A delirium auxiliary prediction and evaluation system based on multi-dimensional data

Abstract

The present invention provides a delirium auxiliary prediction and evaluation system based on multi-dimensional data, which relates to the field of medical data processing. The system includes: a graph establishment module, which is used to establish a first knowledge graph, a second knowledge graph and a third knowledge graph based on the electronic medical records of multiple sample patients; a model establishment module, which is used to establish a delirium prediction model and a delirium auxiliary evaluation model; a medical record acquisition module, which is used to obtain the electronic medical records of the patient to be evaluated; a feature extraction module, which is used to extract the medical record keyword features; a delirium prediction module, which is used to determine the delirium prediction result through the delirium prediction model; a scheme generation module, which is used to determine the clinical verification scheme; a verification acquisition module, which is used to obtain clinical verification data; a delirium evaluation module, which is used to determine the delirium auxiliary evaluation result through the delirium auxiliary evaluation model. The system can solve the technical problem of low accuracy of delirium recognition methods, and has the advantages of realizing automated real-time evaluation and improving evaluation accuracy.

Description

Technical Field

[0001] The present invention relates to the field of medical data processing, and in particular to a delirium auxiliary prediction and evaluation system based on multi-dimensional data. Background Art

[0002] Delirium is a common geriatric syndrome in elderly hospitalized patients and is closely related to poor prognosis. Studies have reported that one-third of general medical elderly hospitalized patients over 70 years old have delirium, but no standardized method for determining delirium in the elderly has been established. Therefore, research on delirium is particularly necessary to provide a reference for clinical treatment and improve the health level of the elderly population.

[0003] Existing research on delirium identification in the elderly is mainly based on bedside scale scoring. However, it is difficult to achieve real-time evaluation with scales and electronic medical record questionnaires, and the evaluation results are affected by the professionalism and proficiency of the evaluators. Therefore, the efficiency and accuracy of existing delirium identification methods are low.

[0004] Therefore, there is a need to provide a delirium auxiliary prediction and assessment system based on multi-dimensional data to realize automated real-time assessment of delirium and improve the accuracy of delirium assessment. Summary of the invention

[0005] In order to solve the technical problem in the prior art that the efficiency and accuracy of delirium recognition methods are low due to the professionalism and proficiency of evaluators, the present invention provides a delirium auxiliary prediction and evaluation system based on multi-dimensional data.

[0006] The present invention provides a delirium auxiliary prediction and evaluation system based on multi-dimensional data, comprising: a graph establishment module, used to obtain electronic medical records of multiple sample patients, and to establish a first knowledge graph based on the electronic medical records of the multiple sample patients, wherein the first knowledge graph is used to record the association information between multiple diseases, drugs and surgeries and delirium; the graph establishment module is also used to establish a second knowledge graph based on the electronic medical records of the multiple sample patients and the first knowledge graph, wherein the second knowledge graph is used to record the association information between any two diseases, any two drugs, any two surgeries, any disease and any drug, any The invention relates to association information about delirium between a disease and surgery, any drug and any surgery; the graph establishment module is also used to establish a third knowledge graph based on the electronic medical records of the multiple sample patients, wherein the third knowledge graph is used to record the association between multiple delirium types and onset characteristics; the model establishment module is used to establish a delirium prediction model based on the electronic medical records of the multiple sample patients and the second knowledge graph; the model establishment module is also used to establish a delirium auxiliary assessment model based on the electronic medical records of the multiple sample patients, the second knowledge graph and the third knowledge graph; the medical record acquisition module is used to acquire the medical records of the patient to be evaluated. electronic medical record; a feature extraction module, used to extract the medical record keyword features of the patient to be evaluated from the electronic medical record of the patient to be evaluated based on the first knowledge graph; a delirium prediction module, used to determine the delirium prediction result of the patient to be evaluated based on the medical record keyword features of the patient to be evaluated and the second knowledge graph through the delirium prediction model, wherein the delirium prediction result includes the risk probability of each delirium type; a scheme generation module, used to determine the clinical verification scheme of the patient to be evaluated based on the delirium prediction result of the patient to be evaluated and the third knowledge graph; a verification acquisition module, used to obtain the clinical verification scheme of the patient to be evaluated based on the delirium prediction result of the patient to be evaluated and the third knowledge graph; A clinical verification plan for the patient to be evaluated, and obtaining the clinical verification data of the patient to be evaluated; a delirium assessment module, used to determine the delirium auxiliary assessment result of the patient to be evaluated based on the medical record characteristics of the patient to be evaluated, the third knowledge graph and the clinical verification data of the patient to be evaluated through the delirium auxiliary assessment model; an intervention prediction module, used to determine the delirium intervention plan according to the medical record characteristics of the patient to be evaluated, the clinical verification data of the patient to be evaluated and the delirium auxiliary assessment result of the patient to be evaluated, wherein the delirium intervention plan includes reversible risk factors and non-drug intervention measures corresponding to the reversible risk factors.

[0007] Furthermore, the graph establishment module establishes a first knowledge graph based on the electronic medical records of the multiple sample patients, including: for each of the sample patients, extracting basic characteristics, disease characteristics, drug characteristics, surgical characteristics and delirium characteristics from the electronic medical records of the sample patients; based on the basic characteristics of any two of the sample patients, calculating the similarity of the basic characteristics of any two of the sample patients; using a clustering algorithm to cluster the multiple sample patients according to the similarity of the basic characteristics of any two of the sample patients, and determining multiple sample patient clusters; for each of the sample patient clusters, based on the disease characteristics, drug characteristics, surgical characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, determining the association information between multiple diseases, drugs and surgeries and delirium; based on the association information between multiple diseases, drugs and surgeries and delirium corresponding to each of the sample patient clusters, establishing the first knowledge graph.

[0008] Furthermore, the atlas establishment module extracts basic features, disease features, drug features, surgical features and delirium features from the electronic medical records of the sample patients, including: establishing a medical terminology vocabulary, a stop word library and a stop word exception rule library; based on the medical terminology vocabulary, the stop word library and the stop word exception rule library, removing invalid stop words in the electronic medical records of the sample patients to generate a first word segmentation result; performing maximum string frequency matching on the first word segmentation result to generate a second word segmentation result; generating a third word segmentation result based on the second word segmentation result through a long short-term memory network; and extracting the basic features, disease features, drug features, surgical features and delirium features according to the third word segmentation result.

[0009] Furthermore, the atlas building module determines the association information between multiple diseases, drugs, surgeries and delirium based on the disease characteristics, drug characteristics, surgical characteristics and delirium characteristics of multiple sample patients included in the sample patient clustering cluster, including: calculating the beta coefficient and odds ratio between each disease and delirium based on the disease characteristics and delirium characteristics of multiple sample patients included in the sample patient clustering cluster, and calculating the association coefficient between each disease and delirium based on the beta coefficient and odds ratio between each disease and delirium; calculating the association coefficient between each disease and delirium based on the sample patient clustering cluster. The patient cluster includes drug characteristics and delirium characteristics of multiple sample patients, and the beta coefficient and odds ratio between each drug and delirium are calculated. Based on the beta coefficient and odds ratio between each drug and delirium, the association coefficient between each drug and delirium is calculated; based on the surgical characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, the beta coefficient and odds ratio between each surgery and delirium are calculated, and based on the beta coefficient and odds ratio between each surgery and delirium, the association coefficient between each surgery and delirium is calculated.

[0010] Furthermore, the graph establishment module establishes a second knowledge graph based on the electronic medical records of the multiple sample patients and the first knowledge graph, including: calculating the joint beta coefficient and the joint odds ratio between any two diseases and delirium based on the disease characteristics and delirium characteristics of the multiple sample patients included in the sample patient clustering cluster, and calculating the correlation coefficient between any two diseases and delirium based on the joint beta coefficient and the joint odds ratio between any two diseases and delirium; calculating the joint beta coefficient and the joint odds ratio between any two drugs and delirium based on the drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient clustering cluster, and calculating the joint beta coefficient and the joint odds ratio between any two drugs and delirium based on the joint beta coefficient and the joint odds ratio between any two drugs and delirium. The joint beta coefficient and joint odds ratio between delirium and any two drugs are calculated based on the surgical characteristics and delirium characteristics of the multiple sample patients included in the sample patient clustering cluster, and the joint beta coefficient and joint odds ratio between any two surgeries and delirium are calculated. The joint beta coefficient and joint odds ratio between any two surgeries and delirium are calculated based on the joint beta coefficient and joint odds ratio between any two surgeries and delirium. The joint beta coefficient and joint odds ratio between any disease and any drug and delirium are calculated based on the disease characteristics, drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient clustering cluster, and the joint beta coefficient and joint odds ratio between any disease and any drug and delirium are calculated based on the joint beta coefficient and joint odds ratio between any disease and any drug and delirium. The joint beta coefficient and joint odds ratio between the drugs and delirium are calculated, and the correlation coefficient between any disease and any drug and delirium is calculated; based on the disease characteristics, surgical characteristics and delirium characteristics of the multiple sample patients included in the sample patient clustering cluster, the joint beta coefficient and joint odds ratio between any disease and any surgery and delirium are calculated, and the correlation coefficient between any disease and any surgery and delirium is calculated based on the joint beta coefficient and joint odds ratio between any disease and any surgery and delirium; based on the surgical characteristics, drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient clustering cluster, the correlation coefficient between any disease and any surgery and delirium is calculated. and the joint beta coefficient and joint odds ratio between any drug and delirium, and calculate the correlation coefficient between any surgery and any drug and delirium based on the joint beta coefficient and joint odds ratio between any surgery and any drug and delirium; based on the correlation coefficient between any two diseases and delirium, the correlation coefficient between any two drugs and delirium, the correlation coefficient between any two surgeries and delirium, the correlation coefficient between any disease and any drug and delirium, the correlation coefficient between any disease and any surgery and delirium, and the correlation coefficient between any surgery and any drug and delirium, establish the second knowledge graph.

[0011] Furthermore, the graph establishment module establishes a third knowledge graph based on the electronic medical records of the multiple sample patients, including: for each of the sample patient clustering clusters, based on the delirium characteristics of any two sample patients included in the sample patient clustering cluster, calculating the similarity of the delirium characteristics of any two sample patients, using a clustering algorithm to cluster the multiple sample patients according to the similarity of the delirium characteristics of any two sample patients, and determining multiple sample patient clustering groups; for each of the sample patient clustering groups, based on the delirium characteristics of each sample patient included in the sample patient clustering group, calculating the correlation coefficient between each onset characteristic and the sample patient clustering group, based on the correlation coefficient between each onset characteristic and the sample patient clustering group, determining the associated onset characteristic corresponding to the sample patient clustering group, based on the correlation coefficient between each onset characteristic and the sample patient clustering group, determining the associated onset characteristic corresponding to the sample patient clustering group; based on the associated onset characteristic corresponding to each of the sample patient clustering groups, establishing the third knowledge graph.

[0012] Furthermore, the feature extraction module extracts medical record keyword features of the patient to be evaluated from the electronic medical record of the patient to be evaluated based on the first knowledge graph, including: extracting basic features of the patient to be evaluated from the electronic medical record of the patient to be evaluated; determining a target sample patient cluster based on the basic features of the patient to be evaluated; determining the medical record keywords corresponding to the target sample patient cluster based on the first knowledge graph; and extracting medical record keyword features of the patient to be evaluated from the electronic medical record of the patient to be evaluated based on the medical record keywords corresponding to the target sample patient cluster.

[0013] Furthermore, the scheme generation module determines the clinical verification scheme of the patient to be evaluated based on the delirium prediction result of the patient to be evaluated and the third knowledge graph, including: determining the target sample patient clustering group based on the delirium prediction result of the patient to be evaluated, the target sample patient clustering cluster and the third knowledge graph; determining the clinical verification scheme of the patient to be evaluated based on the associated disease characteristics corresponding to the target sample patient clustering group.

[0014] Further, the verification acquisition module includes: an automatic dialogue unit, which is used to conduct a dialogue with the patient to be evaluated through an automatic dialogue model, and determine the consciousness characteristics of the patient to be evaluated based on the dialogue information of the patient to be evaluated; an emotion recognition unit, which is used to identify the emotion change characteristics of the patient to be evaluated based on the dialogue information of the patient to be evaluated; a delusion recognition unit, which is used to obtain the video and voice of the patient to be evaluated, identify the abnormal movement characteristics of the patient to be evaluated based on the video of the patient to be evaluated, identify the objection voice characteristics of the patient to be evaluated based on the video and voice of the patient to be evaluated, and determine the delusion characteristics of the patient to be evaluated based on the abnormal movement characteristics and abnormal voice characteristics of the patient to be evaluated; an activity perception unit, including a plurality of acceleration sensors and an activity analysis component arranged at different positions of the patient to be evaluated, wherein the activity analysis component is used to determine the activity characteristics of the patient to be evaluated based on the output data of the plurality of acceleration sensors at a plurality of time points; a biological marker acquisition unit, which is used to acquire the biological marker characteristics of the patient to be evaluated; and an electroencephalogram acquisition unit, which is used to acquire the electroencephalogram characteristics of the patient to be evaluated.

[0015] Furthermore, the delirium assessment module determines the delirium auxiliary assessment result of the patient to be assessed through the delirium auxiliary assessment model based on the medical record characteristics of the patient to be assessed, the third knowledge graph and the clinical verification data of the patient to be assessed, including: determining the delirium auxiliary assessment result of the patient to be assessed through the delirium auxiliary assessment model based on the medical record characteristics of the patient to be assessed, the third knowledge graph and at least one of the consciousness characteristics, emotional change characteristics, delusion characteristics, activity characteristics, biological marker characteristics and brain wave characteristics of the patient to be assessed.

[0016] Compared with the prior art, the delirium auxiliary prediction and evaluation system based on multi-dimensional data provided by the present invention has at least the following beneficial effects:

[0017] 1. By establishing the first knowledge graph, the second knowledge graph, and the third knowledge graph, the electronic medical record data of a large number of sample patients can be systematically integrated and analyzed to dig out the complex relationship between diseases, drugs, surgery, and delirium. This helps to consider various factors more comprehensively when making auxiliary assessments of delirium, and improve the scientificity and accuracy of decision-making. The delirium prediction model can predict the risk of delirium in patients in advance based on the keyword features of the patient's medical record and the second knowledge graph, and give the risk probability of each type of delirium. It can customize a personalized clinical verification plan for the patient based on the patient's delirium prediction results and the third knowledge graph. This can not only improve the pertinence and efficiency of the verification, but also reduce unnecessary examinations and reduce the patient's medical costs. The delirium auxiliary assessment model combines the patient's medical record characteristics, the third knowledge graph, and clinical verification data to more comprehensively evaluate the patient's delirium condition.

[0018] 2. The medical terminology lexicon contains a large number of professional medical words and terms that appear frequently in medical texts and have specific meanings. By using the medical terminology lexicon for word segmentation, it is possible to ensure that these professional words are correctly identified and segmented, thereby improving the accuracy of text processing. The stop word library contains words that appear frequently in the text but do not contribute much to the text topic, such as prepositions, conjunctions, and auxiliary words. Removing these invalid stop words can reduce noise interference, make the text more concise and clear, and facilitate subsequent feature extraction and analysis. The stop word exception rule library is used to process words that belong to the stop word category but have important meanings in a specific context. Through this rule library, it is possible to ensure that these special words are retained during the word segmentation process to avoid information loss caused by accidental deletion. The use of advanced word segmentation technologies such as string frequency maximum matching and long short-term memory network can further improve the word segmentation effect. String frequency maximum matching uses the frequency information of words to optimize the word segmentation results, while LSTM can capture long-distance dependencies in the text to generate more accurate and coherent word segmentation results. Based on high-quality word segmentation results, key information such as basic characteristics, disease characteristics, drug characteristics, surgical characteristics and delirium characteristics can be extracted more accurately.

[0019] 3. By calculating the beta coefficient and odds ratio between each disease, drug, surgery and delirium, and then calculating the correlation coefficient, we can accurately identify which diseases, drugs, and surgeries are high-risk factors for delirium in patients in the sample patient cluster. The analysis of correlation information not only reveals risk factors, but also may provide clues for understanding the potential mechanism between disease and delirium, and provide information assistance for the subsequent extraction of medical record keyword features of patients to be evaluated from their electronic medical records, reducing the acquisition of irrelevant information and improving the efficiency and accuracy of delirium auxiliary evaluation;

[0020] 4. The medical terminology lexicon can include common medical terms, disease names, drug names, surgical names, etc., which are used to identify these professional terms in word segmentation and subsequent processing. The medical terminology lexicon can include a delirium medical record lexicon. Medicine is closely related to human health. Therefore, medical technology will have different focuses in different periods, and new diagnosis and treatment technologies and new field terms are constantly emerging. Therefore, the delirium medical record lexicon also needs to be continuously supplemented. The discovery of new words in the lexicon can also promote the reduction of false negatives in delirium prediction models and delirium auxiliary assessment models. The method of discovering new words in delirium medical records is to segment the delirium medical records in a new cycle through the delirium medical record lexicon and the lexicon system at regular intervals (such as a quarter or half a year). After segmentation, the remaining words that cannot be segmented are used as potential new words. If the number of potential new words in the delirium medical records in a new cycle exceeds a certain threshold of the existing words (e.g., 10%), the maximum frequency matching of these potential new words will be performed again and combined with the long short-term memory network model. At the same time, the results will be annotated by experts. If they are identified as new words in delirium medical records, they will be added to the delirium medical record vocabulary. In this way, the delirium medical record vocabulary is continuously expanded. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0022] Figure 1 It is a module schematic diagram of a delirium auxiliary prediction and assessment system based on multi-dimensional data according to some embodiments of this specification;

[0023] Figure 2 is a schematic diagram of a first knowledge graph according to some embodiments of this specification;

[0024] Figure 3 is a schematic diagram of a sub-graph of the second knowledge graph according to some embodiments of this specification;

[0025] Figure 4 It is a schematic diagram of the third knowledge graph shown in some embodiments of this specification. DETAILED DESCRIPTION

[0026] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0027] Figure 1 is a module schematic diagram of a delirium auxiliary prediction and evaluation system based on multi-dimensional data according to some embodiments of this specification, such as Figure 1 As shown, a delirium auxiliary prediction and evaluation system based on multi-dimensional data may include a map building module, a model building module, a medical record acquisition module, a feature extraction module, a delirium prediction module, a solution generation module, a verification acquisition module and a delirium evaluation module.

[0028] The graph building module can be used to obtain the electronic medical records of multiple sample patients and build a first knowledge graph based on the electronic medical records of the multiple sample patients.

[0029] Among them, the first knowledge graph is used to record the correlation information between various diseases, drugs, surgeries and delirium.

[0030] Some of the multiple sample patients may be patients who do not have delirium, and the other part may be patients who have delirium. Electronic medical records may include basic patient information (e.g., gender, age, blood type, lifestyle, etc.), diagnosis and treatment process information (e.g., medical records, course of disease records), examination result information (e.g., results of medical imaging examinations such as radiology, ultrasound, and endoscopy, and laboratory test results, such as blood routine, urine routine, biochemical tests, etc.), medication information (e.g., drug name, specification, dosage, usage, medication time, etc.), and delirium manifestation information (e.g., impaired consciousness, loss of orientation, sensory confusion, agitation, etc.).

[0031] In some embodiments, the graph building module builds a first knowledge graph based on the electronic medical records of multiple sample patients, including:

[0032] For each sample patient, basic characteristics, disease characteristics, medication characteristics, surgical characteristics, and delirium characteristics were extracted from the sample patient's electronic medical records;

[0033] Based on the basic characteristics of any two sample patients (e.g., gender, age, blood type, weight, living habits, physiological and biochemical indicators, etc.), calculate the similarity of the basic characteristics of any two sample patients;

[0034] Using a clustering algorithm to cluster multiple sample patients according to the similarity of basic characteristics of any two sample patients, multiple sample patient clusters are determined;

[0035] For each sample patient cluster, based on the disease characteristics, drug characteristics, surgical characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, determining the association information between multiple diseases, drugs and surgeries and delirium;

[0036] The first knowledge graph is established based on the correlation information between various diseases, drugs, surgeries and delirium corresponding to each sample patient cluster.

[0037] In some embodiments, the atlas building module extracts basic features, disease features, drug features, surgical features, and delirium features from the electronic medical records of the sample patients, including:

[0038] Establish a medical terminology database, a stop word database, and a stop word exception rule database;

[0039] Based on the medical terminology database, the stop word database and the stop word exception rule database, invalid stop words in the electronic medical records of sample patients are removed to generate the first word segmentation result;

[0040] Perform maximum string frequency matching on the first word segmentation result to generate a second word segmentation result;

[0041] Generate a third word segmentation result based on the second word segmentation result through a long short-term memory network;

[0042] According to the third segmentation results, basic characteristics, disease characteristics, drug characteristics, surgical characteristics and delirium characteristics were extracted.

[0043] Specifically, electronic medical records contain a large amount of expert knowledge, clinical verification and other information, but these medical records are unstructured data and cannot be directly used for data mining. Therefore, it is necessary to convert electronic medical records into structured data, filter out non-clinical terminology, and retain clinically relevant terms.

[0044] The medical terminology lexicon can contain common medical terms, disease names, drug names, surgical names, etc., which are used to identify these professional terms in word segmentation and subsequent processing. The medical terminology lexicon can include a delirium medical record lexicon. Medicine is closely related to human health. Therefore, medical technology will have different focuses in different periods, and new diagnosis and treatment technologies and new field terms are constantly emerging. Therefore, the delirium medical record lexicon also needs to be continuously supplemented. The discovery of new words in the lexicon can also promote the reduction of false negatives in delirium prediction models and delirium auxiliary assessment models. The method of discovering new words in delirium medical records is to segment the delirium medical records in a new cycle through the delirium medical record lexicon and the lexicon system at regular intervals (such as a quarter or half a year). After segmentation, the remaining words that cannot be segmented are used as potential new words. If the number of potential new words in the delirium medical records in a new cycle exceeds a certain threshold of the existing words (e.g., 10%), the maximum frequency matching of these potential new words will be performed again and combined with the long short-term memory network model. At the same time, the results will be annotated by experts. If they are identified as new words in delirium medical records, they will be added to the delirium medical record vocabulary. In this way, the delirium medical record vocabulary is continuously expanded.

[0045] The stop word library can contain some words that frequently appear in the text but do not contribute practically to the analysis, such as "的", "了", "是", etc. Removing these words can improve the efficiency and accuracy of text processing.

[0046] Considering the particularity of medical texts, some words that are stop words in regular texts may have important meanings in medical texts (such as "的" in "的并发症"). The stop word exception rule library is used to define the stop word processing rules in these special cases.

[0047] Using the established medical term library, stop word library, and stop word exception rule library, remove the invalid stop words from the electronic medical record text to ensure that medical terms can be correctly identified and segmented. Segment the electronic medical records of sample patients into a series of meaningful words or phrases.

[0048] The maximum matching by frequency of strings is a word segmentation algorithm. Its basic idea is to scan the string to be segmented from left to right, and each time take out the longest possible string for matching. If a corresponding word can be found in the dictionary, the matching is successful and the word is segmented; if not, the last character of the string is removed and the matching is performed again until a word is found or the remaining string is empty. It matches the longest possible word according to the frequency and length of the word in the text. In this step, further processing is performed on the first word segmentation result to improve the accuracy and efficiency of word segmentation.

[0049] The long short-term memory network is a special type of recurrent neural network that can handle long-term dependencies in sequential data. In this step, the second word segmentation result is used as the input to the long short-term memory network model, and the sequence modeling ability of the long short-term memory network model is used to further optimize and adjust the word segmentation result. Generate the third word segmentation result, that is, a more accurate and reasonable word segmentation result after being processed by the long short-term memory network model.

[0050] For example only, the basic features can include the patient's gender, age, blood type, weight, living habits, physiological and biochemical indicators, etc. The disease features can include information such as the name of the disease the patient has and the symptom description. The drug features can include information such as the name of the drug the patient uses, the dosage, and the medication time. The surgical features can include information such as the name of the surgery the patient undergoes, the surgery time, and the surgery result. The delirium features can include the features related to delirium shown by the patient, for example, disturbance of consciousness, cognitive dysfunction, perceptual disorders, emotional features, behavioral features, etc.

[0051] The atlas building module can calculate the Euclidean distance of the basic features of two sample patients as the similarity of the basic features of the two sample patients.

[0052] The graph building module can cluster multiple sample patients according to the following algorithm process to determine multiple sample patient clusters:

[0053] S11, based on the basic feature similarity of any two sample patients, calculating the variance of the basic feature similarity, determining the number of sample patient clusters according to the variance of the basic feature similarity, and randomly selecting multiple sample patients as multiple cluster centers based on the number of sample patient clusters;

[0054] S12, for each sample patient, clustering the sample patient to the cluster center with the largest basic feature similarity;

[0055] S13. For each sample patient cluster, calculate the corresponding basic feature similarity mean of each sample patient included in the sample patient cluster, wherein the corresponding basic feature similarity mean of the sample patient may be the mean of the basic feature similarity between the sample patient and each other sample patient included in the sample patient cluster, and take the sample patient with the largest basic feature similarity mean as the new cluster center of the sample patient cluster;

[0056] S14, judging whether the cluster center of each sample patient cluster has not changed, if so, completing the clustering, if not, executing S12.

[0057] For example, the number of sample patient clusters can be calculated based on the following formula:

[0058]

[0059] in, is the number of sample patient clusters, is the variance of the basic feature similarity, is the preset parameter, and greater than 0, The number of clusters for pre-set sample patients.

[0060] Using clustering algorithms to group sample patients can group patients with similar characteristics together based on the similarity of basic characteristics. This grouping method helps to more accurately identify the association between diseases, drugs, surgeries and delirium in different patient groups. This is because patients with similar characteristics may face similar risk factors and disease progression pathways. By analyzing the association information between diseases, drugs, surgeries and delirium within clusters, a knowledge graph that is closer to the actual clinical situation can be established, which has higher practicality.

[0061] In some embodiments, the atlas building module determines the association information between multiple diseases, drugs, surgeries and delirium based on the disease characteristics, drug characteristics, surgery characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, including:

[0062] Based on the disease characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, the beta coefficient and odds ratio between each disease and delirium are calculated; based on the beta coefficient and odds ratio between each disease and delirium, the correlation coefficient between each disease and delirium is calculated;

[0063] Based on the drug characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, the beta coefficient and odds ratio between each drug and delirium are calculated; based on the beta coefficient and odds ratio between each drug and delirium, the correlation coefficient between each drug and delirium is calculated;

[0064] Based on the surgical characteristics and delirium characteristics of multiple sample patients included in the sample patient clustering cluster, the beta coefficient and odds ratio between each surgery and delirium were calculated. Based on the beta coefficient and odds ratio between each surgery and delirium, the correlation coefficient between each surgery and delirium was calculated.

[0065] Specifically, the beta coefficient between each disease and delirium can be calculated based on regression analysis. For example, using statistical software (such as SPSS, R, SAS, etc.) to perform logistic regression analysis, setting the dependent variable to the occurrence of delirium, and the independent variable to the presence of the disease, running the logistic regression model, the software will automatically calculate the beta coefficient of the independent variable.

[0066] The odds ratio for each disease and delirium can be calculated based on the following formula:

[0067]

[0068] in, is the odds ratio between the jth disease and delirium corresponding to the i-th sample patient cluster, is the number of sample patients with delirium who have the jth disease in the i-th sample patient cluster, is the number of sample patients with delirium who do not suffer from the jth disease in the i-th sample patient cluster, is the number of sample patients with the jth disease and without delirium in the i-th sample patient cluster, is the number of sample patients in the i-th sample patient cluster who do not suffer from the j-th disease and do not suffer from delirium.

[0069] The association coefficient between each disease and delirium can be calculated based on the following formula:

[0070]

[0071] in, is the correlation coefficient between the j-th disease corresponding to the i-th sample patient cluster and delirium, and All are preset weights. , and are greater than 0, is the beta coefficient between the j-th disease corresponding to the i-th sample patient cluster and delirium.

[0072] The method of calculating the association coefficient between each drug and delirium, and the association coefficient between each surgery and delirium is similar to the method of calculating the association coefficient between each disease and delirium, and will not be repeated here.

[0073] For each sample patient cluster, a disease with a correlation coefficient greater than a preset correlation coefficient with delirium can be used as the associated disease of the sample patient cluster, a drug with a correlation coefficient greater than a preset correlation coefficient with delirium can be used as the associated drug of the sample patient cluster, and a surgery with a correlation coefficient greater than a preset correlation coefficient with delirium can be used as the associated surgery of the sample patient cluster. This is just an example. Figure 2 is a schematic diagram of a first knowledge graph according to some embodiments of this specification, such as Figure 2 As shown, the first knowledge graph may include four types of nodes, the first type of node is a node representing a sample patient cluster, the second type of node is a node representing an associated disease, the third type of node is a node representing an associated drug, and the fourth type of node is a node representing an associated surgery. The node representing the sample patient cluster can be connected to the nodes representing the associated disease, associated drug, and associated surgery of the sample patient cluster through edges.

[0074] The graph building module can also be used to build a second knowledge graph based on the electronic medical records of multiple sample patients and the first knowledge graph.

[0075] Among them, the second knowledge graph is used to record the association information about delirium between any two diseases, any two drugs, any two surgeries, any disease and any drug, any disease and surgery, and any drug and any surgery.

[0076] In some embodiments, the graph building module builds a second knowledge graph based on the electronic medical records of multiple sample patients and the first knowledge graph, including:

[0077] Based on the disease characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, the joint beta coefficient and the joint odds ratio between any two diseases and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any two diseases and delirium, the association coefficient between any two diseases and delirium is calculated, wherein the any two diseases are any two associated diseases corresponding to the sample patient cluster;

[0078] Based on the drug characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, the joint beta coefficient and the joint odds ratio between any two drugs and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any two drugs and delirium, the association coefficient between any two drugs and delirium is calculated, wherein the any two drugs are any two associated drugs corresponding to the sample patient cluster;

[0079] Based on the surgical characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, the joint beta coefficient and the joint odds ratio between any two surgeries and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any two surgeries and delirium, the correlation coefficient between any two surgeries and delirium is calculated, wherein the any two surgeries are any two associated surgeries corresponding to the sample patient cluster;

[0080] Based on the disease characteristics, drug characteristics and delirium characteristics of multiple sample patients included in the sample patient clustering cluster, the joint beta coefficient and the joint odds ratio between any disease and any drug and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any disease and any drug and delirium, the correlation coefficient between any disease and any drug and delirium is calculated, wherein any disease is the associated disease corresponding to the sample patient clustering cluster, and any drug is the associated drug corresponding to the sample patient clustering cluster;

[0081] Based on the disease characteristics, surgical characteristics and delirium characteristics of multiple sample patients included in the sample patient clustering cluster, the joint beta coefficient and the joint odds ratio between any disease and any surgery and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any disease and any surgery and delirium, the correlation coefficient between any disease and any surgery and delirium is calculated, wherein any disease is the associated disease corresponding to the sample patient clustering cluster, and any surgery is the associated surgery corresponding to the sample patient clustering cluster;

[0082] Based on the surgical characteristics, drug characteristics and delirium characteristics of multiple sample patients included in the sample patient clustering cluster, the joint beta coefficient and joint odds ratio between any surgery and any drug and delirium are calculated; based on the joint beta coefficient and joint odds ratio between any surgery and any drug and delirium, the correlation coefficient between any surgery and any drug and delirium is calculated, wherein any drug is the associated drug corresponding to the sample patient clustering cluster, and any surgery is the associated surgery corresponding to the sample patient clustering cluster;

[0083] A second knowledge graph is established based on the correlation coefficient between any two diseases and delirium, the correlation coefficient between any two drugs and delirium, the correlation coefficient between any two surgeries and delirium, the correlation coefficient between any disease and any drug and delirium, the correlation coefficient between any disease and any surgery and delirium, and the correlation coefficient between any surgery and any drug and delirium.

[0084] Specifically, the beta coefficient between any two diseases and delirium can be calculated based on regression analysis. For example, using statistical software to perform logistic regression analysis, setting the dependent variable to the occurrence of delirium, and setting the independent variable to the co-existence of the two diseases, running the logistic regression model, the software will automatically calculate the beta coefficient of the independent variable.

[0085] The joint odds ratio between any two associated diseases and delirium can be calculated based on the following formula:

[0086]

[0087] in, is the joint odds ratio between the e-th associated disease and the f-th associated disease and delirium corresponding to the i-th sample patient cluster, is the number of sample patients with delirium who suffer from both the e-th associated disease and the f-th associated disease in the i-th sample patient cluster, is the number of sample patients with delirium who do not suffer from the e-th associated disease and the f-th associated disease at the same time in the i-th sample patient cluster, is the number of sample patients in the i-th sample patient cluster who suffer from both the e-th associated disease and the f-th associated disease and do not suffer from delirium, is the number of sample patients in the i-th sample patient cluster who do not suffer from both the e-th associated disease and the f-th associated disease and do not suffer from delirium.

[0088] The method of calculating the correlation coefficient between any two related diseases and delirium is similar to the method of calculating the correlation coefficient between a disease and delirium, which will not be repeated here.

[0089] The method of calculating the association coefficient between each drug and delirium, and the association coefficient between each surgery and delirium is similar to the method of calculating the association coefficient between each disease and delirium, and will not be repeated here.

[0090] For example only, Figure 3 is a schematic diagram of a sub-graph of the second knowledge graph according to some embodiments of this specification, such as Figure 3 As shown, the second knowledge graph may include multiple sub-graphs, and one sub-graph corresponds to one sample patient cluster. The sub-graph includes three types of nodes, the first type of node is a node representing the associated disease of the sample patient cluster, the second type of node is a node representing the associated drug of the sample patient cluster, and the third type of node is a node representing the associated surgery of the sample patient cluster. The edge between two nodes can represent the correlation coefficient of the two nodes.

[0091] The graph building module can also be used to build a third knowledge graph based on the electronic medical records of multiple sample patients.

[0092] Among them, the third knowledge graph is used to record the association between various delirium types and disease characteristics.

[0093] In some embodiments, the graph building module builds a third knowledge graph based on the electronic medical records of multiple sample patients, including:

[0094] For each sample patient cluster, based on the delirium characteristics of any two sample patients included in the sample patient cluster, the similarity of the delirium characteristics of any two sample patients is calculated, and a clustering algorithm is used to cluster multiple sample patients according to the similarity of the delirium characteristics of any two sample patients to determine multiple sample patient cluster groups;

[0095] For each sample patient cluster group, based on the delirium characteristics of each sample patient included in the sample patient cluster group, calculate the correlation coefficient between each onset characteristic and the sample patient cluster group, and based on the correlation coefficient between each onset characteristic and the sample patient cluster group, determine the associated onset characteristic corresponding to the sample patient cluster group, wherein one sample patient cluster group may correspond to one delirium type;

[0096] A third knowledge graph is established based on the associated disease characteristics corresponding to each sample patient cluster group.

[0097] Specifically, the Euclidean distance between delirium features of any two sample patients included in the sample patient cluster may be calculated as the similarity of delirium features of any two sample patients included in the sample patient cluster.

[0098] The graph building module can cluster multiple sample patients according to the following algorithm process to determine multiple sample patient cluster groups:

[0099] S21, based on the similarity of delirium characteristics of any two sample patients included in the sample patient clustering cluster, calculating the variance of the delirium characteristic similarity, determining the number of sample patient clustering groups according to the variance of the delirium characteristic similarity, and based on the number of sample patient clustering groups, randomly selecting multiple sample patients from the sample patient clustering cluster as multiple clustering centers;

[0100] S22, for each sample patient included in the sample patient cluster, clustering the sample patient to a cluster center with the greatest delirium feature similarity;

[0101] S23. For each sample patient cluster group, calculate the corresponding delirium feature similarity mean of each sample patient included in the sample patient cluster group, wherein the corresponding delirium feature similarity mean of the sample patient may be the mean of the delirium feature similarity between the sample patient and each other sample patient included in the sample patient cluster group, and take the sample patient with the largest delirium feature similarity mean as the new cluster center of the sample patient cluster group;

[0102] S24, judging whether the cluster center of each sample patient cluster group has not changed, if so, completing the clustering, if not, executing S22.

[0103] The disease characteristics with correlation coefficients greater than the correlation coefficient threshold are used as the associated disease characteristics of the sample patient cluster group. Figure 4 is a schematic diagram of a third knowledge graph according to some embodiments of this specification, such as Figure 4 As shown, the third knowledge graph may include three types of nodes. The first type of node represents sample patient clusters, the second type of node represents sample patient cluster groups, and the third type of node represents associated disease characteristics of sample patient cluster groups.

[0104] The model building module can be used to build a delirium prediction model based on the electronic medical records of multiple sample patients and the second knowledge graph.

[0105] Specifically, the delirium prediction model may be a convolutional neural network (CNN) model. The input of the delirium prediction model may include basic features, disease features, drug features, and surgical features extracted from the patient's electronic medical record, and the output of the delirium prediction model may include the patient's delirium prediction result, wherein the delirium prediction result includes the risk probability of each delirium type. The delirium prediction model may include a sub-graph determination unit and a delirium type prediction unit, wherein the input of the sub-graph determination unit may include basic features, disease features, drug features, and surgical features extracted from the patient's electronic medical record and a second knowledge graph, the output of the sub-graph determination unit may include a target sub-graph determined from the second knowledge graph, the input of the delirium type prediction unit may include basic features, disease features, drug features, and surgical features extracted from the patient's electronic medical record and a target sub-graph, and the output of the delirium type prediction unit may include the patient's delirium prediction result.

[0106] Preferably, the loss function used to train the delirium prediction model is:

[0107]

[0108]

[0109] in, is the loss function of the delirium prediction model, and is the preset weight, and and are greater than 0, , is 0 or 1. When the sub-graph corresponding to the m-th training sample is the n-th sub-graph, ,otherwise, , The probability that the sub-spectrum corresponding to the m-th training sample predicted by the delirium prediction model is the n-th sub-spectrum, is the total number of training samples, is the total number of sub-graphs, is the total number of delirium types, is 0 or 1. When the delirium type corresponding to the mth training sample is the kth delirium type, then ,otherwise, , The probability that the delirium type corresponding to the mth training sample predicted by the delirium prediction model is the kth delirium type.

[0110] The model building module can also be used to build a delirium auxiliary assessment model based on the electronic medical records of multiple sample patients, the second knowledge graph and the third knowledge graph.

[0111] Specifically, multiple candidate delirium auxiliary assessment models can be established based on logistic regression (LR), random forest (RF), extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), gradient boosting decision tree (GBDT), distributed gradient boosting machine (LightGBM) based on decision tree algorithm, and gradient boosting with categorical features (CatBoost).

[0112] Random forest is a machine learning method for classification and regression. The algorithm can divide the training data set into different subsets, and each subset is trained to generate a decision tree. Because the random sampling method ensures that each tree in the random forest is unique, this method significantly reduces the overall variance of the model. Each tree in the random forest provides a classification prediction, and the classification with the most "votes" becomes the prediction of the model. Therefore, random forest can provide good results from a global perspective.

[0113] Adaptive boosting is the original Boosting algorithm. Its core idea is to train different classifiers (weak classifiers) for the same training set, and then combine these weak classifiers to form a stronger final classifier (strong classifier). The adaptive boosting algorithm itself is implemented by changing the data distribution. It determines the weight of each sample based on the classification of each sample in each training set and the accuracy of the last overall classification. The new data set with modified weights is sent to the lower-level classifier for training, and finally the classifiers obtained each time are finally fused as the final decision classifier.

[0114] Gradient boosted decision tree is a Boosting method. Its main idea is that each time a model is built, it is in the direction of the gradient descent of the loss function of the previously built model. The loss function is used to evaluate the performance of the model (generally the degree of fit + regularization term). It is believed that the smaller the loss function, the better the performance. By allowing the loss function to continue to decrease, the model can be continuously modified to improve performance. The best way is to make the loss function decrease along the gradient direction (the gradient direction decreases fastest). The gradient boosted decision tree uses the negative gradient of the loss function in the current model as the residual approximation of the boosted tree algorithm for regression problems to fit a regression tree.

[0115] Extreme gradient boosting is a machine learning method that focuses on the gradient boosting algorithm and is used for supervised learning, such as regression, classification, and ranking. Extreme gradient boosting improves the gradient boosting decision tree algorithm. The principle of the gradient boosting decision tree algorithm is to first train a tree using the training set and the sample true value (i.e., the standard answer), then use this tree to predict the training set to obtain the predicted value of each sample, and then use the deviation between the predicted value and the true value, i.e., the "residual", as the standard answer to train the next tree. Then continue to use the residual to train the third tree, and so on. The total number of trees can be specified manually, or certain indicators (such as the error on the validation set) can be monitored to stop training. In comparison, extreme gradient boosting not only has a good learning effect, but also is very fast and is very effective in controlling overfitting.

[0116] Gradient boosting based on categorical features is a newcomer after XGBoost and Light GBM. CatBoost is also an improved implementation under the GBDT algorithm framework. It is an algorithm based on symmetric decision trees with fewer parameters, support for categorical variables and high accuracy. This algorithm can efficiently and reasonably handle categorical features, and also solves the problems of gradient bias and prediction shift, thereby effectively reducing the occurrence of overfitting and improving the accuracy and generalization ability of the algorithm.

[0117] The leaf growth strategy used by LightGBM finds a leaf with the largest information gain (usually the largest amount of data) from the leaves each time and splits it, and repeats this cycle. Layer-by-layer growth algorithms such as XGBoost treat leaves of the same layer indiscriminately. Many leaves have low split gains and there is no need to search and split them. With the same number of splits, the leaf growth strategy can reduce more errors and obtain better accuracy.

[0118] A series of indicators including sensitivity, specificity, F1 value, Youden index, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio were used to evaluate the performance of multiple candidate delirium assisted assessment models in the test set and validation set, and the candidate delirium assisted assessment model with the best performance was selected as the final delirium assisted assessment model.

[0119] The medical record acquisition module can be used to obtain the electronic medical records of the patient to be evaluated.

[0120] The feature extraction module can be used to extract medical record keyword features of the patient to be evaluated from the electronic medical records of the patient to be evaluated based on the first knowledge graph.

[0121] Specifically include:

[0122] Extract the basic characteristics of the patients to be evaluated from their electronic medical records;

[0123] Based on the basic characteristics of the patients to be evaluated, determine the target sample patient clusters;

[0124] Based on the first knowledge graph, determine the medical record keywords corresponding to the target sample patient cluster;

[0125] Based on the medical record keywords corresponding to the target sample patient clusters, the medical record keyword features of the patients to be evaluated are extracted from their electronic medical records.

[0126] Specifically, the Euclidean distance between the basic characteristics of the patient to be evaluated and the basic characteristics of the cluster center of each sample patient cluster cluster can be calculated, and the sample patient cluster cluster to which the cluster center belongs whose Euclidean distance is less than the Euclidean distance threshold is taken as the target sample patient cluster cluster, and based on the first knowledge graph, the associated diseases, associated drugs, and associated surgeries of the target sample patient cluster cluster are determined, and then the medical record keywords corresponding to the associated diseases, associated drugs, and associated surgeries of the target sample patient cluster cluster are determined.

[0127] The delirium prediction module can be used to determine the delirium prediction result of the patient to be evaluated based on the medical record keyword features of the patient to be evaluated and the second knowledge graph through the delirium prediction model, wherein the delirium prediction result includes the risk probability of each delirium type.

[0128] The scheme generation module can be used to determine the clinical verification scheme for the patient to be evaluated based on the delirium prediction result of the patient to be evaluated and the third knowledge graph.

[0129] Specifically include:

[0130] Determine a target sample patient clustering group based on the delirium prediction results of the patient to be evaluated, the target sample patient clustering clusters, and the third knowledge graph;

[0131] Based on the associated disease characteristics corresponding to the target sample patient cluster group, the clinical verification plan for the patients to be evaluated is determined.

[0132] Specifically, the delirium type whose predicted risk probability is greater than the risk probability threshold in the delirium prediction results of the patient to be evaluated is used as the target delirium type, and the sample patient cluster group corresponding to the target delirium type is used as the target sample patient cluster group.

[0133] The clinical verification scheme for the patient to be evaluated can be determined by a scheme generation model based on the associated disease characteristics corresponding to the target sample patient cluster group, wherein the scheme generation model can be a convolutional neural network (CNN) model. The clinical verification scheme for the patient to be evaluated may include clinical verification means and verification frequency, wherein the clinical verification means may include at least one of conversation, emotion recognition, delusion recognition, activity perception, biological marker acquisition, and brain wave acquisition.

[0134] The verification acquisition module can be used to obtain clinical verification data of the patient to be evaluated based on the clinical verification plan of the patient to be evaluated.

[0135] In some embodiments, the verification acquisition module includes:

[0136] An automatic dialogue unit, used for conducting a dialogue with the patient to be evaluated through an automatic dialogue model, and determining the consciousness characteristics of the patient to be evaluated based on the dialogue information of the patient to be evaluated;

[0137] An emotion recognition unit, used to identify emotion change characteristics of the patient to be evaluated based on the conversation information of the patient to be evaluated;

[0138] a delusion recognition unit, used to obtain a video and voice of the patient to be evaluated, to recognize abnormal movement characteristics of the patient to be evaluated based on the video of the patient to be evaluated, to recognize objection voice characteristics of the patient to be evaluated based on the video and voice of the patient to be evaluated, and to determine the delusion characteristics of the patient to be evaluated based on the abnormal movement characteristics and abnormal voice characteristics of the patient to be evaluated;

[0139] An activity sensing unit, comprising a plurality of acceleration sensors arranged at different positions of the patient to be evaluated and an activity analysis component, wherein the activity analysis component is used to determine the activity characteristics of the patient to be evaluated based on output data of the plurality of acceleration sensors at a plurality of time points;

[0140] A biomarker acquisition unit, used to obtain the biomarker characteristics of the patient to be evaluated (e.g., inflammation-related markers, neuroprotection-related markers, astrocyte and glial activation-related markers, etc.);

[0141] The brain wave acquisition unit is used to acquire the brain wave characteristics of the patient to be evaluated.

[0142] Specifically, the automated conversation model contains a series of carefully designed questions and evaluation criteria designed to capture the patient's cognitive function, attention, memory ability, and the coherence and logic of the conversation. The automated conversation model starts the conversation in a friendly way, such as using simple greetings and self-introductions to guide the patient to relax and participate in the conversation. The automated conversation model preliminarily assesses the patient's attention level and sense of direction by asking simple but direct questions such as "Where are you now?" or "Can you tell me what date it is today?" The automated conversation model asks the patient about recent events or important personal information (such as the names of family members, recent activities, etc.) to assess their memory and orientation. The automated conversation model allows the patient to describe an event, tell a story, or answer complex questions to observe whether their answers are coherent and organized, and whether they can keep the focus of the topic. The automated conversation model can automatically assess whether the patient is frequently distracted during the conversation and has difficulty maintaining attention. The automated conversation model asks the patient whether there are visual, auditory, or other perceptual abnormalities, such as hallucinations or illusions. The automated conversation model can assess whether the patient's speech is confused, the logic is unclear, and whether there are meaningless repetitive words. The automatic dialogue model can record the conversation process and conduct a detailed analysis, including the patient's answer content, speaking speed, tone, reaction time, etc., and comprehensively evaluate the patient's consciousness characteristics by combining the patient's conversation performance and the evaluation criteria of the automatic dialogue model.

[0143] The delusion recognition unit can use a high-definition camera to capture the real-time video stream of the patient to be evaluated, ensuring that the patient's movements, facial expressions and body postures can be clearly observed, and collect the patient's voice data through a microphone array or a dedicated recording device to ensure that the patient's speech content, tone, speed and other characteristics can be accurately captured. Computer vision technology, such as deep learning algorithms, is used to analyze the patient's movements in the video stream. By comparing the normal behavior pattern database, the patient's abnormal movement characteristics, such as frequent gestures, meaningless body swings, and sudden changes in movement, are identified. At the same time, the patient's facial expressions are analyzed to identify emotional expressions that may be related to delusions, such as tension, fear, and doubt. The collected voice data is processed to extract voice features, such as tone, speed, volume, pause pattern, etc. By comparing the normal speech pattern database, the patient's dissenting voice features, such as incoherent speech content, logical confusion, meaningless repetition or fictional narration, are identified. Combined with the voice features, the patient's emotional state is further analyzed to determine whether his speech reveals emotions such as anxiety, paranoia, fear, etc. that may be related to delusions. The abnormal movement features obtained by video analysis and the dissenting voice features obtained by voice analysis are comprehensively evaluated. Consider the patient's overall presentation, including actions, speech, emotions, and other aspects, to determine the characteristics of his or her delusions.

[0144] The emotion recognition unit can identify the emotion type of the patient to be evaluated at multiple conversation time points, video collection time points, and voice collection time points, and calculate the emotion change characteristics of the patient to be evaluated. Preferably, the emotion change characteristics of the patient to be evaluated can be calculated based on the following formula:

[0145]

[0146]

[0147] in, is the average emotional type of the patients to be evaluated, is the emotional type of the patient to be evaluated at time point t, is the total number of sampling time points, The emotion fluctuation parameters of the patient to be evaluated. The emotion change characteristics of the patient to be evaluated may include the average emotion type of the patient to be evaluated and the emotion fluctuation parameters of the patient to be evaluated.

[0148] The activity analysis component can perform empirical mode decomposition on the output data of each acceleration sensor at multiple time points and then denoise it. Based on the denoised output data of each acceleration sensor at multiple time points, variational mode decomposition is performed on the denoised output data of each acceleration sensor at multiple time points to obtain multiple acceleration intrinsic mode functions and acceleration residuals. The activity characteristics of the patient to be evaluated are determined based on the acceleration intrinsic mode functions and acceleration residuals corresponding to each acceleration sensor through the activity analysis model, where the activity characteristics of the patient to be evaluated may include activity frequency and activity amplitude, etc. The activity analysis model may be a long short-term memory (LSTM) model.

[0149] The brain wave acquisition unit can acquire the brain waves of the patient to be evaluated, and perform signal processing on the brain waves of the patient to be evaluated to extract the brain wave features of the patient to be evaluated. Preferably, the brain wave acquisition unit performs variational mode decomposition on the brain waves of the patient to be evaluated to obtain multiple brain wave intrinsic mode functions and brain wave residuals, extracts the frequency domain features and time domain features of each brain wave intrinsic mode function, and extracts the time domain features and frequency domain features of the brain wave residuals, wherein the brain wave features of the patient to be evaluated may include the frequency domain features and time domain features of each brain wave intrinsic mode function and the time domain features and frequency domain features of the brain wave residuals.

[0150] The delirium assessment module can be used to determine the delirium auxiliary assessment result of the patient to be assessed based on the medical record characteristics of the patient to be assessed, the third knowledge graph and the clinical verification data of the patient to be assessed through the delirium auxiliary assessment model.

[0151] Specifically include:

[0152] The delirium auxiliary assessment result of the patient to be assessed is determined by the delirium auxiliary assessment model based on the medical record characteristics of the patient to be assessed, the third knowledge graph, and at least one of the consciousness characteristics, mood change characteristics, delusion characteristics, activity characteristics, biological marker characteristics and brain wave characteristics of the patient to be assessed.

[0153] The intervention prediction module can be used to determine the delirium intervention plan based on the medical record characteristics of the patient to be evaluated, the clinical verification data of the patient to be evaluated, and the auxiliary evaluation results of delirium of the patient to be evaluated, wherein the delirium intervention plan includes reversible risk factors and non-drug intervention measures corresponding to the reversible risk factors. Specifically, the delirium intervention plan can be determined by the plan determination model based on the medical record characteristics of the patient to be evaluated, the clinical verification data of the patient to be evaluated, and the auxiliary evaluation results of delirium of the patient to be evaluated. Among them, the plan determination model can be a Convolutional Neural Network (CNN) model.

[0154] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A delirium auxiliary prediction and evaluation system based on multi-dimensional data, characterized in that: include: A graph building module, used to obtain electronic medical records of multiple sample patients, and build a first knowledge graph based on the electronic medical records of the multiple sample patients, wherein the first knowledge graph is used to record association information between multiple diseases, drugs and surgeries and delirium; The graph building module is further used to build a second knowledge graph based on the electronic medical records of the multiple sample patients and the first knowledge graph, wherein the second knowledge graph is used to record association information about delirium between any two diseases, any two drugs, any two surgeries, any disease and any drug, any disease and surgery, and any drug and any surgery; The graph building module is further used to build a third knowledge graph based on the electronic medical records of the multiple sample patients, wherein the third knowledge graph is used to record the association between multiple delirium types and onset characteristics; A model building module, used to build a delirium prediction model based on the electronic medical records of the multiple sample patients and the second knowledge graph; The model building module is also used to build a delirium auxiliary assessment model based on the electronic medical records of the multiple sample patients, the second knowledge graph and the third knowledge graph; Medical record acquisition module, used to obtain the electronic medical records of patients to be evaluated; A feature extraction module, configured to extract medical record keyword features of the patient to be evaluated from the electronic medical record of the patient to be evaluated based on the first knowledge graph; A delirium prediction module, configured to determine a delirium prediction result of the patient to be evaluated based on the keyword features of the medical records of the patient to be evaluated and the second knowledge graph through the delirium prediction model, wherein the delirium prediction result includes a risk probability of each delirium type; A scheme generating module, used for determining a clinical verification scheme for the patient to be evaluated based on the delirium prediction result of the patient to be evaluated and the third knowledge graph; A verification acquisition module, used for acquiring clinical verification data of the patient to be evaluated based on the clinical verification scheme of the patient to be evaluated; A delirium assessment module, configured to determine a delirium auxiliary assessment result of the patient to be assessed based on the medical record characteristics of the patient to be assessed, the third knowledge graph, and the clinical verification data of the patient to be assessed through the delirium auxiliary assessment model; An intervention prediction module, used to determine a delirium intervention plan according to the medical record characteristics of the patient to be evaluated, the clinical verification data of the patient to be evaluated and the delirium auxiliary evaluation result of the patient to be evaluated, wherein the delirium intervention plan includes reversible risk factors and non-drug intervention measures corresponding to the reversible risk factors; The graph building module builds a first knowledge graph based on the electronic medical records of the multiple sample patients, including: For each of the sample patients, basic characteristics, disease characteristics, drug characteristics, surgical characteristics and delirium characteristics are extracted from the electronic medical records of the sample patients; Based on the basic characteristics of any two of the sample patients, calculating the similarity of the basic characteristics of any two of the sample patients; Using a clustering algorithm to cluster multiple sample patients according to the similarity of basic characteristics of any two of the sample patients, and determining multiple sample patient clusters; For each of the sample patient clusters, determining association information between multiple diseases, drugs, surgeries and delirium based on disease characteristics, drug characteristics, surgery characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster; Establishing the first knowledge graph based on the association information between the multiple diseases, drugs, surgeries and delirium corresponding to each of the sample patient clusters; The graph establishment module clusters multiple sample patients according to the following algorithm flow to determine multiple sample patient clusters: S11, based on the basic feature similarity of any two sample patients, calculating the variance of the basic feature similarity, determining the number of sample patient clusters according to the variance of the basic feature similarity, and randomly selecting multiple sample patients as multiple cluster centers based on the number of sample patient clusters; S12, for each sample patient, clustering the sample patient to the cluster center with the largest basic feature similarity; S13. For each sample patient cluster, calculate the corresponding basic feature similarity mean of each sample patient included in the sample patient cluster, wherein the corresponding basic feature similarity mean of the sample patient is the mean of the basic feature similarity between the sample patient and each other sample patient included in the sample patient cluster, and take the sample patient with the largest basic feature similarity mean as the new cluster center of the sample patient cluster; S14, determining whether the cluster center of each sample patient cluster has not changed, if so, completing clustering, if not, executing S12; The number of sample patient clusters is calculated based on the following formula: Where n is the number of sample patient clusters, σ similar is the variance of the basic feature similarity, η is a preset parameter, and η is greater than 0, and n0 is the number of preset sample patient clusters; The atlas building module determines the association information between multiple diseases, drugs, surgeries and delirium based on the disease characteristics, drug characteristics, surgery characteristics and delirium characteristics of multiple sample patients included in the sample patient cluster, including: Based on the disease characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, the beta coefficient and odds ratio between each disease and delirium are calculated; based on the beta coefficient and odds ratio between each disease and delirium, the correlation coefficient between each disease and delirium is calculated; Based on the drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, the beta coefficient and odds ratio between each drug and delirium are calculated; based on the beta coefficient and odds ratio between each drug and delirium, the correlation coefficient between each drug and delirium is calculated; Based on the surgical characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, the beta coefficient and odds ratio between each surgery and delirium are calculated; based on the beta coefficient and odds ratio between each surgery and delirium, the correlation coefficient between each surgery and delirium is calculated; Among them, based on regression analysis, the beta coefficient between each disease and delirium was calculated; The odds ratio between each disease and delirium was calculated based on the following formula: Among them, OR (i，j) is the odds ratio between the jth disease and delirium corresponding to the i-th sample patient cluster, N ((i，j)，1) N is the number of sample patients with delirium who have the jth disease in the i-th sample patient cluster, ((i，j)，2) N is the number of sample patients with delirium who do not suffer from the jth disease in the i-th sample patient cluster, ((i，j)，3)为 The number of sample patients with the jth disease and without delirium in the i-th sample patient cluster, N ((i，j)，4) is the number of sample patients in the i-th sample patient cluster who do not suffer from the j-th disease and do not suffer from delirium; The association coefficient between each disease and delirium was calculated based on the following formula: C (i，j) =a 11 ×β (i，j) +a 12 ×(OR (i，j) -1) 2 Among them, C (i，j) is the correlation coefficient between the jth disease corresponding to the i-th sample patient cluster and delirium, a 11 and a 12 All are preset weights, a 11 +a 12 =1, a 11 and a 12 are greater than 0, β (i，j) is the beta coefficient between the jth disease and delirium corresponding to the i-th sample patient cluster; The graph building module builds a second knowledge graph based on the electronic medical records of the plurality of sample patients and the first knowledge graph, including: Based on the disease characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, calculating the joint beta coefficient and the joint odds ratio between any two diseases and delirium, and based on the joint beta coefficient and the joint odds ratio between any two diseases and delirium, calculating the association coefficient between any two diseases and delirium; Based on the drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, calculating the joint beta coefficient and the joint odds ratio between any two drugs and delirium, and based on the joint beta coefficient and the joint odds ratio between any two drugs and delirium, calculating the association coefficient between any two drugs and delirium; Based on the surgical characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, calculating the joint beta coefficient and the joint odds ratio between any two surgeries and delirium, and based on the joint beta coefficient and the joint odds ratio between any two surgeries and delirium, calculating the correlation coefficient between any two surgeries and delirium; Based on the disease characteristics, drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, calculating the joint beta coefficient and the joint odds ratio between any disease and any drug and delirium, and based on the joint beta coefficient and the joint odds ratio between any disease and any drug and delirium, calculating the correlation coefficient between any disease and any drug and delirium; Based on the disease characteristics, surgical characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, the joint beta coefficient and the joint odds ratio between any disease and any surgery and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any disease and any surgery and delirium, the correlation coefficient between any disease and any surgery and delirium is calculated; Based on the surgical characteristics, drug characteristics and delirium characteristics of the multiple sample patients included in the sample patient cluster, the joint beta coefficient and the joint odds ratio between any surgery and any drug and delirium are calculated; based on the joint beta coefficient and the joint odds ratio between any surgery and any drug and delirium, the correlation coefficient between any surgery and any drug and delirium is calculated; The second knowledge graph is established based on the correlation coefficient between any two diseases and delirium, the correlation coefficient between any two drugs and delirium, the correlation coefficient between any two surgeries and delirium, the correlation coefficient between any disease and any drug and delirium, the correlation coefficient between any disease and any surgery and delirium, and the correlation coefficient between any surgery and any drug and delirium; Among them, based on regression analysis, the beta coefficient between any two diseases and delirium was calculated; The joint odds ratio between any two associated diseases and delirium was calculated based on the following formula: Among them, OR (i,(e,f)) is the joint odds ratio between the e-th associated disease and the f-th associated disease and delirium corresponding to the i-th sample patient cluster, N (i,(e,f),1) N is the number of sample patients with delirium who suffer from both the e-th associated disease and the f-th associated disease in the i-th sample patient cluster, (i,(e,f),2) N is the number of sample patients with delirium who do not suffer from the e-th associated disease and the f-th associated disease at the same time in the i-th sample patient cluster, (i,(e,f),3) N is the number of sample patients in the i-th sample patient cluster who have both the e-th associated disease and the f-th associated disease and do not suffer from delirium. (i,(e,f),4) is the number of sample patients in the i-th sample patient cluster who do not suffer from both the e-th associated disease and the f-th associated disease and do not suffer from delirium; The association coefficient between any two related diseases and delirium is calculated in the same way as the association coefficient between any disease and delirium.

2. The delirium auxiliary prediction and evaluation system based on multi-dimensional data according to claim 1, characterized in that: The atlas building module extracts basic features, disease features, drug features, surgical features and delirium features from the electronic medical records of the sample patients, including: Establish a medical terminology database, a stop word database, and a stop word exception rule database; Based on the medical terminology word library, the stop word library and the stop word exception rule library, invalid stop words in the electronic medical records of the sample patients are removed to generate a first word segmentation result; Performing maximum string frequency matching on the first word segmentation result to generate a second word segmentation result; Generate a third word segmentation result based on the second word segmentation result through a long short-term memory network; According to the third word segmentation result, the basic characteristics, disease characteristics, drug characteristics, surgery characteristics and delirium characteristics are extracted.

3. A delirium auxiliary prediction and evaluation system based on multi-dimensional data according to claim 1 or 2, characterized in that: The graph building module builds a third knowledge graph based on the electronic medical records of the multiple sample patients, including: For each of the sample patient clusters, based on the delirium characteristics of any two sample patients included in the sample patient cluster, the similarity of the delirium characteristics of any two sample patients is calculated, and a clustering algorithm is used to cluster a plurality of sample patients according to the similarity of the delirium characteristics of any two sample patients to determine a plurality of sample patient cluster groups; For each of the sample patient cluster groups, based on the delirium characteristics of each sample patient included in the sample patient cluster group, calculating the correlation coefficient between each onset characteristic and the sample patient cluster group, and based on the correlation coefficient between each onset characteristic and the sample patient cluster group, determining the associated onset characteristic corresponding to the sample patient cluster group; The third knowledge graph is established based on the associated disease characteristics corresponding to each of the sample patient cluster groups.

4. The delirium auxiliary prediction and evaluation system based on multi-dimensional data according to claim 3, characterized in that: The feature extraction module extracts the medical record keyword features of the patient to be evaluated from the electronic medical record of the patient to be evaluated based on the first knowledge graph, including: Extracting basic characteristics of the patient to be evaluated from the electronic medical records of the patient to be evaluated; Based on the basic characteristics of the patient to be evaluated, determining a target sample patient cluster; Based on the first knowledge graph, determine the medical record keywords corresponding to the target sample patient cluster; Based on the medical record keywords corresponding to the target sample patient cluster, medical record keyword features of the patient to be evaluated are extracted from the electronic medical records of the patient to be evaluated.

5. The delirium auxiliary prediction and evaluation system based on multi-dimensional data according to claim 4, characterized in that: The scheme generation module determines a clinical verification scheme for the patient to be evaluated based on the delirium prediction result of the patient to be evaluated and the third knowledge graph, including: Determine a target sample patient cluster group based on the delirium prediction result of the patient to be evaluated, the target sample patient cluster cluster, and the third knowledge graph; Based on the associated disease characteristics corresponding to the target sample patient cluster group, a clinical verification plan for the patient to be evaluated is determined.

6. The delirium auxiliary prediction and evaluation system based on multi-dimensional data according to claim 5, characterized in that: The verification acquisition module comprises: An automatic dialogue unit, used to conduct a dialogue with the patient to be evaluated through an automatic dialogue model, and determine the consciousness characteristics of the patient to be evaluated based on the dialogue information of the patient to be evaluated; An emotion recognition unit, used to recognize emotion change characteristics of the patient to be evaluated based on the conversation information of the patient to be evaluated; a delusion recognition unit, configured to obtain the video and voice of the patient to be evaluated, identify abnormal movement characteristics of the patient to be evaluated based on the video of the patient to be evaluated, identify objection voice characteristics of the patient to be evaluated based on the video and voice of the patient to be evaluated, and determine the delusion characteristics of the patient to be evaluated based on the abnormal movement characteristics and abnormal voice characteristics of the patient to be evaluated; An activity sensing unit, comprising a plurality of acceleration sensors and an activity analysis component arranged at different positions of the patient to be evaluated, wherein the activity analysis component is used to determine the activity characteristics of the patient to be evaluated based on output data of the plurality of acceleration sensors at a plurality of time points; A biomarker acquisition unit, used to acquire the biomarker characteristics of the patient to be evaluated; The brain wave acquisition unit is used to acquire the brain wave characteristics of the patient to be evaluated.

7. The delirium auxiliary prediction and evaluation system based on multi-dimensional data according to claim 6, characterized in that: The delirium assessment module determines the delirium auxiliary assessment result of the patient to be assessed through the delirium auxiliary assessment model based on the medical record characteristics of the patient to be assessed, the third knowledge graph and the clinical verification data of the patient to be assessed, including: The delirium auxiliary assessment result of the patient to be assessed is determined by the delirium auxiliary assessment model based on the medical record characteristics of the patient to be assessed, the third knowledge graph, and at least one of the consciousness characteristics, mood change characteristics, delusion characteristics, activity characteristics, biological marker characteristics, and brain wave characteristics of the patient to be assessed.

Images