Prediction model and construction method for major adverse cardiovascular events based on thoracic artery calcification

By calcifying the calcification integral of common chest artery and using imagingomics technology, combining imagingomics characteristics and clinical data, a composite prediction model was constructed, which solved the limitations of coronary calcification integral in MACEs prediction in the existing technology, and improved the prediction Accuracy and specificity.

CN113476068BActive Publication Date: 2024-11-01EIGHTH AFFILIATED HOSPITAL SUN YAT SEN UNIV (SHENZHEN FUTIAN)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110847786.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-07-27
Publication Date
2024-11-01
Estimated Expiration
2041-07-27

AI Technical Summary

Technical Problem

The existing coronary calcification scores have limitations in predicting major adverse cardiovascular events (MACEs), especially in patients with coronary calcification scores of 0 and chest artery calcification, resulting in insufficient prediction accuracy.

Method used

Using common chest artery calcification integral (CTACS) and imagingomics technology, CTACS is calculated through segmented CT imaging data, and combined imagingomics characteristics and clinical data, a composite prediction model is constructed to improve the predictive ability of MACEs.

Benefits of technology

CTACS predictive performance in MACEs is better than traditional coronary calcification integrals, and imaging omics integrals further improve the prediction effect. The constructed imaging omics-clinical variable model has good MACEs prediction ability, which improves the specificity and sensitivity of prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113476068B_ABST
    Figure CN113476068B_ABST
Patent Text Reader

Abstract

The present invention discloses a prediction model and construction method for major adverse cardiovascular events based on chest artery calcification, which adopts computer high-throughput imaging features, enriches the description of calcification features, and thus improves the accuracy of calcification indicators in predicting MACEs. The present invention extracts the imaging features of chest artery calcification based on CTACS using an imaging genomics analysis method, and constructs a new parameter for predicting MACEs, namely the imaging genomics score; the parameter is significantly better than traditional calcification assessment parameters such as CTACS and CACS in predicting MACEs. At the same time, an imaging genomics-clinical variable prediction model is constructed based on the imaging genomics score. The model has good prediction performance and can accurately predict whether a patient will have MACEs in the future, assisting physicians in individualized cardiovascular prevention and treatment, timely adjusting treatment plans, avoiding undertreatment or overtreatment, improving patient prognosis and improving quality of life, and has high clinical application value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of biomedicine, and specifically relates to a prediction model for major adverse cardiovascular events based on thoracic artery calcification and a construction method thereof. Technical Background

[0002] In the past 30 years, cardiovascular disease has been the leading cause of death in the world. Major adverse cardiovascular events (MACEs) are closely related to cardiovascular disease mortality and prognosis, and the occurrence of MACEs increases the cardiovascular burden. Therefore, establishing an accurate MACEs prediction model is one of the important measures to reduce the cardiovascular burden.

[0003] At present, clinical relevant MACEs prediction models are established for traditional cardiovascular risk factors; however, with the deepening of research, some scholars have found that even if these risk factors are stably controlled, some patients still have MACEs, which is called residual cardiovascular risk. In the past decade, studies have found that coronary artery calcification is not only a risk factor for cardiovascular disease, but also a predictor of MACEs, and is closely related to residual cardiovascular risk; in addition, some studies have shown that coronary artery calcification is more predictive of cardiovascular disease / events than traditional risk factors. In the Multi-Ethnic Study of Atherosclerosis (MESA), et al. constructed a ten-year cardiovascular risk prediction model by combining the coronary artery calcification score (CACS). The United States and China included coronary artery calcification as a risk indicator for primary prevention in 2019 and 2020, respectively. Therefore, the establishment of a MACEs prediction model combined with calcification indicators is expected to improve the accuracy of prediction.

[0004] However, some scholars have found from studies such as MESA that about one-third of MACEs occur in patients with a coronary artery calcification score of 0. On the other hand, although some scholars have pointed out that coronary artery calcification can be used to guide patients who need secondary / tertiary cardiovascular prevention, it is limited by the intervention of complex factors. However, while coronary artery calcification is being counted, other chest artery calcifications can also be obtained. Studies have shown that chest artery calcification is closely related to MACEs, and the KDIGO guidelines recommend thoracic aortic calcification as an important risk indicator for cardiovascular prognosis in patients with chronic kidney disease. However, the clinical predictive value of chest artery calcification for MACEs remains unclear. In view of the above, we propose whether a comprehensive analysis of chest artery calcification can improve the performance of calcification indicators in predicting MACEs. To this end, the chest total artery calcification score (CTACS) parameter, which is the sum of the calcification scores of each chest artery including coronary artery calcification, will be constructed to explore its predictive value for MACEs.

[0005] Agatston score is the gold standard method for evaluating calcification. This method defines lesions with CT values ​​≥130 as calcification. However, considering that 18F-NaF PET / CT can detect early calcification information and that calcification lesions are essentially heterogeneous lesions, Agatston score evaluation of calcification will inevitably lose some calcification information. The lost part includes undetected microcalcification lesions, which are associated with the occurrence of MACEs. Therefore, improving the description of calcification can help improve the prediction of MACEs. In this case, the imaging omics technology proposed in 2012 was used. Imaging omics technology mines rich imaging features from images and uses them for clinical diagnosis, evaluation and prognosis. In addition, imaging omics in coronary artery and tumor research has also shown that imaging omics features can extract rich calcification imaging information.

[0006] In summary, the establishment of a MACEs prediction model based on calcification scores, through full analysis of chest CT arterial calcification and the use of imaging genomics technology, creates more informative calcification imaging parameters, thereby improving the prediction performance of MACEs, which has important clinical guiding significance. Summary of the invention

[0007] In view of the defects of the prior art, the present invention aims to provide a prediction model and construction method for major adverse cardiovascular events based on chest artery calcification. The present invention proves that CTACS is superior to CACS in predicting MACEs. Secondly, imaging genomics analysis based on CTACS can extract more abundant calcification information for MACEs prediction, and finally establishes a composite prediction model for MACEs.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A prediction model for major adverse cardiovascular events based on thoracic artery calcification and a construction method thereof, the method comprising the following steps:

[0010] S1 collects clinical data of patients who have undergone chest plain scan CT and CT mediastinal window sequence image data, wherein the image data is DICOM images;

[0011] S2 divides the primary and secondary arteries of the CT image data of step S1 into segments, a total of 10 arteries; and uses Agatston integration to calculate the calcification scores of the 10 arteries; the thoracic common artery calcification score (CTACS) is the sum of the calcification scores of the 10 arteries;

[0012] S3 compared the performance of CTACS and CACS in predicting MACEs by using the AUC value of MACEs prediction;

[0013] S4 segmented the 10 arteries in step S2, extracted radiomic features, screened radiomic features related to the occurrence of MACEs through LASSO regression analysis, and obtained the corresponding radiomic score calculation formula;

[0014] S5 The AUC value of MACEs prediction was used to compare the performance of radiomics score with CTACS, CACS and annual growth rate of calcification (AGRC) in predicting MACEs.

[0015] S6 screened the radiomics scores and clinical data collected in step S1 by Spearman correlation analysis, univariate and multivariate Cox regression analysis to obtain parameters associated with the occurrence of MACEs;

[0016] S7 constructs a composite prediction model of radiomics-clinical variables based on the features obtained in step S6, and performs model evaluation.

[0017] It should be noted that the inclusion criteria for patients who have undergone chest CT scans in step S1 include: patients who have undergone at least 2 chest CT scans with an interval of more than 180 days, and the collected mediastinal window sequence image specifications are 3 mm layer thickness and 512*512 pixel matrix; the image analysis software includes I TK-SNAP and python 3.6.5 platforms.

[0018] It should be noted that the clinical data described in step S1 include the following clinical characteristics: age, gender, whether MACEs have occurred, whether there are underlying cardiovascular diseases, whether there are underlying cerebrovascular diseases, whether there are chronic kidney diseases, whether there are diabetes, smoking status (yes / no), drinking status (yes / no), systolic blood pressure / diastolic blood pressure, hypertension grade, urea nitrogen, serum creatinine, lipoprotein a, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, total cholesterol, blood glucose and blood calcium.

[0019] It should be noted that the definition of MACEs includes sudden cardiac death, acute heart failure, acute myocardial infarction, unstable angina requiring hospitalization, malignant arrhythmia requiring hospitalization, and stroke.

[0020] It should be noted that the 10 arteries described in step S2 include the aortic arch, ascending aorta, descending aorta, left subclavian artery, left common carotid artery, brachiocephalic trunk, right subclavian artery, right common carotid artery, coronary artery and pulmonary artery trunk; the CTACS=Σ(AS 冠状动脉 +AS 升主动脉 +AS 主动脉弓 +AS 降主动脉 +AS 肺动脉干 +AS 头臂干 +AS 右颈总动脉 +AS 右锁骨下动脉 +AS 左颈总动脉 +AS 左锁骨下动脉 ).

[0021] It should be noted that the AUC values ​​of step S3 and step S5 were compared using the Delong test, and a p value less than 0.05 was considered to be statistically different.

[0022] It should be noted that the method for extracting radiomic features from the CT mediastinal window in step S4 is: segmenting the 10 arteries layer by layer, stacking the layers to form a three-dimensional arterial segmentation body, and extracting radiomic features from the segmentation body.

[0023] It should be noted that the radiomics score in step S4 = (0.106)*(original_shape_Elongation)+

[0024] (0.00412)*(original_shape_Maximum2DDiameterSlice)+

[0025] (-116)*(wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis)+

[0026] (-67.5)*(wavelet-LLH_gldm_LowGrayLevelEmphasis)+

[0027] (1.52)*(wavelet-LHL_firstorder_Median)+

[0028] (-0.00000284)*(wavelet-LHL_glszm_LargeAreaLowGrayLevelEmphasis)+

[0029] (-3.89)*(wavelet-LHL_glszm_SmallAreaEmphasis)+

[0030] (-0.406)*(wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis)+

[0031] (0.00000000000506)*(wavelet-HHH_glszm_LargeAreaEmphasis)+

[0032] (0.000000000187)*(wavelet-HHH_glszm_ZoneVariance)+

[0033] (0.000237)*(wavelet-LLL_gldm_SmallDependenceHighGrayLevelEmphasis)+

[0034] (4.03)*(log-sigma-1-4-mm-3D_glcm_Imc1)+

[0035] (-0.852)*(log-sigma-1-4-mm-3D_glcm_InverseVariance)

[0036] The method described in 1 above is characterized in that the AGRC in step (5) is (CTACS 随访 —CTACS 基线 ) / CT interval time.

[0037] It should be noted that the model parameters described in step S6 include radiomics score, hypertension grade and blood sugar.

[0038] The beneficial effect of the present invention is that it makes full use of the chest calcification score. CTACS not only includes CACS, but also includes other chest artery calcification scores. The research of the present invention shows that CTACS is better than CACS in predicting MACEs, which is reflected in higher specificity and sensitivity. On the basis of CTACS, the radiomics integral parameter is constructed using radiomics technology. This parameter can not only reflect CTACS, that is, the chest artery calcification situation, but also can further improve the MACEs prediction performance. Finally, the radiomics-clinical variable model constructed in combination with the radiomics integral has good MACEs prediction ability, and is expected to provide patients with an accurate MACEs occurrence prediction platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Figure 2 is the ROC curve of CTACS and CACS for predicting MACEs, where CTACS: thoracic common artery calcium score; CACS: coronary artery calcium score; MACEs: major adverse cardiovascular events;

[0040] Figure 2 Figure 3 is the ROC curve of radiomics score, CTACS, CACS and AGRC for the prediction of MACEs and acute heart failure, where Figure A is the prediction of MACEs and Figure B is the prediction of acute heart failure; AGRC: annual growth rate of calcification;

[0041] Figure 3 Nomogram visualization of the radiomics-clinical variable model;

[0042] Figure 4 Figure 3 is the decision analysis curve diagram of the radiomics-clinical variable model, where Figure A is the training group and Figure B is the validation group. Specific embodiments

[0043] The present invention will be further described below in conjunction with the accompanying drawings. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0044] The present invention provides a prediction model for major adverse cardiovascular events based on thoracic artery calcification and a construction method thereof, the method comprising the following steps:

[0045] S1 collects clinical data of patients who have undergone chest plain scan CT and CT mediastinal window sequence image data, wherein the image data is DICOM images;

[0046] S2 divides the primary and secondary arteries of the CT image data of step S1 into segments, a total of 10 arteries; and uses Agatston integration to calculate the calcification scores of the 10 arteries; the thoracic common artery calcification score (CTACS) is the sum of the calcification scores of the 10 arteries;

[0047] S3 compared the performance of CTACS and CACS in predicting MACEs by using the AUC value of MACEs prediction;

[0048] S4 segmented the 10 arteries in step S2, extracted radiomic features, screened radiomic features related to the occurrence of MACEs through LASSO regression analysis, and obtained the corresponding radiomic score calculation formula;

[0049] S5 The AUC value of MACEs prediction was used to compare the performance of radiomics score with CTACS, CACS and annual growth rate of calcification (AGRC) in predicting MACEs;

[0050] S6 screened the radiomics scores and clinical data collected in step S1 by Spearman correlation analysis, univariate and multivariate Cox regression analysis to obtain parameters associated with the occurrence of MACEs;

[0051] S7 constructs a composite prediction model of radiomics-clinical variables based on the features obtained in step S6, and performs model evaluation.

[0052] It should be noted that the inclusion criteria for patients who have undergone chest CT scans in step S1 include: patients who have undergone at least 2 chest CT scans with an interval of more than 180 days, and the collected mediastinal window sequence image specifications are 3 mm layer thickness and 512*512 pixel matrix; the image analysis software includes I TK-SNAP and python 3.6.5 platforms.

[0053] It should be noted that the clinical data described in step S1 include the following clinical characteristics: age, gender, whether MACEs have occurred, whether there are underlying cardiovascular diseases, whether there are underlying cerebrovascular diseases, whether there are chronic kidney diseases, whether there are diabetes, smoking status (yes / no), drinking status (yes / no), systolic blood pressure / diastolic blood pressure, hypertension grade, urea nitrogen, serum creatinine, lipoprotein a, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, total cholesterol, blood glucose and blood calcium.

[0054] It should be noted that the definition of MACEs includes sudden cardiac death, acute heart failure, acute myocardial infarction, unstable angina requiring hospitalization, malignant arrhythmia requiring hospitalization, and stroke.

[0055] It should be noted that the 10 arteries described in step S2 include the aortic arch, ascending aorta, descending aorta, left subclavian artery, left common carotid artery, brachiocephalic trunk, right subclavian artery, right common carotid artery, coronary artery and pulmonary artery trunk; the CTACS=Σ(AS 冠状动脉 +AS 升主动脉 +AS 主动脉弓 +AS 降主动脉 +AS 肺动脉干 +AS 头臂干 +AS 右颈总动脉 +AS 右锁骨下动脉 +AS 左颈总动脉 +AS 左锁骨下动脉 ).

[0056] It should be noted that the AUC values ​​of step S3 and step S5 were compared using the Delong test, and a p value less than 0.05 was considered to be statistically different.

[0057] It should be noted that the method for extracting radiomic features from the CT mediastinal window in step S4 is: segmenting the 10 arteries layer by layer, stacking the layers to form a three-dimensional arterial segmentation body, and extracting radiomic features from the segmentation body.

[0058] It should be noted that the radiomics score in step S4 = (0.106)*(original_shape_Elongation)+

[0059] (0.00412)*(original_shape_Maximum2DDiameterSlice)+

[0060] (-116)*(wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis)+

[0061] (-67.5)*(wavelet-LLH_gldm_LowGrayLevelEmphasis)+

[0062] (1.52)*(wavelet-LHL_firstorder_Median)+

[0063] (-0.00000284)*(wavelet-LHL_glszm_LargeAreaLowGrayLevelEmphasis)+

[0064] (-3.89)*(wavelet-LHL_glszm_SmallAreaEmphasis)+

[0065] (-0.406)*(wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis)+

[0066] (0.00000000000506)*(wavelet-HHH_glszm_LargeAreaEmphasis)+

[0067] (0.000000000187)*(wavelet-HHH_glszm_ZoneVariance)+

[0068] (0.000237)*(wavelet-LLL_gldm_SmallDependenceHighGrayLevelEmphasis)+

[0069] (4.03)*(log-sigma-1-4-mm-3D_glcm_Imc1)+

[0070] (-0.852)*(log-sigma-1-4-mm-3D_glcm_InverseVariance)

[0071] The method described in 1 above is characterized in that the AGRC in step (5) is (CTACS 随访 —CTACS 基线 ) / CT interval time.

[0072] It should be noted that the model parameters described in step S6 include radiomics score, hypertension grade and blood sugar.

[0073] Example

[0074] It should be noted that, in the present embodiment, all data statistics of the present invention are analyzed under SPSS 25 or R 3.6.3 software. Since the continuous variables involved in the present invention do not obey the normal distribution, the median (quartile interval) is used to represent, and the Mann-Whitney U test is used to compare the two groups. Categorical variables are represented by numerical values ​​(percentages), and the x2 test or Fisher's exact probability method is used to compare the two groups. Cox regression analysis is used to screen model features and build a prediction model. The "glmnet" package is used to run LASSO regression analysis. The "pROC" package is used to draw ROC curves. The "rms" package is used to draw nomograms and calibration curves. "code source stdca.R" is used to draw decision analysis curves. Bilateral p<0.05 is considered to be statistically significant.

[0075] 1. Patient screening and data collection

[0076] A total of 358 subjects were included between January 2014 and August 2018. They were divided into primary, secondary, and tertiary cardiovascular predictors according to whether they had ever experienced MACEs or had cardiovascular and cerebrovascular diseases. Inclusion criteria were: (1) patients who had undergone at least 2 chest CT scans with an interval of more than 180 days; (2) the collected mediastinal window sequence images were 3 mm thick and 512*512 pixel matrix. Exclusion criteria included: (1) patients younger than 18 years old; (2) patients with malignant tumors or malignant tumors found during follow-up; (3) pregnant women; (4) patients with metal stents implanted in the coronary artery and aorta; (5) patients with acute and severe infectious diseases in the baseline records and image information that could not be analyzed by image analysis software; (6) patients with complete lack of clinical data.

[0077] Within one week before and after the baseline CT scan, the clinical data of the subjects were collected: age, gender, whether MACEs had occurred, whether there were underlying cardiovascular diseases, whether there were underlying cerebrovascular diseases, whether there were chronic kidney diseases, whether there were diabetes, smoking status (yes / no), drinking status (yes / no), systolic / diastolic blood pressure, hypertension grade, urea nitrogen, serum creatinine, lipoprotein a, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, total cholesterol, blood glucose and blood calcium.

[0078] 2. Obtain CT mediastinal window sequence images and calculate calcification scores

[0079] All chest plain CT scans were performed using Siemens (SOMATOM Sensation) spiral detection CT scans, with the scanning range from the apex (approximately at the level of the lower edge of the cricoid cartilage) to the pulmonary base (approximately 1-2 cm below the aortic hiatus). We collected medical imaging images of the mediastinal window sequence, which were saved in DICOM (digital imaging and communications in medicine) format. The specifications of the collected DICOM images were as follows: (1) the reconstructed slice thickness was 3 mm; (2) each slice thickness was 512 × 512 pixel matrix; (3) the voxel size ranged from 0.5 to 0.6.

[0080] The thoracic arteries were divided into the following 10 branches: coronary artery, ascending aorta, aortic arch, descending aorta, pulmonary trunk, brachiocephalic trunk, right common carotid artery, right subclavian artery, left common carotid artery and left subclavian artery. The calcification score of the 10 arteries was calculated using the Agatston score. The formula for calculating the calcification score is: Agatston score = Σ (calcification area × weight coefficient), where the weight coefficients are: 1 = 130-199HU, 2 = 200-299HU, 3 = 300-399HU, 4 = ≥ 400HU. Where CTACS = Σ (AS 冠状动脉 +AS 升主动脉 +AS 主动脉弓 +AS 降主动脉 +AS 肺动脉干 +AS 头臂干 +AS 右颈总动脉 +AS 右锁骨下动脉 +AS 左颈总动脉 +AS 左锁骨下动脉 ). The AGRC = (CTACS 随访 —CTACS 基线 ) / CT interval time.

[0081] 3. Artery segmentation and extraction of radiomics features

[0082] Two researchers (one radiologist with 9 years of experience; the other a student receiving mentoring training) independently used ITK-SNAP (version 3.8) to segment the artery, and one of the researchers repeated the segmentation again after an interval of 1 month. The segmentation was based on outlining the arterial wall contours. The outlined arterial regions were two-dimensional regions of interest (ROIs), and consecutive ROIs were stacked to form three-dimensional volumes of interest (VOIs). Subsequently, the pyradiomics package and microwave transformation were used to extract radiomic features for each VOI in the python 3.6.5 platform. 944 radiomic features were extracted for each VOI.

[0083] In addition, the intra-group / inter-group correlation coefficients were calculated by comparing the arterial segmentation features of the two researchers to clarify the reliability and repeatability of the imaging features. The intra-group correlation coefficients were 0.951-0.983, and the inter-group correlation coefficients were 0.915-0.934, both greater than 0.9, indicating that arterial segmentation and feature extraction have good stability.

[0084] 4. Evaluation of CTACS in the prediction of MACEs

[0085] The prediction performance of CTACS and CACS was compared by the area under the ROC curve of MACEs prediction. Among them, the AUC value of CTACS for MACEs prediction was 0.815, specificity / sensitivity: 0.670 / 0.904; while the AUC value of CACS for MACEs prediction was 0.678, specificity / sensitivity: 0.624 / 0692. Through the Delong test, the difference in AUC values ​​between the two was p<0.05, which proved the purpose of the study: CTACS is superior to coronary artery calcification score in the prediction of MACEs.

[0086] 5. Grouping of training group, validation group and additional validation group

[0087] Of the 358 subjects, 246 subjects with rich clinical variables were selected, and numerical values ​​were randomly generated by computer and arranged in ascending order. At the same time, they were randomly divided into a training group of 198 cases and a validation group of 48 cases in a stratified manner according to a 4:1 ratio (the first 80% and the last 20% of the ascending order); the stratification basis included (1) the incidence of MACEs; (2) CTACS distribution. The remaining 112 subjects were automatically divided into an additional validation group.

[0088] 6. Construction of radiomics scores and evaluation of their predictive ability

[0089] In the training group, the LASSO regression method was used to screen the radiomics features related to the occurrence of MACEs from the extracted radiomics features, and the radiomics score was calculated to predict whether the research subjects were at high risk of MACEs.

[0090] Among them, a total of 13 radiomic features used to construct the radiomics score were obtained after ten-fold cross validation. The radiomics features include: original_shape_Elongation, original_shape_Maximum2DDiameterSlice, wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis, wavelet-LLH_gldm_LowGrayLevelEmphasis, wavelet-LHL_firstorder_Median, wavelet-LHL_glszm_LargeAreaLowGrayLevelEmphasis, wavelet-LHL_glszm_S mallAreaEmphasis, wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis, wavelet-HHH_glszm_LargeAreaEmphasis, wavelet-HHH_glszm_ZoneVaria nce, wavelet-LLL_gldm_SmallDependenceHighGrayLevelEmphasis, log-sigma-1-4-mm-3D_glcm_Imc1 and log-sigma-1-4-mm-3D_glcm_InverseVariance.

[0091] The radiomics score was obtained by linearly adding the coefficients obtained by LASSO regression. The calculation formula is as follows:

[0092] The radiomics score = (0.106) * (original_shape_Elongation) +

[0093] (0.00412)*(original_shape_Maximum2DDiameterSlice)+

[0094] (-116)*(wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis)+

[0095] (-67.5)*(wavelet-LLH_gldm_LowGrayLevelEmphasis)+

[0096] (1.52)*(wavelet-LHL_firstorder_Median)+

[0097] (-0.00000284)*(wavelet-LHL_glszm_LargeAreaLowGrayLevelEmphasis)+

[0098] (-3.89)*(wavelet-LHL_glszm_SmallAreaEmphasis)+

[0099] (-0.406)*(wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis)+

[0100] (0.00000000000506)*(wavelet-HHH_glszm_LargeAreaEmphasis)+

[0101] (0.000000000187)*(wavelet-HHH_glszm_ZoneVariance)+

[0102] (0.000237)*(wavelet-LLL_gldm_SmallDependenceHighGrayLevelEmphasis)+

[0103] (4.03)*(log-sigma-1-4-mm-3D_glcm_Imc1)+

[0104] (-0.852)*(log-sigma-1-4-mm-3D_glcm_InverseVariance)

[0105] The constructed radiomics score was strongly correlated with CTACS, with a correlation coefficient of r = 0.849 and a linear R 2 =0.614, indicating that the radiomics score can reflect the CTACS situation to some extent.

[0106] The discriminability of the radiomics score was judged by the AUC value and tested in the training group, validation group, and additional validation group. The AUC values ​​in the training group, validation group, and additional validation group were 0.841, 0.845, and 0.718, respectively, indicating that the radiomics score had good performance in predicting MACEs.

[0107] The prediction performance of radiomics score, CTACS, CACS and AGRC was compared by the area under the ROC curve of MACEs prediction. The results showed that the AUC value of radiomics score was significantly higher than that of CTACS, CACS and AGCR, with AUC values ​​of 0.879, 0.815, 0.678 and 0.654, respectively, and the p values ​​of the difference compared with the AUC value of radiomics score were all less than 0.05. In addition, in the prediction of acute heart failure, the AUC values ​​of radiomics score compared with CTACS, CACS and AGRC were 0.008, 0.007 and 0.082, respectively. The results proved the purpose of the study 2: radiomics score is better than traditional calcification evaluation indicators in predicting MACEs, especially in heart failure.

[0108] 7. Construction of radiomics-clinical variable prediction model and evaluation of prediction performance

[0109] In the training group, 11 features related to the occurrence of MACEs were screened by Spearman correlation analysis, including age, systolic blood pressure, hypertension grade, cardiovascular disease, cerebrovascular disease, chronic kidney disease, diabetes, urea nitrogen, eGFR (estimated glomerular filtration rate), blood glucose and radiomics score. Subsequently, radiomics score, hypertension grade and blood glucose were screened as independent predictive features of MACEs by univariate and multivariate Cox regression analysis.

[0110] The screened radiomics scores, hypertension grades and blood glucose were used to establish a radiomics-clinical variable MACEs prediction model through multivariate Cox regression analysis.

[0111] The model is evaluated and tested in the training group and the validation group, and the evaluation and testing methods include a C index and a calibration curve. Among them, the C index reflects the distinguishing ability of the model's predicted probability, and the C index in the training group is 0.872, and in the validation group is 0.660. The calibration curve reflects the consistency of the model's predicted probability, where the closer the predicted probability curve (red line in the figure) and the actual probability of occurrence (gray dotted line in the figure) are, the higher the prediction accuracy. As can be seen from the figure, whether in the training group or the validation group, the predicted probability and the actual probability are relatively close, suggesting that the prediction model has good prediction performance.

[0112] Finally, the decision analysis curve was used to evaluate the clinical applicability of the prediction model. The decision analysis curves of the training group and the validation group showed that the clinical benefits obtained by using the prediction model for clinical decision-making were better than those of the no treatment plan and / or the full treatment plan.

[0113] For those skilled in the art, various corresponding changes can be made according to the above technical solutions and concepts, and all of these changes should be included in the protection scope of the claims of the present invention.

Claims

1. A method for constructing a prediction model for major adverse cardiovascular events based on thoracic artery calcification. It is characterized in that The method comprises the following steps: S1 collects clinical data of patients who have undergone chest plain scan CT and CT mediastinal window sequence image data, wherein the image data is DICOM images; S2 divides the primary and secondary arteries of the CT mediastinal window sequence image data of step S1 into segments, which are 10 in total; and uses Agatston integration to calculate the calcification scores of the 10 arteries; the calcification score of the thoracic common artery is the sum of the calcification scores of the 10 arteries; S3 compared the performance of thoracic common artery calcium score CTACS and coronary artery calcium score CACS in predicting MACEs by using the AUC value of MACEs prediction; The definition of MACEs included sudden cardiac death, acute heart failure, acute myocardial infarction, unstable angina requiring hospitalization, malignant arrhythmia requiring hospitalization, and stroke; The 10 arteries in step S2 include the aortic arch, ascending aorta, descending aorta, left subclavian artery, left common carotid artery, brachiocephalic trunk, right subclavian artery, right common carotid artery, coronary artery and pulmonary artery trunk; the CTACS=Σ(AS 冠状动脉 +AS 升主动脉 +AS 主动脉弓 +AS 降主动脉 +AS 肺动脉干 +AS 头臂干 +AS 右颈总动脉 +AS 右锁骨下动脉 +AS 左颈总动脉 +AS 左锁骨下动脉 ), where AS represents the Agatston integral; S4 segmented the 10 arteries in step S2, extracted radiomic features, screened radiomic features related to the occurrence of MACEs through LASSO regression analysis, and obtained the corresponding radiomic score calculation formula; The radiomics score in step S4 = (0.106)*(original_shape_El ongation)+ (0.00412)*(original_shape_Maximum2DDiameterSlice)+ (-116)*(wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis)+ (-67.5)*(wavelet-LLH_gldm_LowGrayLevelEmphasis)+ (1.52)*(wavelet-LHL_firstorder_Median)+ (-0.00000284)*(wavelet-LHL_glszm_LargeAreaLowGrayLeve lEmphasis)+ (-3.89)*(wavelet-LHL_glszm_Smal lAreaEmphasis)+ (-0.406)*(wavelet-LHL_glszm_Smal lAreaLowGrayLevelEmph asis)+ (0.00000000000506)*(wavelet-HHH_glszm_LargeAreaEmphas is)+ (0.000000000187)*(wavelet-HHH_glszm_ZoneVariance)+ (0.000237)*(wavelet-LLL_gldm_Smal lDependenceHighGrayL evelEmphasis)+ (4.03)*(log-sigma-1-4-mm-3D_glcm_Imc1)+ (-0.852)*(log-sigma-1-4-mm-3D_glcm_InverseVariance); S5 compared the performance of radiomics score with CTACS, CACS and annual calcification growth rate ARGC in predicting MACEs by using the AUC value of MACEs prediction; AGRC=(CTACS 随访 —CTACS 基线 ) / CT interval time; among them, CTACS 随访 The second chest CT scan is for the total artery calcification score, CTACS 基线 The score of common artery calcification in the initial chest CT scan; S6 screened the radiomics scores and clinical data collected in step S1 by Spearman correlation analysis, univariate and multivariate Cox regression analysis to obtain parameters associated with the occurrence of MACEs; S7 constructs a composite prediction model of radiomics-clinical variables based on the parameters obtained in step S6, and performs model evaluation.

2. The method for constructing a prediction model for major adverse cardiovascular events based on thoracic artery calcification according to claim 1, It is characterized in that The inclusion criteria for patients who have undergone chest CT scans in step S1 include: (1) patients who have undergone at least 2 chest CT scans with an interval greater than 180 days; and (2) the specifications of the collected mediastinal window sequence images are 3 mm slice thickness and 512*512 pixel matrix.

3. The method for constructing a prediction model for major adverse cardiovascular events based on thoracic artery calcification according to claim 1, It is characterized in that The clinical data in step S1 include the following clinical characteristics: Age, gender, whether MACEs had occurred, whether there was underlying cardiovascular disease, whether there was underlying cerebrovascular disease, whether there was chronic kidney disease, whether there was diabetes, smoking status (yes / no), drinking status (yes / no), systolic / diastolic blood pressure, hypertension grade, urea nitrogen, serum creatinine, lipoprotein a, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, total cholesterol, blood glucose and blood calcium.

4. The method for constructing a prediction model for major adverse cardiovascular events based on thoracic artery calcification according to claim 1, It is characterized in that The AUC values ​​of step S3 and step S5 were compared using the Delong test, and a p value less than 0.05 was considered to be statistically different.

5. The method for constructing a prediction model for major adverse cardiovascular events based on thoracic artery calcification according to claim 1, It is characterized in that The radiomics feature extraction method described in step S4 is: segmenting the 10 arteries layer by layer, stacking the layers to form a three-dimensional artery segmentation body, and extracting radiomics features from the segmentation body.

6. The method for constructing a prediction model for major adverse cardiovascular events based on thoracic artery calcification according to claim 1, It is characterized in that The parameters described in step S6 include radiomics score, hypertension grade and blood sugar.