A method, device and equipment for early warning detection of severe diseases of digestive system

Abstract

The application relates to the technical field of disease prediction, in particular to a digestive system severe disease early warning detection method, device and equipment, wherein the method comprises the following steps: acquiring patient multi-modal data information, wherein the patient multi-modal data information comprises clinical symptoms, laboratory examination and imaging examination; a clustering model is constructed and trained to obtain a plurality of clustering sets; the patient multi-modal data information is input into a target clustering set in the clustering model; whether the patient multi-modal data is abnormal is judged; if the patient multi-modal data information is abnormal, a probability model is constructed; the target clustering set is generated to generate a severity label and is input into the probability model to obtain a probability result of severe disease severity evaluation; hierarchical early warning is carried out according to the probability result; and clinical decision is made according to the hierarchical early warning. Therefore, the problems that a same method cannot be used to evaluate a plurality of digestive system diseases, the evaluation results of diseases are different, and diseases are not timely evaluated in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of disease prediction technology, and in particular to a method, device and equipment for early warning detection of severe digestive system diseases. Background Technology

[0002] Severe digestive system diseases are the most common in clinical practice and are a major category of diseases that endanger people's lives and health.

[0003] For the management of critically ill patients, proposing scientific and standardized rescue procedures and strengthening basic and psychological nursing care can improve the success rate of emergency and critical care rescue.

[0004] In related technologies, the assessment of severe digestive system diseases mainly uses scoring systems, such as the CTSI score, SOFA score, and MELD score. Each scoring system is applicable to different disease types, and currently no single scoring system is suitable for all four common severe digestive system diseases. Therefore, before using a scoring system, gastroenterologists need to determine the exact type of severe disease based on the patient's symptoms and examination data, and then select the appropriate scoring system to assess the severity of the disease. Clinically, when determining the type of disease, on the one hand, isolated indicators are often relied upon as the basis for judgment, such as diagnosing acute pancreatitis as serum amylase levels three times higher than the upper limit of normal. This method has relatively poor specificity, as approximately 20% of non-acute pancreatitis patients also have elevated serum amylase levels. On the other hand, differences in judgment among physicians of different experience levels regarding the same case also affect the selection of scoring systems, hindering the assessment and treatment of the patient's condition and wasting medical resources.

[0005] Furthermore, for severe digestive system diseases, it is crucial to identify high-risk patients as early as possible and prevent mild or moderate cases from progressing to severe illness. The aforementioned manual assessment of the condition is time-consuming in practice, hindering early intervention. Summary of the Invention

[0006] This application provides a method for early warning detection of severe digestive system diseases, in order to solve the problems in the prior art, such as the inability to use the same method to evaluate multiple types of digestive system diseases, the inconsistency of results when evaluating diseases, and the lack of timeliness.

[0007] The first aspect of this application provides a method for early warning detection of severe digestive system diseases, comprising the following steps: acquiring multimodal patient data information, wherein the multimodal patient data information includes clinical symptoms, laboratory tests, and imaging examinations; constructing and training a clustering model to obtain multiple cluster sets; inputting the multimodal patient data information into a target cluster set in the clustering model; determining whether the multimodal patient data is abnormal; wherein, if the multimodal patient data information is abnormal, constructing a probability model; generating severity labels for the target cluster sets and inputting them into the probability model to obtain a probability result for severe disease severity assessment; performing graded early warning based on the probability result; and making clinical decisions based on the graded early warning.

[0008] Optionally, the step of inputting patient multimodal data information into the target cluster set in the clustering model includes: constructing a similarity matrix based on the patient multimodal data information to obtain multimodal similarity matrix data; fusing the multimodal similarity matrix data to generate a degree matrix; constructing a Laplacian matrix based on the similarity matrix and the degree matrix; performing eigenvalue decomposition on the Laplacian matrix to obtain feature vectors; performing K-means clustering on the feature vectors to obtain the final category label; and outputting the target clustering result.

[0009] Optionally, the formula for the similarity matrix is:

[0010] The similarity matrix for clinical symptom data is as follows:

[0011]

[0012] in, It is the set of symptoms of patient i. Let j represent the set of symptoms of patient j, and let elements be... This indicates the similarity of clinical symptoms between the i-th patient and the j-th patient;

[0013] The similarity matrix for laboratory test data is:

[0014]

[0015] in, It's a bandwidth parameter. This represents the set of laboratory test data for the i-th patient. Let j represent the set of laboratory test data for the j-th patient, with elements... This indicates the similarity of laboratory test data between the i-th patient and the j-th patient;

[0016] The similarity matrix for imaging examination data is:

[0017]

[0018] Where B is the number of bins in the histogram. It is the frequency of the i-th image in the K-th bin. It is the frequency of the j-th image in the k-th bin; element This indicates the similarity of imaging data between the i-th patient and the j-th patient.

[0019] Optionally, the formula for the multimodal similarity matrix is:

[0020]

[0021]

[0022] Where M is 3, it indicates that there are 3 modalities of data. It represents the weights of the similarity matrix for the m-th modality.

[0023] Optionally, the formula for the degree matrix is:

[0024]

[0025] The formula for the Laplace matrix is:

[0026] ,

[0027]

[0028] The objective function formula for clustering is:

[0029]

[0030] Where Tr represents the summation of the diagonals, and F is the eigenvector extracted from the Laplacian matrix and then regularized. Let F be the transpose of F, and I be the unit diagonal matrix.

[0031] Optionally, the generation of severity labels for the target cluster set includes: performing secondary clustering on the target cluster set to obtain severity labels, wherein the severity labels include primary labels, secondary labels, and tertiary labels.

[0032] Optionally, the probability model includes:

[0033]

[0034]

[0035]

[0036] Where M is the total number of decision trees, and K is the number of classifications, set to 3. It is the prediction contribution of the m-th decision tree to the class K of sample x. It is the sum of the prediction results of M decision trees. It is the probability that sample x belongs to class k.

[0037] Optionally, the step of providing graded early warning based on probability results includes: determining an initial disease severity category based on the probability results output in the probability model, wherein the initial disease severity category includes level 1 disease, level 2 disease, and level 3 disease, with a priority set as level 1 disease > level 2 disease, and level 2 disease > level 3 disease; determining a first target value, a second target value, and a third target value based on the probability results of the level 1 disease, level 2 disease, and level 3 disease; if the first target value is greater than or equal to a first preset value, then the final disease severity category is determined; otherwise, it is determined whether the sum of the first target value and the second target value is greater than or equal to a second preset value: wherein, if the sum of the first target value and the second target value is greater than or equal to the second preset value, then the final disease severity category is determined. The system determines the disease severity category based on the first target value and the second target value. If the difference is greater than or equal to a third preset value, the disease severity category of the first target value is determined as the final disease severity category. Otherwise, the final disease severity category is determined according to the priority of the disease severity categories of the first and second target values. If the sum of the first and second target values ​​is less than a second preset value, a level 2 disease is determined as the final disease severity category. The system then determines the warning level based on the final disease severity category. If the final disease severity category is level 1, a level 1 warning is activated. If the final disease severity category is level 2, a level 2 warning is activated. If the final disease severity category is level 3, a level 3 warning is activated.

[0038] A second aspect of this application provides a device for early warning and detection of severe digestive system diseases, comprising: an acquisition module for acquiring multimodal patient data information, wherein the multimodal patient data information includes clinical symptoms, laboratory tests, and imaging examinations; and a construction module for constructing and training a clustering model to obtain multiple cluster sets, inputting the multimodal patient data information into a target cluster set in the clustering model, and determining whether the multimodal patient data is abnormal, wherein if the multimodal patient data information is abnormal, a probability model is constructed, and severity labels are generated from the target cluster sets and input into the probability model to obtain a probability result for assessing the severity of the severe disease.

[0039] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform a method for early warning detection of severe digestive system diseases as described in the above embodiments.

[0040] The beneficial effects achieved by employing the present invention are as follows: By integrating multimodal data such as patients' clinical symptoms, laboratory tests, and imaging examinations, the embodiments of this application achieve a comprehensive assessment of patients' health status, overcoming the limitations of using a single data source and significantly improving the accuracy and reliability of disease assessment. The use of spectral clustering to construct a clustering model for automatic classification reduces the workload of physicians in manually classifying diseases, while also shortening examination time, thus providing patients with more time for treatment. The use of a probabilistic model to assess the severity of patients' diseases provides clinicians with objective and quantitative assessment criteria, helping to differentiate patients with different degrees of severity and optimize the allocation of medical resources. The tiered early warning system based on probabilistic results helps gastroenterologists make rapid clinical decisions, improving treatment efficiency and patient prognosis. Therefore, it solves the problems in the prior art, such as the inability to use the same method to assess multiple types of digestive system diseases, discrepancies in assessment results, and delays in timeliness.

[0041] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0042] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0043] Figure 1 This is a flowchart of a method for early warning detection of severe digestive system diseases according to an embodiment of this application;

[0044] Figure 2 This is a flowchart illustrating the probability results of obtaining a severity assessment of a severe digestive system disease, according to an embodiment of this application.

[0045] Figure 3 This is a flowchart of a clustering model construction method provided according to an embodiment of this application;

[0046] Figure 4 This is a flowchart illustrating a method for determining a target value based on probability model results, according to an embodiment of this application.

[0047] Figure 5 This is a flowchart illustrating a method for determining a final disease severity category according to an embodiment of this application;

[0048] Figure 6 This is a flowchart of a graded early warning method for acute pancreatitis based on disease clustering results provided in an embodiment of this application;

[0049] Figure 7This is a flowchart of a graded early warning method for gastrointestinal bleeding based on disease clustering results provided in an embodiment of this application;

[0050] Figure 8 This is a flowchart of a graded early warning method for disease clustering results of enterogenic sepsis according to an embodiment of this application;

[0051] Figure 9 This is a flowchart of a graded early warning method for disease clustering results of acute-on-chronic liver failure, provided according to an embodiment of this application;

[0052] Figure 10 This is a schematic diagram of the structure of a critical digestive system disease early warning detection device provided in an embodiment of this application.

[0053] Figure 11 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application.

[0054] Explanation of the symbols in the attached figure: P(1): probability that the patient has a level 1 disease; P(2): probability that the patient has a level 2 disease; P(3): probability that the patient has a level 3 disease. Detailed Implementation

[0055] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0056] The following description, with reference to the accompanying drawings, illustrates an embodiment of a method and device for early warning detection of severe digestive system diseases according to this application. Addressing the issues mentioned in the background art, such as the long time required for manual assessment of illness, the inability to identify the type of severe digestive system disease early, and the inability to timely assess the severity of the patient's disease, this application provides an early warning detection method for severe digestive system diseases. This method integrates multimodal data, including the patient's clinical symptoms, laboratory tests, and imaging examinations, to achieve a comprehensive assessment of the patient's health status, overcoming the limitations of using a single data source and significantly improving the accuracy and reliability of disease assessment. It utilizes spectral clustering to construct a clustering model for automatic classification, reducing the workload of physicians in manually classifying diseases and shortening examination time, thus allowing patients more time for treatment. Furthermore, it uses a probabilistic model to assess the severity of the patient's disease, providing clinicians with objective and quantitative assessment data, helping to differentiate patients with different degrees of severity and optimize the allocation of medical resources. The tiered early warning system based on probabilistic results can help gastroenterologists make rapid clinical decisions, improve treatment efficiency, and improve patient prognosis. This solves the problems in existing technologies, such as the inability to use the same method to assess multiple types of digestive system diseases, discrepancies in assessment results, and delays in assessment.

[0057] Specifically, Figure 1 This is a flowchart illustrating a method for early warning detection of severe digestive system diseases provided in an embodiment of this application.

[0058] like Figure 1 As shown, this method for early warning detection of severe digestive system diseases includes the following steps:

[0059] In step S101, patient multimodal information is acquired.

[0060] The patient's multimodal data information includes clinical symptoms, laboratory tests, and imaging examinations. Clinical symptoms, often referred to as clinical manifestations, refer to a series of abnormal changes in the body after the body develops a disease, and can serve as an important basis for disease diagnosis. Laboratory tests refer to physical or chemical examinations conducted in the laboratory to determine the contents, properties, concentration, quantity, and other characteristics of the submitted substances. Imaging examinations refer to the use of media such as ultrasound and X-rays to visualize the structure and density of human tissues and organs in the form of images, used to determine whether there are lesions in a local area.

[0061] Specifically, clinical symptoms include heart rate, blood pressure, body temperature, bowel movements, abdominal pain, fatigue, bleeding, hematemesis, dyspnea, and syncope. Laboratory tests include complete blood count (white blood cell count and differential, red blood cell count and hemoglobin, platelet count), C-reactive protein, serum amylase, lipase, blood glucose, urine amylase, liver function tests (direct bilirubin, indirect bilirubin), and kidney function tests (creatinine, urine output). Imaging studies include upright abdominal X-ray, abdominal CT scan, and abdominal ultrasound.

[0062] It is understood that the embodiments of this application can obtain and analyze common clinical symptoms, laboratory test and imaging data indicators of severe digestive system diseases, thereby obtaining similar characteristics between diseases, which facilitates subsequent clustering and evaluation.

[0063] In step S102, a clustering model is constructed and trained to obtain multiple cluster sets. The patient's multimodal data information is input into the target cluster set in the clustering model to determine whether the patient's multimodal data is abnormal. If the patient's multimodal data information is abnormal, a probability model is constructed, and the severity label of the target cluster set is generated and input into the probability model to obtain the probability result of the severity assessment of the severe disease.

[0064] Specifically, such as Figure 2 As shown, clustering analysis is performed using features from multimodal data. The clustering model is constructed using a multimodal spectral clustering algorithm. The target cluster set is the output of the clustering model and is divided into six categories: one normal set and five abnormal sets. The abnormal sets consist of four common severe digestive system diseases and other digestive system diseases. The four common severe digestive system diseases are acute pancreatitis, gastrointestinal bleeding, enterogenic sepsis, and acute-on-chronic liver failure. The probabilistic model is constructed using the LightGBM model, an efficient gradient boosting tree framework developed by Microsoft. Compared to XGBoost and CatBoost algorithms, LightGBM is more suitable for large-scale data and high-dimensional features, and is more applicable to medical scenarios. In this embodiment, LightGBM is used as the probabilistic model to output the probabilities of three categories, facilitating subsequent secondary evaluation of the diseases.

[0065] It is understood that the embodiments of this application, by fusing multimodal data information for cluster analysis, can describe the same object from different perspectives, compensate for the noise and information gaps of a single modality, and improve the comprehensive characterization of patient features. Robustness is enhanced through multimodal cross-validation. Using the disease classification results as probability output provides a measure of uncertainty and supports subsequent probability calibration, offering more refined decision support.

[0066] In this embodiment of the application, inputting patient multimodal data information into the target cluster set in the clustering model includes: constructing a similarity matrix based on the patient multimodal data information to obtain multimodal similarity matrix data; fusing the multimodal similarity matrix data to generate a degree matrix; constructing a Laplacian matrix based on the similarity matrix and the degree matrix; performing eigenvalue decomposition on the Laplacian matrix to obtain feature vectors; performing K-means clustering on the feature vectors to obtain the final category label; and outputting the target clustering result.

[0067] Specifically, such as Figure 3 As shown, the similarity matrix is ​​used to quantify the similarity between samples. When constructing the similarity matrix based on multimodal patient data, the Jaccard similarity coefficient is used for clinical symptom data. For laboratory test data, a Gaussian kernel function generated by Euclidean distance is used. For imaging test data, the histogram intersection similarity method is used. A weighted average method is used to fuse the three similarity matrices to obtain the multimodal similarity matrix. During the weighted linear fusion process, cross-validation is used to optimize the weights. The degree matrix is ​​a diagonal matrix derived from the similarity matrix and is used to describe the connection strength of each node in the graph. The degree matrix is ​​generated based on the multimodal similarity matrix. The Laplacian matrix is ​​used to describe the structural features of the graph. A normalized Laplacian matrix is ​​constructed based on the degree matrix, and the eigenvectors F corresponding to the first K smallest eigenvalues ​​of the Laplacian matrix are calculated. In this embodiment, K is set to 6. After regularization, the eigenvectors F are clustered using the K-means clustering algorithm, outputting K cluster sets.

[0068] It is understood that the embodiments of this application employ Jaccard similarity coefficient, Gaussian kernel function generated by Euclidean distance, and histogram intersection similarity method to construct similarity matrices for multimodal data, respectively. Jaccard similarity coefficient is suitable for binary features; for clinical symptom data, symptomatic data is represented by 1, and asymptomatic data by 0, with symptom severity expressed numerically, effectively measuring the similarity of sets. Laboratory test data is generally represented by specific numerical values; the Gaussian kernel function generated by Euclidean distance is sensitive to data and can smooth similarity through the kernel function. Imaging examination data outputs a set of images; the histogram intersection similarity method is suitable for matching the distribution of image data.

[0069] This application improves the accuracy of the modal similarity matrix by customizing the processing of each modality's data. Cross-validation is used to optimize weights, ensuring a scientifically reasonable contribution of different modalities and avoiding biases caused by subjective weight settings, thus enhancing the reliability of the fused multimodal similarity matrix. The transformation from similarity matrix to degree matrix and then to Laplacian matrix converts the data into a graph structure, using multimodal spectral clustering to capture global relationships between samples and uncover inherent similarities. Selecting the top K smallest eigenvectors of the Laplacian matrix effectively extracts the low-dimensional embedding representation of the data, removing noise while preserving the main structure. K-means clustering is computationally efficient in low-dimensional space; regularizing the eigenvectors before clustering effectively improves the K-means clustering effect.

[0070] In this embodiment, the formula for the similarity matrix is:

[0071] The similarity matrix for clinical symptom data is as follows:

[0072]

[0073] in, It is the set of symptoms of patient i. Let j represent the set of symptoms of patient j, and let elements be... This indicates the similarity of clinical symptoms between the i-th patient and the j-th patient;

[0074] The similarity matrix for laboratory test data is:

[0075]

[0076] in, It's a bandwidth parameter. This represents the set of laboratory test data for the i-th patient. Let j represent the set of laboratory test data for the j-th patient, with elements... This indicates the similarity of laboratory test data between the i-th patient and the j-th patient;

[0077] The similarity matrix for imaging examination data is:

[0078]

[0079] Where B is the number of bins in the histogram. It is the frequency of the i-th image in the K-th bin. It is the frequency of the j-th image in the k-th bin; element This indicates the similarity of imaging data between the i-th patient and the j-th patient.

[0080] In this embodiment of the application, multimodal similarity matrix data is obtained, wherein the formula for the multimodal similarity matrix is:

[0081]

[0082]

[0083] Where M is 3, it indicates that there are 3 modalities of data. It represents the weights of the similarity matrix for the m-th modality.

[0084] In this embodiment of the application, the formula for the degree matrix is:

[0085]

[0086] The formula for the Laplace matrix is:

[0087] ,

[0088]

[0089] The objective function formula for clustering is:

[0090]

[0091] Where Tr represents the summation of the diagonals, and F is the eigenvector extracted from the Laplacian matrix and then regularized. Let F be the transpose of F, and I be the unit diagonal matrix.

[0092] In this embodiment of the application, generating severity labels for the target cluster set includes: performing secondary clustering on the target cluster set to obtain severity labels, wherein the severity labels include first-level labels, second-level labels, and third-level labels.

[0093] In this embodiment, the K-means method is used to perform secondary clustering on the target cluster set. In this embodiment, K is set to 3, resulting in three cluster sets: Level 1 severity, Level 2 severity, and Level 3 severity. Based on the cluster set to which the patient belongs, Level 1, Level 2, and Level 3 labels are assigned to the patient data.

[0094] It is understood that the embodiments of this application perform secondary clustering on the target cluster set. The first clustering classifies the diseases, but does not directly correspond to the severity of the diseases. Secondary clustering of the target clusters can further summarize severity labels, making the data suitable for subsequent supervised learning and achieving a smooth transition from unsupervised to supervised learning. In addition, the severity labels generated by secondary clustering reduce the subjective bias of manual annotation and intuitively reflect the true level of disease based on data distribution.

[0095] In this application embodiment, the probability model includes:

[0096]

[0097]

[0098]

[0099] Where M is the total number of decision trees, and K is the number of classifications, set to 3. It is the prediction contribution of the m-th decision tree to the class K of sample x. It is the sum of the prediction results of M decision trees. This represents the probability that sample x belongs to class k. The probabilistic model is built using the LightGBM model, a gradient boosting framework based on decision trees. Gradient boosting is an iterative decision tree ensemble method where a new decision tree is trained in each iteration to fit the residuals of the previous iteration, ultimately resulting in a model capable of accurate classification. The predicted values ​​output by the model are the original predicted values. The Softmax normalization method is used to convert the original predicted values ​​into probabilities, and the sum of all probabilities is 1.

[0100] Specifically, there are three types of classifications: first-level tags, second-level tags, and third-level tags. If... The original predicted value of a sample is [1.2, 0.5, -0.8]. After transformation using the Softmax normalization method, the probability is [0.62, 0.30, 0.08]. This means that the probability of the sample being a first-level label is 0.62, the probability of being a second-level label is 0.30, the probability of being a third-level label is 0.08, and the sum of the predicted probabilities of the three categories is 1.

[0101] It is understood that this application's embodiments output probability values ​​for patient data as primary, secondary, and tertiary labels using the LightGBM probability model. Primary labels correspond to primary diseases, secondary labels to secondary diseases, and tertiary labels to tertiary diseases, enabling rapid assessment of the severity of a patient's condition and providing data support for subsequent graded early warning. These probability results provide doctors with objective reference points, enabling them to develop personalized rehabilitation plans for patients and improve their recovery outcomes.

[0102] In step S103, a graded early warning is issued based on the probability results of the severity assessment of the critical illness, and clinical decisions are made based on the graded early warning.

[0103] Among them, the probability results of the severity assessment of severe diseases are three-category probability results. The three-category probability results are then evaluated a second time to determine the final disease severity category. Each disease severity category corresponds to a warning level, and the warning plan is activated based on the final disease severity category.

[0104] Specifically, if the clustering result is an anomaly, and the anomaly type is acute pancreatitis, the probability output in the probability model for the three categories is: Level 1 disease probability 0.2, Level 2 disease probability 0.5, and Level 3 disease probability 0.3. Then, according to the evaluation process, the final disease severity category is determined to be Level 2, and the Level 2 early warning plan is activated. For acute pancreatitis, a severe digestive system disease, the Level 2 early warning plan involves triage to a general ward for relevant supportive treatment.

[0105] Understandably, the embodiments of this application incorporate probabilistic uncertainty management. Based on the probability results output by the probabilistic model, a secondary evaluation is performed on the probability results to determine the final disease severity category, reducing the risk of misdiagnosis. A multi-level early warning mechanism is employed to optimize resource allocation and improve the accuracy of clinical decision-making.

[0106] In this embodiment, a graded early warning system based on probability results is implemented, including: setting a priority level of level 1 disease greater than level 2 disease, and level 2 disease greater than level 3 disease; determining a first target value, a second target value, and a third target value based on the probability results of level 1, level 2, and level 3 diseases; if the first target value is greater than or equal to a first preset value, then determining the final disease severity category; otherwise, determining whether the sum of the first target value and the second target value is greater than or equal to a second preset value; wherein, if the sum of the first target value and the second target value is greater than or equal to the second preset value, then determining whether the difference between the first target value and the second target value is greater than or equal to a third preset value. If the difference between the target value and the second target value is greater than or equal to the third preset value, then the disease severity category of the first target value is determined as the final disease severity category; otherwise, the final disease severity category is determined according to the priority of the disease severity categories of the first target value and the second target value. Otherwise, if the sum of the first target value and the second target value is less than the second preset value, then the second-level disease is determined as the final disease severity category. The warning level is determined according to the final disease severity category. If the final disease severity category is a first-level disease, a first-level warning is activated; if the final disease severity category is a second-level disease, a second-level warning is activated; if the final disease severity category is a third-level disease, a third-level warning is activated.

[0107] Specifically, such as Figure 4 As shown, if the probability model outputs three classification probabilities of [0.62, 0.30, 0.08], then the first target value is determined to be 0.62, the second target value to be 0.30, and the third target value to be 0.08. Figure 5As shown, the first preset value is 0.45, the second preset value is 0.7, and the third preset value is 0.1. Because the first target value is greater than the first preset value, the disease severity category of the first target value is determined as the final disease severity category, that is, the final disease severity category is determined as a level one disease, and the corresponding level one early warning scheme for the disease is activated.

[0108] It is understood that the embodiments of this application evaluate the three-class classification probability results using multiple preset values.

[0109] A priority system is also used. When two probabilities are close, the one with higher priority is chosen to avoid delaying treatment and causing the patient to develop a severe illness. This application's embodiments are not only applicable to situations where the probability results are clear, but also to situations where the probability results of two or three categories tend to be balanced. Based on the probability differences and disease priority, an appropriate disease level is selected as the final disease severity level, improving the patient's treatment effect and survival rate.

[0110] According to the embodiments of this application, a method for early warning and detection of severe digestive system diseases is proposed. This method integrates multimodal data such as clinical symptoms, laboratory tests, and imaging examinations to achieve a comprehensive assessment of the patient's health status, overcoming the limitations of using a single data source and significantly improving the accuracy and reliability of disease assessment. It utilizes spectral clustering to construct a clustering model for automatic classification, reducing the workload of physicians in manually classifying diseases and shortening examination time, thus allowing patients more time for treatment. A probabilistic model is used to assess the severity of the patient's disease, providing clinicians with objective and quantitative assessment criteria, helping to differentiate patients with different degrees of severity and optimize the allocation of medical resources. A tiered early warning system based on probabilistic results can help gastroenterologists make rapid clinical decisions, improve treatment efficiency, and improve patient prognosis. Therefore, this method solves the problems in existing technologies, such as the inability to use the same method to assess multiple types of digestive system diseases, inconsistencies in assessment results, and delays in timeliness.

[0111] The following is a detailed description of a method for early warning detection of severe digestive system diseases:

[0112] S1. Obtain multimodal information about patients.

[0113] S2. Input the patient's multimodal information into the clustering model and obtain the target clustering results.

[0114] Specific steps are as follows Figure 3 As shown, if the clustering result is normal, it indicates that the patient's digestive system is normal, and the program terminates. If the clustering result is abnormal and indicates other diseases, these are not considered as disease objects in this model, and a doctor will conduct a disease assessment, at which point the program terminates.

[0115] S3. If the clustering result is one of the following: acute pancreatitis, gastrointestinal bleeding, enterogenic sepsis, or acute-on-chronic liver failure, then the patient data is input into the secondary clustering model to generate severity labels. Based on the severity of the disease, the severity labels are categorized as Level 1, Level 2, and Level 3.

[0116] S4. Input patient data with severity labels into the probability model and output the probability that the patient has a level 1 disease, level 2 disease, or level 3 disease.

[0117] S5. Obtain the target value from the probabilities of level 1, level 2, and level 3 diseases to determine the final disease severity of the patient.

[0118] S6. Classify and issue early warnings based on the results of a single clustering and the final severity of the disease.

[0119] Specifically, if a patient's clustering result indicates acute pancreatitis, then the tiered early warning scheme is as follows: Figure 6 As shown, if the final disease severity is grade three, a grade three alert is activated, and the patient receives routine drug treatment. If the final disease severity is grade two, a grade two alert is activated, and the patient is triaged to a general ward for relevant supportive treatment. If the final disease severity is grade one, a grade one alert is activated, and vital sign monitoring and maintenance are carried out in the gastroenterology ICU, along with early fluid resuscitation, suppression of inflammatory response, protection of the intestines, and early treatment of the underlying cause.

[0120] If a patient's cluster analysis indicates massive gastrointestinal bleeding, then the tiered early warning scheme is as follows: Figure 7 As shown, if the final disease severity is grade three, a grade three alert is activated, and the patient receives standardized drug treatment and their vital signs are monitored. If the final disease severity is grade two, a grade two alert is activated, and the patient receives volume expansion, blood transfusion, and necessary drug treatment. If the final disease severity is grade one, a grade one alert is activated, and the patient receives volume expansion, blood transfusion, and necessary drug treatment, the endoscopic treatment procedure for massive gastrointestinal bleeding is initiated, an endoscopic diagnosis and treatment decision is made, and emergency surgical treatment is performed.

[0121] If the patient's initial clustering result is enterogenic sepsis, then the tiered early warning scheme is as follows: Figure 8 As shown, if the final disease severity is grade three, a grade three alert is activated, and treatment for the primary disease and medication are initiated. If the final disease severity is grade two, a grade two alert is activated, and the patient is admitted to the hospital for fluid resuscitation. If the final disease severity is grade one, a grade one alert is activated, and the patient is admitted to the ICU for close monitoring of vital signs and comprehensive organ function support. Treatment measures include fluid resuscitation, infection control, nutritional support, and bowel cleansing.

[0122] If a patient's clustering result indicates acute-on-chronic liver failure, then the tiered early warning scheme is as follows: Figure 9 As shown, if the final disease severity is grade III, a grade III alert is activated, and the patient undergoes etiological analysis and receives symptomatic and supportive drug treatment. If the final disease severity is grade II, a grade II alert is activated, and the patient is admitted to the hospital for etiological analysis and symptomatic and supportive drug treatment. Liver function, coagulation function, complete blood count, and blood ammonia levels are closely monitored, and prevention and treatment of complications are implemented. If the final disease severity is grade I, a grade I alert is activated, and the patient is admitted to the ICU for close monitoring of vital signs and supportive treatment. For patients whose condition progressively worsens and is unresponsive to medical treatment, a liver transplant evaluation should be conducted as soon as possible. While awaiting a liver donor, supportive treatment and management of complications are strengthened.

[0123] In summary, this application's embodiments, through the constructed clustering and probability models, differentiate patient conditions based on multimodal data, analyzing the disease type and severity. This pattern-based disease identification can advance the critical illness identification window by 1 to 3 hours. Multi-level early warning schemes have been developed for four common severe digestive system diseases, standardizing the diagnosis and treatment pathways for these diseases and providing more intervention time for critically ill patients. Furthermore, the multi-level early warning scheme enables intelligent triage, improves ICU bed allocation efficiency, avoids excessive testing, and optimizes medical resources.

[0124] It should be noted that the foregoing explanation of an embodiment of a method for early warning detection of severe digestive system diseases also applies to an early warning detection device for severe digestive system diseases in this embodiment, and will not be repeated here.

[0125] Next, referring to the accompanying drawings, an early warning and detection device for severe digestive system diseases according to an embodiment of this application is described.

[0126] Figure 10 This is a schematic diagram of the structure of a critical digestive system disease early warning detection device proposed in an embodiment of this application.

[0127] like Figure 10 As shown, the digestive system critical disease early warning detection device 10 includes: an acquisition module 100, a construction module 200, and an early warning module 300.

[0128] The acquisition module 100 is used to acquire multimodal data information of patients, including clinical symptoms, laboratory tests, and imaging examinations. The construction module 200 is used to construct and train a clustering model to obtain multiple cluster sets. The multimodal data information of patients is input into the target cluster set in the clustering model to determine whether the multimodal data of patients is abnormal. If the multimodal data information of patients is abnormal, a probability model is constructed, and severity labels are generated from the target cluster set and input into the probability model to obtain the probability result of the severity assessment of the severe disease. The early warning module 300 is used to conduct graded early warning based on the probability result and make clinical decisions based on the graded early warning.

[0129] In summary, the method for early warning and detection of severe digestive system diseases proposed in this application integrates multimodal data such as patients' clinical symptoms, laboratory tests, and imaging examinations to achieve a comprehensive assessment of patients' health status. This overcomes the limitations of using a single data source and significantly improves the accuracy and reliability of disease assessment. The method utilizes spectral clustering to construct a clustering model for automatic classification, reducing the workload of physicians in manually classifying diseases and shortening examination time, thus providing patients with more time for treatment. The method uses a probabilistic model to assess the severity of patients' diseases, providing clinicians with objective and quantitative assessment criteria, helping to differentiate patients with different degrees of severity and optimize the allocation of medical resources. The tiered early warning system based on probabilistic results helps gastroenterologists make rapid clinical decisions, improving treatment efficiency and patient prognosis. Therefore, this method solves the problems in existing technologies, such as the inability to use the same method to assess multiple types of digestive system diseases, inconsistencies in assessment results, and delays in timeliness.

[0130] Figure 11 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0131] The memory 1101, the processor 1102, and the computer program stored on the memory 1101 and executable on the processor 1102.

[0132] When the processor 1102 executes the program, it implements the method for early warning and detection of severe digestive system diseases provided in the above embodiments.

[0133] Furthermore, electronic devices also include:

[0134] Communication interface 1103 is used for communication between memory 1101 and processor 1102.

[0135] The memory 1101 is used to store computer programs that can run on the processor 1102.

[0136] The memory 1101 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage.

[0137] If the memory 1101, processor 1102, and communication interface 1103 are implemented independently, then the communication interface 1103, memory 1101, and processor 1102 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized into address buses, data buses, control buses, etc. For ease of representation, Figure 11 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0138] Optionally, in a specific implementation, if the memory 1101, processor 1102, and communication interface 1103 are integrated on a single chip, then the memory 1101, processor 1102, and communication interface 1103 can communicate with each other through an internal interface.

[0139] The processor 1102 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0140] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0141] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0142] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0143] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0144] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for early warning detection of severe digestive system diseases, characterized in that, Includes the following steps: Acquire multimodal patient data, which includes clinical symptoms, laboratory tests, and imaging examinations; A clustering model is constructed and trained to obtain multiple cluster sets. The patient multimodal data information is input into the target cluster set in the clustering model. A similarity matrix is ​​constructed based on the patient multimodal data information to obtain multimodal similarity matrix data. The multimodal similarity matrix data is fused to generate a degree matrix. A Laplace matrix is ​​constructed based on the similarity matrix and the degree matrix. Feature decomposition is performed on the Laplace matrix to obtain feature vectors. K-means clustering is performed on the feature vectors to obtain the final category labels. The target clustering result is output to determine whether the patient multimodal data is abnormal. If the patient multimodal data information is abnormal, a probability model is constructed. Severity labels are generated from the target cluster set and input into the probability model to obtain the probability result of the severity assessment of the severe disease. The formula for the similarity matrix is: The similarity matrix for clinical symptom data is as follows: ； in, It is the set of symptoms of patient i. Let j represent the set of symptoms of patient j, and let elements be... This indicates the similarity of clinical symptoms between the i-th patient and the j-th patient; The similarity matrix for laboratory test data is: ； in, It's a bandwidth parameter. This represents the set of laboratory test data for the i-th patient. Let j represent the set of laboratory test data for the j-th patient, with elements... This indicates the similarity of laboratory test data between the i-th patient and the j-th patient; The similarity matrix for imaging examination data is: ； Where B is the number of bins in the histogram. It is the frequency of the i-th image in the b-th bin. It is the frequency of the j-th image in the b-th bin; element This indicates the similarity of the imaging data between the i-th patient and the j-th patient; The formula for the multimodal similarity matrix is: ； ； Where Z is 3, it indicates that there are 3 modalities of data. These are the weights of the similarity matrix for the z-th modality; The probability model includes: ； ； ； Where M is the total number of decision trees, and K is the number of classifications, set to 3. It is the prediction contribution of the m-th decision tree to the class k of sample x. It is the sum of the prediction results of M decision trees. It is the probability that sample x belongs to class k; Based on the probability results, a graded early warning is issued, and based on the graded early warning, clinical decisions are made.

2. The method for early warning detection of severe digestive system diseases according to claim 1, characterized in that, The formula for the degree matrix is: ； The formula for the Laplace matrix is: ； ； The objective function formula for clustering is: ； Where Tr represents the summation of the diagonals, and F is the eigenvector extracted from the Laplacian matrix and then regularized. Let F be the transpose of F, and I be the unit diagonal matrix.

3. The method for early warning detection of severe digestive system diseases according to claim 1, characterized in that, Generating severity labels for the target cluster set includes: performing secondary clustering on the target cluster set to obtain severity labels, wherein the severity labels include first-level labels, second-level labels, and third-level labels.

4. The method for early warning detection of severe digestive system diseases according to claim 1, characterized in that, Based on the probability results, a tiered early warning system is implemented, including: The initial disease severity category is determined based on the probability results output in the probability model. The initial disease severity category includes level 1 disease, level 2 disease, and level 3 disease, with the priority set as level 1 disease being greater than level 2 disease, and level 2 disease being greater than level 3 disease. Based on the probability results of the first, second, and third levels of disease, a first target value, a second target value, and a third target value are determined. If the first target value is greater than or equal to a first preset value, the final disease severity category is determined; otherwise, it is determined whether the sum of the first target value and the second target value is greater than or equal to a second preset value. If the sum of the first target value and the second target value is greater than or equal to the second preset value, it is determined whether the difference between the first target value and the second target value is greater than or equal to a third preset value. If the difference between the first target value and the second target value is greater than or equal to the third preset value, the disease severity category of the first target value is determined as the final disease severity category; otherwise, the final disease severity category is determined according to the priority of the disease severity categories of the first target value and the second target value. Otherwise, if the sum of the first target value and the second target value is less than the second preset value, the second level of disease is determined as the final disease severity category. The warning level is determined based on the final disease severity category. If the final disease severity category is Level 1, a Level 1 warning is activated; if the final disease severity category is Level 2, a Level 2 warning is activated; and if the final disease severity category is Level 3, a Level 3 warning is activated.

5. A device for early warning and detection of severe digestive system diseases, characterized in that, include: The acquisition module is used to acquire multimodal patient data information, wherein the multimodal patient data information includes clinical symptoms, laboratory tests, and imaging examinations; A construction module is used to build and train a clustering model to obtain multiple cluster sets. The patient multimodal data information is input into the target cluster set in the clustering model. A similarity matrix is ​​constructed based on the patient multimodal data information to obtain multimodal similarity matrix data. The multimodal similarity matrix data is fused to generate a degree matrix. A Laplace matrix is ​​constructed based on the similarity matrix and the degree matrix. Feature decomposition is performed on the Laplace matrix to obtain feature vectors. K-means clustering is performed on the feature vectors to obtain the final category labels. The target clustering result is output to determine whether the patient multimodal data is abnormal. If the patient multimodal data information is abnormal, a probability model is constructed. Severity labels are generated from the target cluster set and input into the probability model to obtain the probability result of severe disease severity assessment. The formula for the similarity matrix is: The similarity matrix for clinical symptom data is as follows: ； in, It is the set of symptoms of patient i. Let j represent the set of symptoms of patient j, and let elements be... This indicates the similarity of clinical symptoms between the i-th patient and the j-th patient; The similarity matrix for laboratory test data is: ； in, It's a bandwidth parameter. This represents the set of laboratory test data for the i-th patient. Let j represent the set of laboratory test data for the j-th patient, with elements... This indicates the similarity of laboratory test data between the i-th patient and the j-th patient; The similarity matrix for imaging examination data is: ； Where B is the number of bins in the histogram. It is the frequency of the i-th image in the b-th bin. It is the frequency of the j-th image in the b-th bin; element This indicates the similarity of the imaging data between the i-th patient and the j-th patient; The formula for the multimodal similarity matrix is: ； ； Where Z is 3, it indicates that there are 3 modalities of data. These are the weights of the similarity matrix for the z-th modality; The probability model includes: ； ； ； Where M is the total number of decision trees, and K is the number of classifications, set to 3. It is the prediction contribution of the m-th decision tree to the class k of sample x. It is the sum of the prediction results of M decision trees. It is the probability that sample x belongs to class k; The early warning module is used to provide graded early warnings based on the probability results and to make clinical decisions based on the graded early warnings.

6. An electronic device, characterized in that, include: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for early warning detection of severe digestive system diseases as described in any one of claims 1-4.

Images