results, cardiac function grading results and airway grading results of the person being tested.

[0075] After obtaining the grading results, the perioperative risk can be comprehensively predicted based on the ASA-PS grading results, cardiac function grading results and airway grading results.

[0076] It can collect patient information comprehensively and accurately, conduct intelligent analysis through deep learning, and output intuitive and easy-to-understand grading results, greatly improving medical efficiency and accuracy.

[0077] All the above optional technical solutions can be combined in any way to form optional embodiments of the present disclosure. Here, the following is an embodiment of the device of the present disclosure, which can be used to execute the method embodiment of the present disclosure. For details not disclosed in the device embodiment of the present disclosure, please refer to the method embodiment of the present disclosure.

[0078] Figure 4 Schematic diagram of a perioperative risk grading model construction device provided by an embodiment of the present disclosure. Figure 4 As shown, the device comprises:

[0079] The acquisition module 41 is used to acquire multiple groups of historical classification data, wherein the historical classification data includes the collection information, classification result information and perioperative real status information of the measured person; wherein the collection information includes basic vital sign information, combined disease information and organ function information;

[0080] An extraction module 42, used for extracting the consistency between the grading result information and the perioperative real status information;

[0081] A data division module 43 is used to divide the multiple groups of historical grading data into a positive sample set and a negative sample set based on the consistency degree, wherein the positive sample set is a sample set constructed from historical grading data whose consistency degree between the grading result and the perioperative true state information is greater than a preset value, and the negative sample set is a sample set constructed from historical grading data whose consistency degree between the grading result information and the perioperative true state information is less than a preset value;

[0082] The training module 44 is used to input the positive sample set and the negative sample set into the initial classification model, calculate the loss function value according to the positive sample classification results and the negative sample classification results, and adjust the model parameters through the optimization algorithm to minimize the loss function value to obtain a trained classification model.

[0083] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.

[0084] Figure 5 Schematic diagram of an electronic device 5 provided in an embodiment of the present disclosure. Figure 5 As shown, the electronic device 5 of this embodiment includes: a processor 501, a memory 502, and a computer program 503 stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program 503, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 501 executes the computer program 503, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0085] Exemplarily, the computer program 503 may be divided into one or more modules / units, which are stored in the memory 502 and executed by the processor 501 to complete the present disclosure. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 503 in the electronic device 5.

[0086] The electronic device 3 may be a terminal, a gateway, a cloud server or other electronic device. The electronic device 5 may include but is not limited to a processor 501 and a memory 502. Those skilled in the art will appreciate that Figure 5 It is only an example of the electronic device 5 and does not constitute a limitation of the electronic device 5. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0087] The processor 501 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0088] The memory 502 may be an internal storage unit of the electronic device 3, for example, a hard disk or memory of the electronic device 3. The memory 502 may also be an external storage device of the electronic device 3, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 3. Further, the memory 502 may also include both an internal storage unit and an external storage device of the electronic device 3. The memory 502 is used to store computer programs and other programs and data required by the electronic device. The memory 502 may also be used to temporarily store data that has been output or is to be output.

[0089] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present disclosure. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0090] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0091] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0092] In the embodiments provided in the present disclosure, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0093] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0094] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0095] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0096] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should all be included in the protection scope of the present disclosure.

Claims

1. A method for constructing a perioperative risk grading model, characterized by: Acquire multiple groups of historical grading data, wherein the historical grading data includes the collected information, grading result information and perioperative real status information of the measured person; wherein the collected information includes basic vital signs information, combined disease information and organ function information; the grading result information includes ASA-PS grading, cardiac function grading and airway grading; Extracting the consistency degree between the grading result information and the perioperative true status information, including: obtaining the preoperative ASA-PS grading result, cardiac function grading result and airway grading result of the measured person; determining the perioperative assessment status information based on the ASA-PS grading result, the cardiac function grading result and the airway grading result, the perioperative assessment status information including preoperative anesthesia assessment status information, surgical process assessment status information and postoperative assessment status information; obtaining the perioperative true status information of the same measured person, the perioperative true status information including preoperative anesthesia true status information, surgical process true status information and postoperative true status information; comparing the perioperative assessment status information with the perioperative true status information to obtain the consistency degree between the grading result information and the perioperative true status information, specifically including: extracting the preoperative anesthesia assessment status information, the surgical process true status information and the postoperative true status information respectively. Evaluation status information and a first key indicator in the postoperative evaluation status information; scoring the first key indicator and constructing a first key indicator scoring matrix; respectively extracting second key indicators in the preoperative anesthesia real state information, the surgical process real state information and the postoperative real state information; scoring the second key indicator and constructing a second key indicator scoring matrix; calculating the similarity between the first key indicator scoring matrix and the second key indicator scoring matrix as the consistency degree; based on the consistency degree, dividing multiple groups of historical graded data into positive sample sets and negative sample sets, wherein the positive sample set is a sample set constructed from historical graded data whose consistency degree between the graded result information and the perioperative real state information is greater than a preset value, and the negative sample set is a sample set constructed from historical graded data whose consistency degree between the graded result information and the perioperative real state information is less than a preset value; Inputting the positive sample set and the negative sample set into an initial classification model, calculating a loss function value according to the positive sample classification results and the negative sample classification results, and adjusting the model parameters through an optimization algorithm to minimize the loss function value, thereby obtaining a trained classification model; Wherein, the grading model includes: a cascaded feature extraction module and a multi-head attention mechanism layer, and an airway grading module, a cardiac function grading module and an ASA-PS grading module cascaded in parallel after the multi-head attention mechanism layer; the inputting of the positive sample set and the negative sample set into the initial grading model includes: respectively inputting the positive sample set and the negative sample set into the feature extraction module, respectively extracting the dynamic features and the static features in the basic vital sign information, the combined disease information and the organ function information in each group of historical grading data; in the multi-head attention mechanism layer, the dynamic features and the static features are adaptively weighted and matched to the airway grading module, the cardiac function grading module and the ASA-PS grading module input layer; the feature extraction module includes a physical sign entity feature extraction unit, an image feature extraction unit and a temporal feature extraction unit; The inputting the positive sample set and the negative sample set into the feature extraction module respectively comprises: Input the collected information into the feature extraction module, use the physical sign entity feature extraction unit to extract physical sign entity features, use the image feature extraction unit to extract airway image features, and use the physical sign entity features and the airway image features as the static features; input the physical sign entity features and the airway image features into the time series feature extraction unit respectively, and extract the time series feature sequence and time series change features of the static features as the dynamic features; The multi-head attention mechanism layer adaptively weights and matches the dynamic features and the static features to the airway grading module, the cardiac function grading module and the ASA-PS grading module input layer, including: constructing one-to-one corresponding attention matrices based on the airway grading module, the cardiac function grading module and the ASA-PS grading module respectively; calculating the dot product of the static features and the dynamic features with the attention matrices respectively, and obtaining the attention weights of the static features and the dynamic features in the airway grading module, the cardiac function grading module and the ASA-PS grading module respectively; combining the static features and the dynamic features based on the attention weights to construct the input features of the airway grading module, the cardiac function grading module and the ASA-PS grading module respectively.

2. The method for constructing a perioperative risk grading model according to claim 1, characterized in that: Before extracting the consistency between the grading result information and the perioperative real status information, the method further includes: Extracting the corresponding personnel types of each of the historical classification data respectively, the personnel types including: pregnant women, children and the elderly; Marking the historical classification data with a personnel type based on the personnel type; The model parameters of the grading model are adjusted using the personnel type mark to learn the corresponding relationship between the personnel type and the grading result.

3. A perioperative risk grading method, characterized in that: include: Obtaining collected information of the person to be tested, the collected information including basic physical sign information, combined disease information and organ function information; The collected information is input into the grading model constructed by the perioperative risk grading model construction method according to claim 1 or 2 to obtain the grading result of the person to be tested.

4. The perioperative risk grading method according to claim 3, characterized in that: The classification results include the ASA-PS classification results, cardiac function classification results and airway classification results of the person to be tested; After the grading result is obtained, the surgical risk result is predicted based on the ASA-PS grading result, the cardiac function grading result and the airway grading result.

5. An electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein: The processor, the communication interface and the memory communicate with each other via the communication bus, wherein: The memory is used to store computer programs; The processor is used to execute the perioperative risk grading model construction method described in claim 1 or 2 and / or the perioperative risk grading method described in claim 3 or 4 by running the computer program stored in the memory.

Images