Early grading decision-making method for acute pulmonary embolism deterioration risk

Abstract

The invention provides an early grading decision-making method for an acute pulmonary embolism deterioration risk. The method comprises the steps of obtaining a data set; the data set comprises clinical information, laboratory detection results and CT pulmonary artery imaging of the acute pulmonary embolism patient; three-dimensional structures of the pulmonary artery, the pulmonary vein, the left ventricle, the right ventricle, the left atrium and the right atrium in CT pulmonary artery imaging are segmented; constructing a three-dimensional pulmonary vein and pulmonary artery blood vessel tree model; obtaining difference filling data of contrast agent imaging; obtaining a pulmonary circulation reflux score; constructing a four-cavity center plane; calculating RV / LV; screening the preprocessed clinical information and laboratory detection results; fusing the screened characteristic data related to the exacerbation of the acute pulmonary embolism disease, the pulmonary circulation reflux score and the RV / LV; inputting the training set into the model for training, and inputting the verification set into the trained model for optimization; and inputting the test set into the final optimized model to obtain a prediction result of the grading diagnosis index of the acute pulmonary embolism patient.

Description

Technical Field

[0001] The present invention relates to the technical field of acute pulmonary embolism exacerbation risk assessment, and in particular to an early grading decision-making method for acute pulmonary embolism exacerbation risk. Background Art

[0002] Acute pulmonary embolism is the third leading cause of death among cardiovascular emergencies, with a 30-day mortality rate as high as 3.5%. Despite significant progress in technologies such as catheter-based treatment in recent years, the number of deaths from acute pulmonary embolism has not decreased. Clinical studies have shown that the core contradiction leading to this phenomenon is the lack of the ability to identify patients with severe deterioration early and accurately. Previous studies have mostly used death as the endpoint, which has little practical guidance for treatment. Therefore, we should pay more attention to the early identification of acute pulmonary embolism on the verge of deterioration, that is, patients who have just shown signs of deterioration rather than those who have already died, and provide targeted treatment within the critical time window (such as thrombolysis, thrombectomy, etc.). Such operations will greatly help improve the prognosis of patients with pulmonary embolism.

[0003] Currently, the mortality risk assessment for acute pulmonary embolism mainly relies on clinical scores such as the PESI score and single biomarker testing (such as brain natriuretic peptide). However, these methods have significant limitations:

[0004] (1) The existing single imaging or blood biomarker has insufficient predictive power and is not recommended by the guidelines;

[0005] (2) Published models seriously underutilize modern methodological approaches to imaging data, which limits the predictive value of the models;

[0006] (3) Previous models mostly used death as the endpoint for prediction, which failed to identify the group of patients who were on the verge of deterioration and before death, seriously affecting the implementation of early treatment. Summary of the Invention

[0007] The present invention aims to solve the core technical problems of the existing technology in the risk assessment of acute pulmonary embolism exacerbation, such as insufficient representation of national conditions, one-sided endpoint event assessment, insufficient accuracy, single modality, and separate use of imaging and vital signs data.

[0008] To this end, the purpose of the present invention is to propose an early grading decision-making method for the risk of exacerbation of acute pulmonary embolism.

[0009] In order to achieve the above-mentioned purpose, the technical solution of the present invention provides an early graded decision-making method for the risk of exacerbation of acute pulmonary embolism. The early graded decision-making method for the risk of exacerbation of acute pulmonary embolism includes: step S1: acquiring a data set; wherein, the data set is clinical information, laboratory test results, and CT pulmonary artery imaging of patients with acute pulmonary embolism; the data set is marked with a graded diagnostic index of acute pulmonary embolism disease; the graded diagnostic index corresponds to the severity of early deterioration of the disease in patients with acute pulmonary embolism; and different graded diagnostic indices correspond to different medical treatments for acute pulmonary embolism disease; step S2: uniformly converting the CT pulmonary artery imaging into a medical imaging standard format; step S3: preprocessing the converted CT pulmonary artery imaging; step S4: segmenting the preprocessed CT pulmonary artery imaging to divide Cut out the three-dimensional structure of the pulmonary artery, pulmonary vein, left ventricle, right ventricle, left atrium and right atrium in the CT pulmonary artery imaging; step S5: extract the pulmonary vein and pulmonary artery trunk and its branch vessels based on the segmentation result of step S4, and perform pixel measurement on the extracted pulmonary vein and pulmonary artery trunk and its branch vessels to construct a three-dimensional pulmonary vein and pulmonary artery vascular tree model; step S6: based on the pixel measurement values ​​of the constructed three-dimensional pulmonary vein vascular tree model and the pixel measurement values ​​of the left atrium, obtain the differential filling data of the contrast agent imaging of the pulmonary vein and the left atrium in the three-dimensional pulmonary vein vascular tree model; step S7: based on the differential filling data, obtain the pulmonary circulation reflow score; wherein, the pulmonary circulation reflow score is based on The number of abnormal pulmonary veins in the four pulmonary veins is defined; the four pulmonary veins are: right superior pulmonary vein, right inferior pulmonary vein, left superior pulmonary vein, and left inferior pulmonary vein; step S8: automatic construction of the four-chamber heart plane; step S9: sequentially extracting pixels of the heart chambers and automatically segmenting the heart edges, and then calculating the spatial vertical diameter based on the ventricular septum; step S10: based on the right ventricular diameter and the left ventricular diameter in the spatial vertical diameter, automatically calculating the ratio of the right ventricular diameter to the left ventricular diameter; step S11: preprocessing the clinical information and laboratory test results of the patient with acute pulmonary embolism; step S12: performing random forest algorithm on the preprocessed clinical information and laboratory test results The method is used to screen feature data to screen out feature data related to the deterioration of acute pulmonary embolism; step S13: fusing the screened feature data related to the deterioration of acute pulmonary embolism, the pulmonary circulation reflow score, and the ratio of the right ventricular diameter to the left ventricular diameter, and forming the fused features into a set of predictive variables; step S14: dividing the set of predictive variables into a training set, a validation set, and a test set according to a preset ratio; step S15: building a prediction model for the deterioration of acute pulmonary embolism; wherein the prediction model for the deterioration of acute pulmonary embolism is a logistic regression machine learning model; the prediction model for the deterioration of acute pulmonary embolism uses a logistic regression machine learning algorithm to extract features from the set of predictive variables;Step S16: Input the training set into the prediction model for training, input the validation set into the trained prediction model for optimization, and save the final optimized prediction model; Step S17: Input the test set into the final optimized prediction model to obtain the prediction result of the graded diagnostic index of patients with acute pulmonary embolism.

[0010] Preferably, the clinical information includes one of the following or a combination thereof: age, gender, heart rate, blood pressure, respiratory rate, and blood oxygen saturation; the laboratory test results include one of the following or a combination thereof: troponin I value, brain natriuretic peptide value, and amino-terminal pro-brain natriuretic peptide value.

[0011] Preferably, the medical image standard format is DICOM format.

[0012] Preferably, the pulmonary circulation reflow score is defined based on the number of abnormal pulmonary veins among the four pulmonary veins, specifically: when none of the four pulmonary veins is abnormal, the pulmonary circulation reflow score is defined as level 0; when a single pulmonary vein is affected among the four pulmonary veins, the pulmonary circulation reflow score is defined as level 1; when two of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as level 2; when three or more of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as level 3.

[0013] Preferably, the time period corresponding to the early deterioration of the disease in the patient with acute pulmonary embolism is: 30 days after the patient with acute pulmonary embolism is admitted to the hospital or within 30 days after the admission.

[0014] Beneficial effects of the present invention:

[0015] (1) Prediction accuracy enhanced by synergistic imaging and physiological features: The innovative development of pulmonary circulation reflow score and feature correlation can explore the correlation between features and pulmonary embolism deterioration, which can be used as prior knowledge to enhance the prediction accuracy of the prediction model for acute pulmonary embolism deterioration.

[0016] (2) Task adaptability of multimodal feature fusion: The multimodal feature fusion of the present invention mines fusion features specific to the task of predicting the exacerbation of acute pulmonary embolism. The acute pulmonary embolism exacerbation prediction model of the present invention designs a prediction model based on multimodal features, which can more effectively utilize multimodal information.

[0017] (3) Data robustness and clinical practicality: When training the prediction model for acute pulmonary embolism exacerbation, multimodal diagnostic and treatment information is fully integrated; when applied to early prediction, only structured data is required as input for the prediction model for acute pulmonary embolism exacerbation. Therefore, to a certain extent, the need for multimodal data integrity in the prediction model for acute pulmonary embolism exacerbation in actual applications is reduced, and the problem of being unable to achieve accurate and reliable prediction when some examination data is missing is solved.

[0018] (4) Possessing core value of clinical transformation: Multicenter clinical trials have shown that compared with the traditional intelligent prediction method for the risk of exacerbation of acute pulmonary embolism, the early grading decision-making method for the risk of exacerbation of acute pulmonary embolism of the present invention can reduce the exacerbation-related mortality rate of acute pulmonary embolism patients within 30 days or within 30 days. Through precise risk stratification, overtreatment of low-risk patients is avoided, the allocation of medical resources is optimized, and the missed diagnosis of high-risk patients is reduced, thereby improving the overall prognosis of pulmonary embolism. In addition, it supports the closed-loop management from emergency department diagnosis to the realization of "scanning is warning - monitoring is intervention".

[0019] Additional aspects and advantages of the invention will become apparent from the description which follows, or may be learned by practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic flow chart showing an early stratification decision-making method for the risk of exacerbation of acute pulmonary embolism according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, Figure 1 As shown, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0023] Figure 1 FIG2 is a schematic flow chart showing an early classification decision-making method for the risk of exacerbation of acute pulmonary embolism according to an embodiment of the present invention. Figure 1 As shown in the figure, the early stratification decision-making method for the risk of worsening acute pulmonary embolism includes:

[0024] Step S1: Acquire a data set; wherein the data set comprises clinical information, laboratory test results, and CT pulmonary artery imaging of patients with acute pulmonary embolism; the data set is annotated with a graded diagnostic index for acute pulmonary embolism; the graded diagnostic index corresponds to the severity of early exacerbation of the disease in patients with acute pulmonary embolism; and different graded diagnostic indices correspond to different medical treatments for acute pulmonary embolism;

[0025] Step S2: uniformly converting the CT pulmonary artery images into a medical imaging standard format;

[0026] Step S3: preprocessing the converted CT pulmonary artery imaging;

[0027] Step S4: segmenting the pre-processed CT pulmonary artery image to segment the three-dimensional structures of the pulmonary artery, pulmonary vein, left ventricle, right ventricle, left atrium, and right atrium in the CT pulmonary artery image;

[0028] Step S5: Constructing a 3D pulmonary vein and artery vascular tree model; specifically, extracting the pulmonary vein and artery trunks and their branches based on the segmentation results of step S4, and performing pixel measurement on the extracted pulmonary vein and artery trunks and their branches to construct a 3D pulmonary vein and artery vascular tree model;

[0029] Step S6: obtaining differential filling data of the pulmonary veins (pulmonary veins in the 3D pulmonary vein tree model) and the left atrium based on the pixel image measurement values ​​of the 3D pulmonary vein tree model and the left atrium;

[0030] Step S7: Obtaining a pulmonary recirculation score based on the differential filling data; wherein the pulmonary recirculation score is defined based on the number of abnormal pulmonary veins in the four pulmonary veins; the four pulmonary veins are: right superior pulmonary vein, right inferior pulmonary vein, left superior pulmonary vein, and left inferior pulmonary vein;

[0031] Step S8: Automated construction of the four-chamber heart plane;

[0032] Step S9: sequentially extracting pixels of the cardiac cavity and automatically segmenting the cardiac edge, and then calculating the spatial vertical diameter based on the ventricular septum;

[0033] Step S10: automatically calculating the ratio of the right ventricular diameter to the left ventricular diameter based on the right ventricular diameter and the left ventricular diameter in the vertical axis of space; wherein the ratio of the right ventricular diameter to the left ventricular diameter is RV / LV;

[0034] Step S11: pre-processing clinical information and laboratory test results of patients with acute pulmonary embolism;

[0035] Step S12: screening characteristic data of the pre-processed clinical information and laboratory test results using a random forest algorithm to screen out characteristic data related to the exacerbation of acute pulmonary embolism;

[0036] Step S13: fusing the characteristic data (characteristic data related to the exacerbation of acute pulmonary embolism), the pulmonary circulation reflow score, and the ratio of the right ventricular diameter to the left ventricular diameter, and forming the fused characteristics into a set of predictive variables;

[0037] Step S14: Divide the set of prediction variables into a training set, a validation set, and a test set according to a preset ratio;

[0038] Step S15: building a prediction model for the worsening of acute pulmonary embolism; wherein the prediction model for the worsening of acute pulmonary embolism is a logistic regression machine learning model; the prediction model for the worsening of acute pulmonary embolism uses a logistic regression machine learning algorithm to perform feature extraction on the set of prediction variables;

[0039] Step S16: input the training set into the prediction model for training, input the validation set into the trained prediction model for optimization, and save the final optimized prediction model;

[0040] Step S17: Input the test set into the final optimized prediction model to obtain the prediction results of the graded diagnostic index of patients with acute pulmonary embolism.

[0041] In this embodiment, the method for early grading and decision-making of the risk of worsening of acute pulmonary embolism provided by the present invention is, more precisely, a method for early grading and decision-making of the risk of worsening of acute pulmonary embolism patients based on multimodal data fusion based on the newly developed pulmonary circulation reflow score. Specifically, by deeply integrating the anatomical imaging features of CT pulmonary artery imaging (CTPA for short) with vital signs data (i.e., clinical information and laboratory test results of patients with acute pulmonary embolism), an integrated "diagnosis-prediction" model is constructed, which enables patients to complete individualized worsening risk stratification and intervention time window prediction at the same time as pulmonary embolism is diagnosed, breaking through the static and lagging bottlenecks of traditional scoring systems, and solving the core technical problems of insufficient accuracy in the assessment of the risk of worsening of acute pulmonary embolism and the fragmented use of imaging and vital signs data in the existing technology.

[0042] In one embodiment of the present invention, the clinical information includes one of the following or a combination thereof: age, gender, heart rate, blood pressure, respiratory rate, and blood oxygen saturation; the laboratory test results include one of the following or a combination thereof: troponin I value, brain natriuretic peptide value, and amino-terminal pro-brain natriuretic peptide value.

[0043] In one embodiment of the present invention, the medical image standard format is the DICOM format.

[0044] In one embodiment of the present invention, the pulmonary circulation reflow score is defined based on the number of abnormal pulmonary veins among the four pulmonary veins, specifically: when none of the four pulmonary veins is abnormal, the pulmonary circulation reflow score is defined as level 0; when only one of the four pulmonary veins is affected, the pulmonary circulation reflow score is defined as level 1; when two of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as level 2; when three or more of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as level 3.

[0045] In one embodiment of the present invention, the time period corresponding to the early deterioration of the disease in the patient with acute pulmonary embolism is: 30 days after the patient with acute pulmonary embolism is admitted to the hospital or within 30 days after the admission.

[0046] The following is a specific embodiment to illustrate the technical solution of the present invention. The early classification decision-making method for the risk of worsening acute pulmonary embolism in this specific embodiment is implemented by the following steps:

[0047] (1) Step S1: Collect a large sample of pulmonary embolism patient data in the northeastern and northern regions of my country (i.e., obtain a data set). The large sample of pulmonary embolism patient data includes: patient clinical information, laboratory test results, and CT pulmonary artery imaging.

[0048] CT pulmonary artery angiography (CTPA), as the preferred diagnostic tool for acute pulmonary embolism, provides multidimensional quantitative data on cardiac chambers and pulmonary vasculature, and has become an indispensable tool for the assessment of cardiopulmonary diseases both domestically and internationally. Therefore, we collected CT pulmonary artery angiography data from patients with pulmonary embolism.

[0049] The clinical information includes one of the following or a combination thereof: age, gender, heart rate, blood pressure, respiratory rate, and blood oxygen saturation; the laboratory test results include one of the following or a combination thereof: troponin I value, brain natriuretic peptide value, and amino-terminal pro-brain natriuretic peptide value.

[0050] (2) Step S2: The original image data generated when these patients undergo CT pulmonary artery imaging examinations are uniformly converted into a medical imaging standard format (DICOM format) to ensure data compatibility and consistency of subsequent processing.

[0051] (3) Step S3: Use professional medical image processing software to preprocess the CTPA images in DICOM format.

[0052] (4) Step S4: Using the medical image segmentation technology based on the UNET model in the software, the three-dimensional structures of the pulmonary artery, pulmonary vein, left ventricle, right ventricle, left atrium and right atrium in the image are accurately identified and segmented.

[0053] (5) Step S5: Based on the segmentation results of step S4, the pulmonary veins and pulmonary arteries and their branches are extracted. In these selected pulmonary veins and pulmonary arteries and their branches, including the trunk, left and right pulmonary veins, and lobar pulmonary veins, precise pixel measurements are performed using the tools provided by the image processing software to construct a clear three-dimensional (3D) pulmonary vein and pulmonary artery vascular tree model.

[0054] (6) Step S6: Based on the pixel measurement values ​​of the constructed three-dimensional pulmonary venous tree model and the pixel measurement values ​​of the left atrium, differential filling data of the contrast agent imaging of the pulmonary veins and the left atrium in the three-dimensional pulmonary venous tree model are obtained.

[0055] (7) Step S7: Obtaining a pulmonary recirculation score based on the differential filling data; wherein the pulmonary recirculation score is defined based on the number of abnormal pulmonary veins in the four pulmonary veins; the four pulmonary veins are: right superior pulmonary vein, right inferior pulmonary vein, left superior pulmonary vein, and left inferior pulmonary vein;

[0056] Specifically, the pulmonary circulation reflow score is obtained through the length of the differential filling range and the specific value of the density difference in the differential filling data.

[0057] The pulmonary circulation reflow score is defined based on the number of abnormal pulmonary veins among the four pulmonary veins. Specifically, when none of the four pulmonary veins is abnormal, the pulmonary circulation reflow score is defined as level 0; when only one of the four pulmonary veins is affected, the pulmonary circulation reflow score is defined as level 1; when two of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as level 2; when three or more of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as level 3 (Note: the highest level is level 3, even if all four veins are affected, it is still classified as level 3).

[0058] At the junction where the four major pulmonary veins (superior right, right inferior, superior left, and inferior left pulmonary veins) join the left atrium (LA), pulmonary veins that meet the pulmonary circulation return score criteria are considered abnormal and must meet the following two strict criteria: 1. Inhomogeneous filling: the presence of abnormal low-density pixel areas filled with contrast agent, and the late quantification shows that the abnormal low-density pixel areas extend beyond the long axis of the vessel by more than 2 cm; 2. Significant difference with left atrial imaging as the control: using the left atrial pixel values ​​visible in the image as the comparison reference, the average CT attenuation value of the two exceeds 160 Hounsfield units (HU).

[0059] (8) Step S8: Automated construction of the four-chamber heart plane.

[0060] The four-chamber plane is constructed to automatically identify the apex and ventricular septum, followed by image torsion. The four-chamber plane is automatically constructed using the following steps: Volumetric data with a 0.5–1 mm slice thickness is acquired during a CT scan of the pulmonary arteries with intravenous iodinated contrast agent, covering the area from the tracheal carina to 1 cm below the apex. During reconstruction, the left ventricular apex and the center of the mitral valve are first located in the transverse plane. A sagittal oblique plane is generated along the long axis of the left ventricle (LV) and perpendicular to the interventricular septum between the right and left ventricles. The coronal oblique plane is then adjusted within this plane so that the reconstruction plane is parallel to the interventricular septum and passes through the LV and right ventricular apexes and the centers of the mitral and tricuspid valves, ultimately creating a standard four-chamber plane. This requires the following: ① clear visualization of the left ventricular apex; ② visualization of the interventricular and atrial septal structures as continuous linear structures; and ③ clear visualization of the bilateral atrioventricular valves (the anterior leaflet of the mitral valve and the septal leaflet of the tricuspid valve), with discernible valve opening and closing.

[0061] (9) Step S9: Pixel extraction of the cardiac cavity and automated segmentation of the cardiac edge are performed in sequence, and then the spatial vertical diameter based on the ventricular septum is calculated.

[0062] Specifically, pixel extraction of the cardiac chambers and automatic segmentation of the cardiac edges were performed successively. The following is a detailed description of the refined identification and measurement method of the four cardiac chambers after the CT four-chamber cardiac plane was constructed: On the standard four-chamber cardiac plane (layer thickness 1.0 mm, iterative reconstruction), first positioning was performed through anatomical landmarks: Left ventricle (LV): The boundary is bounded by the right edge of the ventricular septum and the free wall endocardium, and the anterior / posterior papillary muscles can be seen protruding into the chamber during systole; Right ventricle (RV): The triangular cardiac chamber is attached to the anterior and lateral sides of the left ventricle; Left atrium (LA): Located above and behind the mitral valve, with the left atrial appendage opening as the lateral landmark; Right atrium (RA): Located above and behind the tricuspid valve, with the terminal ridge separating the anterior trabecular muscle area from the posterior smooth area. Measurement standards for relevant indicators: Cardiac chamber dimensions: LV diameter: the longest diameter from the endocardial surface of the ventricular septum to the endocardial surface of the lateral wall, perpendicular to the long axis (base-apex line); RV diameter: the longest diameter from the right ventricular surface of the ventricular septum to the endocardial surface of the free wall, parallel to the tricuspid annulus (1 / 3 of the base); LA diameter: the longest diameter from the midpoint of the line connecting the posterior edge of the mitral annulus to the posterior superior wall endocardium, perpendicular to the atrial septum; RA diameter: the longest diameter from the midpoint of the tricuspid annulus to the posterior wall endocardium, avoiding the terminal ridge and the opening of the inferior vena cava.

[0063] (10) The right ventricular diameter is then automatically divided by the left ventricular diameter to calculate the right ventricular diameter / left ventricular diameter, that is, the ratio of the right ventricular diameter to the left ventricular diameter.

[0064] (11) Step S11: Preprocess the clinical information and laboratory test results of patients with acute pulmonary embolism. If the obtained clinical information and laboratory test results of the patient have missing data under a certain modality, mark them as missing.

[0065] (12) Step S12: The pre-processed clinical information and laboratory test results are subjected to a random forest algorithm to screen characteristic data to screen out characteristic data related to the exacerbation of acute pulmonary embolism.

[0066] (13) Step S13: Fusing the filtered feature data related to the exacerbation of acute pulmonary embolism, the pulmonary circulation reflow score, and the ratio of the right ventricular diameter to the left ventricular diameter, and forming the fused features into a set of predictive variables.

[0067] (14) Step S14: Divide the set of prediction variables into a training set, a validation set, and a test set according to a preset ratio.

[0068] (15) Step S15: Build a prediction model for the worsening of acute pulmonary embolism; wherein, the prediction model for the worsening of acute pulmonary embolism is a logistic regression machine learning model; the prediction model for the worsening of acute pulmonary embolism uses a logistic regression machine learning algorithm to extract features from the set of prediction variables.

[0069] (16) Step S16: Input the training set into the prediction model for training, input the validation set into the trained prediction model for optimization, and save the final optimized prediction model.

[0070] The consistency of the model across different patient groups was assessed using calibration curves, and the predictive ability of the acute pulmonary embolism exacerbation prediction model was evaluated using area under the curve, sensitivity, specificity, and decision analysis curves. The prediction tool constructed in the training set was validated in internal and external validation sets. The consistency of the model across different patient groups was assessed using calibration curves, and the predictive performance of the acute pulmonary embolism exacerbation prediction model was evaluated using area under the curve, sensitivity, specificity, and decision analysis curves.

[0071] (17) Step S17: Input the test set into the final optimized prediction model to obtain the prediction results of the graded diagnostic index of patients with acute pulmonary embolism.

[0072] The characteristic variables in the test set were input into a prediction model for acute pulmonary embolism exacerbation. Using a classification and regression tree algorithm, a graded diagnostic index (GDI) for acute pulmonary embolism patients was obtained. The GDI corresponds to the severity of early exacerbation of acute pulmonary embolism, and different GDIs correspond to different medical treatments for acute pulmonary embolism.

[0073] The evolution and core goal of acute pulmonary embolism prognostic assessment is to follow patients' disease outcomes within 30 days. Traditional prediction models (such as PESI and sPESI) primarily focus on the endpoint of death. However, evidence-based medicine reveals that non-high-risk acute pulmonary embolism patients (hemodynamically stable patients) generally experience a phase of interventionable disease progression before death. To overcome the bottleneck in prognostic assessment, the applicant proposes the use of early deterioration events as a key surrogate endpoint, defined as deterioration occurring within 30 days of hospital admission.

[0074] The prediction of the graded diagnostic index of patients with acute pulmonary embolism is to grade the probability of occurrence of deterioration events. This specific embodiment can divide deterioration events into two result judgment modes: binary classification or three-category classification. Grading mode 1: The binary classification is to divide all patients into a high deterioration risk group with a deterioration probability of 64% and a low deterioration risk group with a deterioration risk of 3%. The high-risk group is recommended to be transferred to the intensive care unit and arrange for rescue treatment such as thrombolysis and thrombectomy as soon as possible; the low-risk group is recommended to receive oral medication and be discharged as soon as possible. This mode is suitable for graded decision-making for basically all patients seeking medical treatment. Grading mode 2: The three-category classification is to divide all patients into a high deterioration risk group with a deterioration probability of 87%, a medium deterioration risk group with a deterioration risk of 44%, and a low deterioration risk group with a deterioration risk of 3%. The high-risk group is recommended to be transferred to the intensive care unit and arrange for rescue treatment such as thrombolysis and thrombectomy as soon as possible; the medium-risk group is recommended to be kept in the general ward for vital sign monitoring and close observation. If the condition worsens, they should be transferred to the intensive care unit in time and arrange for rescue treatment such as thrombolysis and thrombectomy as soon as possible. If they continue to improve, they can be considered for discharge; the low-risk group is recommended to take oral medication and be discharged as soon as possible. This model is suitable for doctors' hierarchical decision-making in hospitals with a distinction between intensive care units and general wards.

[0075] Specifically, the prediction model for the worsening of acute pulmonary embolism includes three important indicators: pulmonary circulation reflow score ≥1, right ventricular / left ventricular diameter ratio ≥1.2, and heart rate ≥110 beats / min. The output mode of the prediction model for the worsening of acute pulmonary embolism is as follows: Grading mode 1 - two-category grading: If any two or all three of the three indicators are included, the patient is in the high-risk group for worsening; if one indicator is included or none of them are included, the patient is in the low-risk group for worsening. Among them, the high-risk group is recommended to be transferred to the intensive care unit and arrange for rescue treatment such as thrombolysis and thrombectomy as soon as possible; the low-risk group is recommended to take oral medication and be discharged as soon as possible. Grading mode 2 - three-category grading: If all three indicators are included, the patient is in the high-risk group for worsening; if any two of the three indicators are included, the patient is in the moderate-risk group for worsening; if one indicator is included or none of them are included, the patient is in the low-risk group. The high-risk group is recommended to be transferred to the ICU and receive emergency treatments such as thrombolysis and thrombectomy as soon as possible. The medium-risk group is recommended to be kept in a general ward for vital sign monitoring and close observation. If the condition worsens, they should be promptly transferred to the ICU and receive emergency treatments such as thrombolysis and thrombectomy as soon as possible. If the condition continues to improve, discharge can be considered. This will assist different doctors in making tiered decisions during diagnosis and treatment.

[0076] In this specific embodiment, the present invention is based on a prospective pulmonary embolism cohort totaling thousands of people constructed by multiple hospitals. The present invention has developed a pulmonary circulation reflow score for the first time. This index is innovative volumetric data developed based on CTPA reconstruction technology. It has the spatial efficiency of quantifying pulmonary vascular and cardiac function, and has prospectively verified its significant potential for early prediction of pulmonary embolism deterioration in multi-center data. Therefore, based on the core index of the pulmonary circulation reflow score, the present invention also brings together data such as the ratio of right ventricular diameter to left ventricular diameter (RV / LV), clinical information of patients with acute pulmonary embolism, and laboratory test results to construct multimodal big data. It has also been constructed, tested and verified in a large sample of prospective patients from multiple centers in my country, and developed an integrated tool for "multimodal real-time diagnosis and prediction of imaging-vital signs fusion", which enables patients to complete individualized deterioration risk stratification and intervention time window prediction at the same time as pulmonary embolism is diagnosed. The prediction model for the worsening of acute pulmonary embolism is developed based on CTPA and combined with the advancement of pulmonary embolism treatment endpoints to predict early worsening rather than death; the grading method, which has been trained and verified with large samples, is expected to become an important tool for identifying and grading the worsening of pulmonary embolism, providing a scientific basis and technical support for improving patient prognosis; this prediction model for the worsening of acute pulmonary embolism combines unique advantages such as innovation, applicability in my country, real-time, accuracy, and advancement of treatment, while taking into account economic value, breaking through the static and lagging bottlenecks of traditional scoring systems.

[0077] In summary, the main technical features and uses of the present invention are as follows: The present invention provides a deterioration risk assessment technology specifically for patients with acute pulmonary embolism, and its core technological innovation lies in the construction of a risk prediction model based on the fusion of multimodal medical parameters based on the newly developed pulmonary circulation reflow score. Traditional prediction models mainly focus on the endpoint event of death to advance and expand clinical intervention treatment events in patients with acute pulmonary embolism, and identify the stage of deterioration of the disease that is generally experienced before death as a key alternative endpoint to guide early treatment and improve disease prognosis. By integrating the clinical indicator heart rate with the imaging features of CT pulmonary angiography, a quantitative scoring system based on multivariate logistic regression is established, which solves the pain point of the existing technology's insufficient sensitivity in identifying the deterioration risk of patients in the compensated blood pressure period. Specific technical features include: 1) Based on the newly developed pulmonary circulation reflow score, the CTPA imaging parameters and real-time heart rate indicators are jointly analyzed for the construction of a highly discriminatory binary and ternary risk determination rule; 2) Through the verification of a prospective multi-center cohort, an early warning of pulmonary embolism-related deterioration within 30 days is achieved (AUC reaches 0.80-0.82 and 0.84-0.89); 3) Based on the simple decision threshold of "meeting 2-3 of the 3 indicators to determine the high-risk patient group for deterioration", the complexity of clinical operations is significantly reduced. This technology can be embedded in the medical information system to assist emergency and respiratory physicians in quickly identifying hidden high-risk patients who need intensive monitoring or intervention, avoid misdiagnosis of the disease, and shorten the clinical decision-making time to 4-6 hours. It has important value in optimizing the hierarchical diagnosis and treatment system for acute PE.

[0078] Furthermore, the implementation prospects and preliminary ideas of this invention are as follows: The implementation prospects of this invention focus on reshaping the clinical decision-making system for patients with acute pulmonary embolism. Its core value lies in filling the technical gap of early warning of the risk of deterioration of pulmonary embolism patients based on the newly developed pulmonary circulation reflow score and through the multimodal data fusion algorithm. Based on the clinical status of new pulmonary embolism patients in my country each year, this technology integrates CTPA imaging features with real-time heart rate data to construct a high-discrimination (AUC 0.80-0.82) dynamic early warning model, which can accurately identify high-risk groups with a doubling of the risk of deterioration within 30 days.

[0079] Furthermore, the implementation path of this invention is divided into three phases: 1) The first phase involves completing clinical validation of the algorithm and developing an independent module that complies with medical device software specifications. This module will be connected to the hospital's PACS system and electronic medical records, enabling automatic association of CTPA imaging reports with vital sign data and generating real-time risk stratification. 2) The second phase involves the simultaneous development of lightweight mobile tools (such as WeChat mini-programs and web-based tools) to implement a closed-loop management system of "5-minute risk assessment and stratified treatment." This technology is expected to improve the early identification rate of pulmonary embolism patients, reduce the use of intensive care resources, and reduce in-hospital pulmonary embolism-related mortality, in line with the national medical quality and safety improvement goals and the national requirements for the construction of a tiered diagnosis and treatment system for critical and emergency diseases.

[0080] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An early stratification decision-making method for the risk of exacerbation of acute pulmonary embolism, characterized by: include: Step S1: Acquire a data set; wherein the data set comprises clinical information, laboratory test results, and CT pulmonary artery imaging of patients with acute pulmonary embolism; the data set is annotated with a graded diagnostic index for acute pulmonary embolism; the graded diagnostic index corresponds to the severity of early exacerbation of the disease in patients with acute pulmonary embolism; and different graded diagnostic indices correspond to different medical treatments for acute pulmonary embolism; Step S2: uniformly converting the CT pulmonary artery images into a medical imaging standard format; Step S3: preprocessing the converted CT pulmonary artery imaging; Step S4: segmenting the pre-processed CT pulmonary artery imaging to segment the three-dimensional structures of the pulmonary artery, pulmonary vein, left ventricle, right ventricle, left atrium and right atrium in the CT pulmonary artery imaging; Step S5: extracting the pulmonary veins and pulmonary arteries and their branches based on the segmentation results of step S4, and performing pixel measurement on the extracted pulmonary veins and pulmonary arteries and their branches to construct a three-dimensional pulmonary vein and pulmonary artery vascular tree model; Step S6: Based on the pixel measurement values ​​of the constructed three-dimensional pulmonary venous tree model and the pixel measurement values ​​of the left atrium, differential filling data of the contrast agent imaging of the pulmonary veins and the left atrium in the three-dimensional pulmonary venous tree model are obtained; Step S7: Based on the differential filling data, a pulmonary circulation reflow score is obtained; wherein the pulmonary circulation reflow score is defined based on the number of abnormal pulmonary veins in the four pulmonary veins; the four pulmonary veins are: right superior pulmonary vein, right inferior pulmonary vein, left superior pulmonary vein, and left inferior pulmonary vein; Step S8: Automated construction of the four-chamber heart plane; Step S9: sequentially extracting pixels of the cardiac cavity and automatically segmenting the cardiac edge, and then calculating the spatial vertical diameter based on the ventricular septum; Step S10: automatically calculating the ratio of the right ventricular diameter to the left ventricular diameter based on the right ventricular diameter and the left ventricular diameter in the vertical axis of the space; Step S11: pre-processing the clinical information and laboratory test results of the patient with acute pulmonary embolism; Step S12: screening characteristic data of the pre-processed clinical information and laboratory test results using a random forest algorithm to screen out characteristic data related to the exacerbation of acute pulmonary embolism; Step S13: fusing the filtered feature data related to the exacerbation of acute pulmonary embolism, the pulmonary circulation reflow score, and the ratio of the right ventricular diameter to the left ventricular diameter, and forming the fused features into a set of predictive variables; Step S14: dividing the set of prediction variables into a training set, a validation set, and a test set according to a preset ratio; Step S15: building a prediction model for the worsening of acute pulmonary embolism; wherein the prediction model for the worsening of acute pulmonary embolism is a logistic regression machine learning model; the prediction model for the worsening of acute pulmonary embolism uses a logistic regression machine learning algorithm to perform feature extraction on the set of prediction variables; Step S16: inputting the training set into the prediction model for training, inputting the validation set into the trained prediction model for optimization, and saving the final optimized prediction model; Step S17: inputting the test set into the finally optimized prediction model to obtain the prediction results of the graded diagnostic index of patients with acute pulmonary embolism.

2. The early stratification decision-making method for the risk of exacerbation of acute pulmonary embolism according to claim 1, characterized in that: The clinical information includes one or a combination of the following: age, gender, heart rate, blood pressure, respiratory rate, and blood oxygen saturation; The laboratory test results include one or a combination of the following: troponin I value, brain natriuretic peptide value, and amino-terminal pro-brain natriuretic peptide value.

3. The early stratification decision-making method for the risk of exacerbation of acute pulmonary embolism according to claim 1, characterized in that: The medical image standard format is the DICOM format.

4. The early stratification decision-making method for the risk of exacerbation of acute pulmonary embolism according to claim 1, characterized in that: The pulmonary reflow score is defined based on the number of anomalous pulmonary veins among the four pulmonary veins, specifically: When none of the four pulmonary veins has abnormalities, the pulmonary circulation reflow score is defined as grade 0; When only one of the four pulmonary veins is affected, the pulmonary circulation reflow score is defined as grade 1; When two of the four pulmonary veins were affected, the pulmonary circulation reflow score was defined as grade 2; When three or more of the four pulmonary veins are affected, the pulmonary circulation reflow score is defined as grade 3.

5. The early stratification decision-making method for the risk of worsening of acute pulmonary embolism according to any one of claims 1 to 4, characterized in that: The time period corresponding to the early deterioration of the disease in patients with acute pulmonary embolism is: 30 days after the patient with acute pulmonary embolism is admitted to the hospital or within 30 days after the admission.

Images