CN118398211A - Kidney stone risk prediction method and system based on CT image histology - Google Patents

Abstract

The present invention provides a method and system for predicting kidney stone risk based on CT radiomics, comprising: collecting multi-axial CT images of multiple patients and outlining them to obtain results of regions of interest, screening out the most relevant target radiomics features based on LASSO, building models through algorithms such as logistic regression and random forest, and screening out the optimal algorithm for building the prediction model based on comparison of prediction performance; collecting clinical variables of multiple patients respectively, and screening out independent risk factors from the clinical variables based on univariate analysis models and multivariate regression analysis in turn; combining the target radiomics features with the independent risk factors to obtain combined features, and obtaining classification results of risk prediction corresponding to the combined features based on the constructed prediction model; and building a nomogram based on the combined features to explain the classification results; the present invention can improve prediction accuracy and interpretability.

Description

A method and system for predicting kidney stone risk based on CT radiomics

Technical Field

The present invention relates to the technical field of image processing, and in particular to a method and system for predicting kidney stone risk based on CT radiomics.

Background technique

Urinary stones are the most common disease in urology, with high incidence and recurrence rates, and kidney stones are the most common type. At present, most patients with kidney stones seek medical treatment due to symptoms such as renal colic and hematuria, or are accidentally discovered during physical examinations or treatment of other diseases. The diagnosis is often late, which is not conducive to the treatment of kidney stones. Studies have reported that about 50% of patients with symptomatic stones require surgical intervention. Therefore, prevention is crucial for the management of kidney stone disease. Screening high-risk groups is the key to preventing kidney stones. It is conducive to early intervention, such as dietary adjustments to reduce the incidence rate; it is conducive to the timely detection of stones, so that patients can receive timely treatment, which is conducive to controlling the disease and limiting surgical injuries.

Computed Tomography (CT) is the gold standard for diagnosing kidney stones and the preferred imaging evaluation method for patients with renal colic. However, some patients with smaller stones are missed due to thick CT scan layers or neglect by radiologists. Abroad, about 2 million people go to the emergency department each year for renal colic or stone-related back pain. About half of them receive CT scans, and only 20% are diagnosed with kidney stones. Therefore, there is an urgent need for an effective method to assist in screening high-risk groups. Researchers have tried to screen people at high risk of kidney stones through medical imaging, such as monitoring renal papillary CT values and tracking data.

However, when detecting kidney stones through CT, if the diameter of the kidney stone is too small, smaller than the CT scan layer thickness or reconstruction layer thickness, it cannot be detected; and CT cannot screen out people at high risk of kidney stones. Measuring the renal papillary CT value can help screen people at high risk of kidney stones, but existing studies only suggest that increased renal papillary CT values increase the risk of kidney stones and cannot predict the probability of stone risk, making it difficult to interpret the prediction results of whether or not to have kidney stones; in addition, only measuring the renal papillary CT value may lead to omission of information and low accuracy.

Summary of the invention

The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and to propose a kidney stone risk prediction method and system based on CT imaging genomics, which can improve the accuracy of kidney stone prediction and improve the interpretability of kidney stone prediction.

In a first aspect, the present invention provides a method for predicting kidney stone risk based on CT radiomics, comprising:

Collect multi-axial CT images of multiple patients respectively, delineate regions of interest on the CT images, obtain region of interest results, and extract target imaging omics features from the region of interest results;

Clinical variables of multiple patients were collected respectively, and the independent risk of kidney stones in clinical variable images was obtained based on the univariate analysis model. Then, the cross-influencing factors of the independent risks were controlled based on the multivariate analysis model to obtain independent risk factors.

Combining the target radiomics feature with the independent risk factor to obtain a combined feature, and obtaining a classification result of risk prediction corresponding to the combined feature based on a prediction model;

A nomogram is constructed according to the combined features, and the classification results are probability visualized according to the nomogram to obtain the risk probability, and the classification results are explained according to the risk description corresponding to the risk probability.

The present invention obtains multi-axial CT images of renal tissue to comprehensively detect renal tissue, thereby avoiding the omission of useful CT image information, thereby improving the accuracy of kidney stone risk prediction; and, a prediction model is used to first obtain the independent risk affecting kidney stones from clinical variables, and then based on the same prediction model, the cross-influencing factors affecting the independent risk are controlled to obtain more accurate independent risk factors, thereby further improving the accuracy of kidney stone risk prediction; then, risk prediction is performed in combination with the target imaging genomics features of the CT image and the independent risk factors, and a nomogram is used to obtain the risk probability, and the classification results are interpreted according to the risk description corresponding to the risk probability, thereby improving the interpretability of kidney stone risk prediction.

Furthermore, based on the univariate analysis model, the independent risk of suffering from kidney stones based on clinical variable images is obtained respectively, and then based on the multivariate analysis model, the cross-influencing factors of the independent risk are controlled to obtain independent risk factors, including:

The clinical variables are input into the univariate analysis model in sequence, and the first significant difference index of kidney stones corresponding to the univariate images is output respectively. The corresponding clinical variables of the first significant difference index that meets the preset index threshold are input into the multivariate analysis model, the cross-influencing factors between the single factors are controlled, and the second significant difference index is output. According to the second significant difference index, independent risk factors are selected from the clinical variables.

Further, constructing a nomogram according to the combined features, and performing probability visualization on the classification results according to the nomogram to obtain the risk probability, includes:

The scores of the target radiomics feature and the independent risk factor in the nomogram are obtained respectively, the target radiomics feature score and the independent risk factor score are obtained correspondingly, the total score of the target radiomics feature score and the independent risk factor score is obtained, and the corresponding risk probability is obtained according to the total score.

Further, obtaining the target radiomics feature score includes:

According to the coefficient of the target radiomics feature, weighting the target radiomics feature to obtain a target radiomics feature score;

Wherein, the target radiomics feature score is expressed as:

Radiomics score＝a1*F1+a2*F2+...+ai*Fi+b;

Among them, a1 to ai are the coefficients of the target imaging features, F1 to Fi are the target imaging features; the number of target imaging features is i; and b is the intercept.

Further, obtaining the classification result of the risk prediction corresponding to the combined feature based on the prediction model includes:

The combined feature is input into an imaging omics model based on the prediction model, and a classification result of risk prediction corresponding to the combined feature is output; the imaging omics model is obtained after training based on the input of the target imaging omics feature and the intercept corresponding to the target imaging omics feature.

Furthermore, the radiomics model is obtained by training according to the input of the target radiomics feature and the intercept corresponding to the target radiomics feature, and includes:

A training set and a validation set of the target imaging features and the intercept corresponding to the target imaging features are obtained, and a plurality of different machine learning algorithm models are trained and validated according to the training set and the validation set, and the algorithm model with the smallest AUC index difference between the training result and the validation result is selected as the imaging model.

The present invention selects the final imaging omics model according to the algorithm model with the smallest AUC value and the AUC difference between the training set and the validation set, which can ensure the stability and generalization ability of the imaging omics model, thereby improving the accuracy of risk prediction.

Further, the step of extracting target radiomics features from the region of interest result includes:

Extract multiple radiomics features of the region of interest, and have two experts extract the same radiomics feature to obtain two groups of extraction results, respectively. Evaluate the intra-group consistency of the two groups of extraction results to obtain the target radiomics feature; the multiple radiomics features include: first-order features, shape features and texture features.

Furthermore, the intra-group consistency of the two groups of extraction results is evaluated to obtain target imaging omics features, including:

The imaging features with intra-group correlation coefficients not less than a preset threshold are retained to obtain multiple candidate imaging features, and data dimension reduction or parameter compression is performed on the multiple candidate imaging features to screen out multiple target imaging features with the highest correlation with kidney stones.

Furthermore, the step of respectively acquiring multi-axial CT images of multiple patients, outlining regions of interest on the CT images, and obtaining region of interest results includes:

Acquiring multi-axial CT images of renal tissues of multiple patients respectively, wherein the CT images include: cross-sectional, coronal and sagittal CT images of renal tissues;

The air and fat tissue surrounding the kidney tissue in the CT images of the cross section, the coronal plane and the sagittal plane are simultaneously shielded by a preset image threshold, and the kidney tissue is outlined respectively to obtain a region of interest.

In a second aspect, the present invention provides a kidney stone risk prediction system based on CT radiomics, comprising: a target radiomics feature extraction unit, an independent risk factor extraction unit, a classification result acquisition unit and an interpretation unit; wherein,

The target imaging genomics feature extraction unit is used to respectively collect multi-axial CT images of multiple patients, outline the region of interest on the CT images, obtain the region of interest results, and extract the target imaging genomics features from the region of interest results;

The independent risk factor extraction unit is used to collect clinical variables of multiple patients respectively, obtain independent risks of kidney stones in clinical variable images based on a univariate analysis model, and then control the cross-influencing factors of the independent risks based on a multivariate analysis model to obtain independent risk factors;

The classification result acquisition unit is used to combine the target imaging genomics feature with the independent risk factor to obtain a combined feature, and obtain a classification result of the risk prediction corresponding to the combined feature based on a prediction model;

The explanation unit is used to construct a nomogram according to the combined features, and to perform probability visualization of the classification results according to the nomogram to obtain the risk probability, and to explain the classification results according to the risk description corresponding to the risk probability.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic flow chart of a method for predicting kidney stone risk based on CT imaging genomics provided in this embodiment;

FIG2 is a schematic diagram of extracting a region of interest provided by this embodiment;

FIG3 is a schematic diagram of the AUC of the training results and verification results of different machine learning algorithm models provided in this embodiment;

FIG4 is a schematic diagram of the AUC of the prediction model of independent risk factors and target radiomics features provided in this embodiment;

FIG5 is a schematic diagram of a calibration curve and a clinical decision curve based on an imaging omics model provided in this embodiment;

FIG6 is a schematic diagram of a nomogram I provided in this embodiment;

FIG7 is a schematic diagram of a nomogram II provided in this embodiment;

FIG8 is a flow chart of another method for predicting kidney stone risk based on CT radiomics provided in this embodiment;

FIG9 is a schematic diagram of the structure of a kidney stone risk prediction system based on CT imaging omics provided in this embodiment.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

It is worth noting that the existing related technologies use renal papillary CT values to predict the risk of kidney stones, but they cannot predict the probability of kidney stone risk, which makes it difficult to interpret whether one has kidney stones. When the related technologies using renal papillary CT values cannot scan smaller kidney stones, they cannot predict the risk of kidney stones, and they only rely on renal papillary CT values for risk prediction, omitting other characteristic information, resulting in low accuracy.

Based on this, the present invention provides a method and system for predicting kidney stone risk based on CT imaging genomics, which comprehensively detects kidney tissue by acquiring CT images of multi-axial surfaces of kidney tissue, avoiding the omission of useful CT image information, thereby improving the accuracy of kidney stone risk prediction; and, using a univariate analysis model to first obtain the independent risk affecting kidney stones from clinical variables, and then based on a multivariate analysis model, control the cross-influencing factors affecting the independent risk to obtain more accurate independent risk factors, thereby further improving the accuracy of kidney stone risk prediction; then, combining the target imaging genomics features of the CT image and the independent risk factors to predict risk, since the classification result of risk prediction is based on an end-to-end result, either there is a risk of disease or there is no risk of disease, its interpretability is poor, therefore, after risk prediction, the present invention uses a nomogram to obtain the risk probability, and interprets the classification result according to the risk description corresponding to the risk probability, that is, explains why there is a risk of disease or why there is no risk of disease, thereby improving the interpretability of kidney stone risk prediction.

In order to better illustrate the technical solution of the present invention, it will be described in detail from the following embodiments.

Example 1

Referring to FIG. 1 , it is a flow chart of a method for predicting kidney stone risk based on CT radiomics provided in this embodiment, including steps S11 to S14, specifically:

Step S11, respectively collect multi-axial CT images of multiple patients, delineate regions of interest on the CT images, obtain region of interest results, and extract target imaging genomics features from the region of interest results.

In some embodiments, multi-axial CT images of multiple patients are collected separately, and regions of interest are outlined on the CT images to obtain region of interest results, including: collecting multi-axial CT images of renal tissue of multiple patients separately, the CT images including: cross-sectional, coronal and sagittal CT images of renal tissue; simultaneously masking the air and fat tissue surrounding the renal tissue in the cross-sectional, coronal and sagittal CT images by a preset image threshold, and outlining the renal tissue separately to obtain the region of interest.

In some embodiments, patients who were treated from 2020 to 2022 were retrospectively collected, of which 222 had kidney stones and 291 had no kidney stones. The cases were preliminarily screened through the scientific research big data platform.

In some embodiments, CT images are acquired by a RUIKE system and saved in DCM format; wherein the CT images are acquired using Siemens SOMATOM Definition AS+ 128-slice CT or 64-slice CT (Siemens, Erlangen, Germany). CT imaging acquisition parameters include: tube voltage 120 kV; automatic tube voltage; ref mAs 200-500; matrix 512×512; reconstruction thickness is mostly 2 mm, and a few are 5 mm.

In some embodiments, 3Dslicer software is used by two experts to delineate a region of interest (ROI), wherein the ROI is the kidney and the segmentation mode is manual segmentation. The segmentation margin is adjusted to include as much renal tissue as possible while avoiding stones and perirenal tissue.

In some embodiments, to make ROI segmentation easier and faster, the Threshold function of the 3Dslicer software is used to mask the image, and the selected image threshold is 0, so that the kidney boundary can be clearly segmented without outlining the air and fat tissue around the kidney.

In some embodiments, referring to FIG. 2 , it is a schematic diagram of the region of interest extraction provided by the present embodiment. When outlining the region of interest in FIG. 2 , it is necessary to comply with the following principles of image segmentation: 1) the segmentation of the ROI should be performed in the plain scan CT image, and stones and perirenal tissue should be avoided during the image segmentation process. 2) Image segmentation involves full-thickness segmentation, that is, all CT layers including renal tissue are segmented and included in the ROI. According to the above principles, full-thickness outlining is performed on the cross-section, coronal and sagittal planes, and the full-thickness outlining results are combined to obtain a three-dimensional outlining result of the renal tissue.

It is worth mentioning that radiomics refers to the high-throughput extraction of radiomics features from medical images through computer image processing technology based on medical image analysis, including first-order features, texture features, shape features, etc.

In some embodiments, radiomics features are extracted through a plug-in of 3Dslicer software, including: first-order features, shape features, and texture features.

In some embodiments, radiomics features are extracted through a plug-in of 3Dslicer software, including first-order features (18 features), shape features (14 features), gray-level co-occurrence matrix (GLCM) features (24 features), gray-level size zone matrix (GLSZM) features (16 features), gray-level run length matrix (GLRLM) features (16 features), neighboring gray-level difference matrix (NGTDM) features (5 features), and gray-level dependency matrix (GLDM) features (14 features). A total of 7 categories and 107 features were extracted. For detailed feature names, see Table 1, Radiomics Feature Table.

Table 1 Radiomics characteristics

In some embodiments, target radiomics features are extracted from the results of the region of interest, including: after extracting multiple radiomics features of the region of interest, two experts extract the same radiomics features to obtain two groups of extraction results respectively, and the two groups of extraction results are evaluated for intra-group consistency to obtain the target radiomics features; the multiple radiomics features include: first-order features, shape features and texture features.

In some embodiments, the intra-group consistency of the two groups of extraction results is evaluated to obtain target radiomics features, including: retaining radiomics features with an intra-group correlation coefficient not less than a preset threshold, obtaining multiple candidate radiomics features, and performing data dimension reduction or parameter compression on the multiple candidate radiomics features to screen out multiple target radiomics features with the highest correlation with kidney stones.

In some embodiments, the intraclass correlation coefficient (ICC) is used to evaluate the intra-group and inter-group consistency of feature extraction. Expert A and expert B extract radiomic features of the same CT image to evaluate the inter-group consistency; one of the experts repeats the extraction of the radiomic features extracted by the other expert at least one month later, and then evaluates the intra-group consistency.

In some embodiments, radiomics features with an intra-group correlation coefficient less than a preset threshold of 0.75 are deleted, and the remaining radiomics features are used as candidate radiomics features.

In some embodiments, the parameters of multiple candidate radiomics features are compressed by the least absolute shrinkage and selection operator (LASSO) to screen out multiple target radiomics features with the highest correlation with kidney stones.

In some embodiments, principal component analysis is used to perform data dimension reduction on multiple candidate radiomics features to screen out multiple target radiomics features that are most correlated with kidney stones.

In some embodiments, the target imaging genomics features refer to Table 2, which is a table of target imaging genomics features and corresponding coefficients provided in this embodiment.

Table 2 Target imaging omics features and corresponding coefficients

Step S12: clinical variables of multiple patients are collected respectively, and independent risks of kidney stones in clinical variable images are obtained based on a univariate analysis model. Then, based on a multivariate analysis model, cross-influencing factors of independent risks are controlled to obtain independent risk factors.

In some embodiments, clinical variables include: baseline information and past medical history; baseline information includes: age, gender, height and weight; past medical history includes: at least one of hypertension, diabetes or urolithiasis.

In some embodiments, clinical variables of patients are collected via the CDMS digital medical record system.

In some embodiments, based on a univariate analysis model, the independent risk of kidney stones in clinical variable images is obtained respectively, and then based on a multivariate analysis model, the cross-influencing factors of the independent risks are controlled to obtain independent risk factors, including: inputting the clinical variables into the univariate analysis model in sequence, outputting the first significant difference index corresponding to the univariate image of kidney stones respectively, inputting the corresponding clinical variables of the first significant difference index that meets the preset index threshold into the multivariate analysis model, controlling the cross-influencing factors between the single factors, outputting the second significant difference index, and selecting the independent risk factors from the clinical variables according to the second significant difference index.

It is worth noting that the first significant difference index is the significance difference index output when all clinical variables are used as the input of the univariate analysis model when the clinical variables are subjected to univariate analysis according to the univariate analysis model, and the corresponding clinical variables of the first significant difference index that meet the preset index threshold are selected. In order to analyze the influence between the single factors screened out from all clinical variables, the clinical variables corresponding to the first significant difference index that meets the preset index threshold are used as the input of the multifactor analysis model, and the second significant difference index is output. That is, the second significant difference index is the significance difference index output by the multifactor analysis model for the clinical variables corresponding to the first significant difference index that meets the preset index threshold.

In other words, the single factors that affect kidney stones are screened out from clinical variables through the univariate analysis model, and then the influence between the screened single factors is controlled through the multivariate analysis model to obtain independent risk factors affected by the single factors.

In some embodiments, during training, when the univariate analysis model and the multivariate analysis model use the same linear regression algorithm model, the influence of a single factor in the clinical variables on the risk of kidney stones is determined according to the univariate analysis model based on the linear regression algorithm model. In order to ensure that the influence of the variables is closer to the complex situation in the real world, a multivariate analysis model based on the linear regression algorithm model is further used to control the cross-influence of other factors.

In some embodiments, the first significant difference indicator and the second significant difference indicator are P values. It is worth noting that in statistics, the P value is the probability of obtaining a test statistic when the null hypothesis is true.

In some embodiments, see Table 3, which is an analysis result table of the univariate analysis model and the multivariate analysis model based on the logistic regression algorithm model provided in this embodiment. In Table 3, the analysis of the univariate analysis model shows that the P value of gender and body mass index (Body Mass Index, BMI) is less than 0.05, which is statistically significant, then the gender and BMI values are used as inputs of the multivariate analysis model, the P values of gender and BMI values are output, and then the clinical variables whose P values meet the multivariate preset thresholds are selected as the final independent risk factors.

In some embodiments, after all clinical variables are input to the univariate analysis model and the P value is output, all clinical variables are used as inputs to the multivariate analysis model, the multivariate analysis model outputs the P values of all clinical variables, the clinical variables corresponding to the P values of the univariate analysis model that meet the preset threshold are selected, and the clinical variables corresponding to the P values of the multivariate analysis model that meet the multifactor preset threshold are selected, and the intersection of the two clinical variables is taken as the final independent risk factor. In Table 3, after the univariate analysis model outputs the P value for all clinical variables, the multivariate analysis model inputs all clinical variables and outputs the P value of all clinical variables. The analysis results show that among the P values output by the univariate analysis model, the P values of gender and BMI are less than 0.05 and have statistical differences, and among the P values output by the multivariate analysis model, the P values of gender and BMI are also less than 0.05 and have statistical differences, that is, the P values output by the multivariate analysis model and the univariate analysis model respectively meet the intersection of the corresponding clinical variables of the preset threshold and the multifactor preset threshold, only gender and BMI, therefore, the independent risk factors that can be screened out include only: gender and BMI.

Table 3 Analysis results of the single factor analysis model and multi-factor analysis model based on the regression algorithm model

Step S13: Combine the target radiomics features with the independent risk factors to obtain a combined feature, and obtain a classification result of the risk prediction corresponding to the combined feature based on the prediction model.

It is worth noting that the classification result is the classification result output by the corresponding prediction model based on the combined features. In some embodiments, the imaging omics model based on the logistic regression algorithm model inputs the combined features and outputs the classification result.

In some embodiments, based on the prediction model, the classification result of the risk prediction corresponding to the combined feature is obtained, including: inputting the combined feature into an imaging omics model based on the prediction model, and outputting the classification result of the risk prediction corresponding to the combined feature; the imaging omics model is obtained after training based on the input target imaging omics feature and the intercept corresponding to the target imaging omics feature.

In some embodiments, patients are divided into training data and validation data in a ratio of 7:3, and the target imaging features of the corresponding patients are assigned to the training set and the validation set.

In some embodiments, the imaging omics model is obtained after training based on the input target imaging omics features and the intercept corresponding to the target imaging omics features, including: obtaining a training set and a validation set of the target imaging omics features and the intercept corresponding to the target imaging omics features, training and validating multiple different machine learning algorithm models according to the training set and the validation set, and selecting the algorithm model with the smallest difference in AUC (Area under the curve, area under the ROC curve) indicator between the training result and the validation result as the imaging omics model.

In some embodiments, multiple different machine learning algorithm models include: a logistic regression algorithm model, an Adaboost algorithm model, a decision tree algorithm model, and a support vector machine algorithm model.

In some embodiments, the logistic regression algorithm model, Adaboost algorithm model, decision tree algorithm model and support vector machine algorithm model are trained and verified. See Figure 3, which is a schematic diagram of the AUC of the training results and verification results of different machine learning algorithm models provided in this embodiment. In Figure 3 (a), the AUC of the training set of the prediction model based on logistic regression is 0.858 (95% CI 0.819-0.897), that is: the area under the curve is estimated to be 0.858, and there is a 95% confidence that the true AUC value is between 0.819 and 0.897; the AUC of the verification set is 0.806 (95% CI 0.734-0.877).

In Figure 3, the AUC values of the decision tree algorithm model (Figure 3b), the support vector machine algorithm model (Figure 3c), and the Adaboost algorithm model (Figure 3d) in the training set were 0.873 (95% CI

The AUC values in the validation set were 0.770 (95% CI 0.696-0.845), 0.834 (95% CI 0.772-0.896), and 0.654 (95% CI 0.578 -0.730), respectively.

Among the four algorithms selected, the Adaboost algorithm showed the best prediction performance, with an AUC value of 0.933 in the training set and 0.834 in the validation set. However, there was a large difference in its prediction performance between the training set and the validation set, and there was a possibility of overfitting. Therefore, in order to ensure the stability and generalization ability of the imaging omics model, the final prediction model was selected based on the AUC value and the AUC difference between the training set and the validation set. The differences in the AUC of the four algorithms between the training set and the validation set were: logistic regression (0.052), decision tree (0.103), support vector machine (0.099) and Adaboost (0.145), so logistic regression was selected to build the imaging omics model.

In some embodiments, a prediction model based on a logistic regression algorithm model is established for independent risk factors, and compared with a prediction model based on a logistic regression algorithm model established for target imaging genomics features. See Figure 4, which is a schematic diagram of the AUC of the prediction model of independent risk factors and target imaging genomics features provided in this embodiment. The results show that Figure 4 (a) is a schematic diagram of the comparison of the AUC index of the prediction model based on the training set, in which the AUC value of the prediction model of the independent risk factor is 0.617. Figure 4 (a) is a schematic diagram of the comparison of the AUC index of the prediction model based on the validation set, in which the AUC value of the prediction model of the independent risk factor is 0.619, and its AUC index is much lower than the prediction model of the imaging genomics feature (the AUC value in the training set is 0.858, and the AUC value in the validation set is 0.806), thereby further showing the superiority of the imaging genomics prediction model.

In some embodiments, the independent risk factors are combined with the target radiomics features to obtain the combined features, and the radiomics model based on the logistic regression algorithm model is constructed for the combined features to improve the predictive performance of the model. The results show that after combining the independent risk factors, the predictive performance of the radiomics model is slightly improved. It is worth noting that the radiomics features have a large weight. The independent risk factors are determined by univariate and multivariate analysis models, and the independent risk factors are combined with the radiomics features to construct the radiomics model. The radiomics model outputs the classification results of the risk prediction of kidney stones. Since the classification results are either at risk of disease or without risk of disease, the interpretability is poor. Therefore, a nomogram is drawn to visualize the classification results, that is, the input independent risk factors and the target radiomics features in the radiomics model are put into a graph, and the scores are assigned according to the coefficients of the independent risk factors and the target radiomics features, and then the sum of the scores of the independent risk factors and the target radiomics features is calculated, thereby obtaining the probability of suffering from the disease. Thereby, the prediction results of the model can be understood without reading statistical tables or equations, which simplifies the use of the model and is more intuitive.

In some embodiments, to further verify the predictive performance and practical value of the model, the radiomics model is evaluated by a calibration curve and a clinical decision curve.

In some embodiments, referring to FIG5 , it is a schematic diagram of a calibration curve and a clinical decision curve based on an imaging omics model provided in the present embodiment. In the calibration curve, the secondary diagonal corresponding to the 45° dashed line represents the probability under ideal conditions. The closer the prediction curve is to the ideal curve, the better the prediction performance. The calibration curve shows that the prediction performance of the imaging omics model is close to the probability under ideal conditions, and the prediction performance is good. In the clinical decision curve, none means no intervention at all, the gray curve means all interventions, and the area under the curve represents the net benefit. The clinical decision curve shows that when the threshold probability is 0.1-1.0, the net benefit obtained using the imaging omics model is higher than all interventions and no interventions, indicating that the constructed imaging omics model has clinical application value.

Step S14: construct a nomogram based on the combined features, and perform probability visualization of the classification results based on the nomogram to obtain the risk probability, and interpret the classification results based on the risk description corresponding to the risk probability.

In some embodiments, a nomogram is constructed according to the combined features, and the classification results are probabilistically visualized according to the nomogram to obtain the risk probability, including: obtaining the scores of the target radiomics features and the independent risk factors in the nomogram respectively, obtaining the target radiomics feature scores and the independent risk factor scores accordingly, obtaining the total score of the target radiomics feature scores and the independent risk factor scores, and obtaining the corresponding risk probability according to the total score. See Figure 6, which is a schematic diagram of the nomogram I provided in this embodiment.

In some embodiments, obtaining the target radiomics feature score includes: weighting the target radiomics feature according to a coefficient of the target radiomics feature to obtain the target radiomics feature score.

In some embodiments, the target radiomics feature score is expressed as:

Radiomics score＝a1*F1+a2*F2+...+ai*Fi+b;

Among them, a1 to ai are the coefficients of the target imaging features, F1 to Fi are the target imaging features; the number of target imaging features is i; and b is the intercept.

Exemplarily, see Figure 7, which is a schematic diagram of the nomogram II provided in this embodiment. For a female patient with a height of 168 cm and a weight of 60 kg, first of all, the gender is female, and the score in the nomogram is point A corresponding to gender 0, and the BMI value is calculated to be 21.25, which is between 18.5 and 23.9. The BMI score of 21.25 in the nomogram is 2, and the score in the nomogram is point B corresponding to BMI variable 2. The target imaging genomics features are extracted, and the target imaging genomics parameter calculation formula is substituted by the corresponding coefficients in Table 3 to calculate the target imaging genomics score. Assuming that the target imaging genomics score of -2 points corresponds to point C, the final total score is A+B+C≈4.02 points, so that the probability of suffering from kidney stones can be obtained <0.1, which is described as a low-risk population.

In some embodiments, referring to FIG8 , there is a flowchart of another method for predicting kidney stone risk based on CT radiomics provided in this embodiment. In FIG8 , ROI is first delineated based on the CT image to obtain the ROI result, and the target radiomics feature and the intercept corresponding to the target radiomics feature are extracted on the ROI result. The patients are divided into a training set and a validation set according to a ratio of 7:3, and the target radiomics feature and the intercept corresponding to the target radiomics feature of the corresponding patient are divided into the training set and the validation set, and the optimal radiomics model is screened out according to the training set and the validation set; at the same time, the clinical variables of the patients are collected, and after extracting independent risk factors from the clinical variables, the radiomics model is constructed according to the target radiomics feature and the independent risk factors; and based on the radiomics model, the classification result of kidney stone risk prediction is obtained; finally, a nomogram is drawn to obtain an explanatory description of the classification result, and the classification result and the description of the classification result are used as the final result analysis.

Compared with the prior art, this embodiment extracts the radiomic features of kidney images through high-throughput radiomics, not only limited to the renal papillary CT value, but also extracts the first-order features, shape features and texture features including the renal CT value, converts them into quantitative data, and screens the extracted radiomic features to avoid overfitting of the constructed radiomics model, and compares algorithms such as logistic regression and random forest to build the most stable radiomics model, which is more accurate than current research. Secondly, a nomogram is further drawn on the basis of the constructed radiomics model, which simplifies the use of the prediction model and visualizes the prediction probability, so that the risk of patients suffering from kidney stones can be more intuitively reflected, which is more conducive to communication with patients, thereby improving medical compliance.

Example 2

Referring to FIG. 9 , it is a schematic diagram of the structure of a kidney stone risk prediction system based on CT imaging genomics provided in this embodiment, comprising: a target imaging genomics feature extraction unit 21 , an independent risk factor extraction unit 22 , a classification result acquisition unit 23 and an interpretation unit 24 .

It is worth noting that the target imaging genomics feature extraction unit 21 outputs the target imaging genomics feature through the collected CT images, and transmits the target imaging genomics feature to the classification result acquisition unit 23; the independent risk factor extraction unit 22 outputs the independent risk factor through the collected clinical variables, and transmits the independent risk factor to the classification result acquisition unit 23; after the classification result acquisition unit 23 acquires the target imaging genomics feature and the independent risk factor, it outputs the classification result, and transmits the classification result to the interpretation unit 24; after receiving the classification result, the interpretation unit 24 outputs a description of the classification result.

The target imaging genomics feature extraction unit 21 is used to respectively collect multi-axial CT images of multiple patients, outline the region of interest on the CT images, obtain the region of interest results, and extract the target imaging genomics features from the region of interest results.

In some embodiments, multi-axial CT images of multiple patients are collected separately, and regions of interest are outlined on the CT images to obtain region of interest results, including: collecting multi-axial CT images of renal tissue of multiple patients separately, the CT images including: cross-sectional, coronal and sagittal CT images of renal tissue; simultaneously masking the air and fat tissue surrounding the renal tissue in the cross-sectional, coronal and sagittal CT images by a preset image threshold, and outlining the renal tissue separately to obtain the region of interest.

In some embodiments, target radiomics features are extracted from the results of the region of interest, including: extracting multiple radiomics features of the region of interest, and having two experts extract the same radiomics feature to obtain two groups of extraction results respectively, evaluating the intra-group consistency of the two groups of extraction results to obtain the target radiomics features; the multiple radiomics features include: first-order features, shape features and texture features.

In some embodiments, the intra-group consistency of the two groups of extraction results is evaluated to obtain target radiomics features, including: retaining radiomics features with an intra-group correlation coefficient not less than a preset threshold, obtaining multiple candidate radiomics features, and performing data dimension reduction or parameter compression on the multiple candidate radiomics features to screen out multiple target radiomics features with the highest correlation with kidney stones.

The independent risk factor extraction unit 22 is used to collect clinical variables of multiple patients respectively, obtain the independent risk of kidney stones in the clinical variable images based on the univariate analysis model, and then control the cross-influencing factors of the independent risk based on the multivariate analysis model to obtain the independent risk factors.

In some embodiments, based on a univariate analysis model, the independent risk of kidney stones in clinical variable images is obtained respectively, and then based on a multivariate analysis model, the cross-influencing factors of the independent risks are controlled to obtain independent risk factors, including: inputting the clinical variables into the univariate analysis model in sequence, outputting the first significant difference index corresponding to the univariate image of kidney stones respectively, inputting the corresponding clinical variables of the first significant difference index that meets the preset index threshold into the multivariate analysis model, controlling the cross-influencing factors between the single factors, outputting the second significant difference index, and selecting the independent risk factors from the clinical variables according to the second significant difference index.

The classification result acquisition unit 23 is used to combine the target imaging genomics feature with the independent risk factor to obtain a combined feature, and obtain a classification result of the risk prediction corresponding to the combined feature based on the prediction model.

In some embodiments, based on the prediction model, the classification result of the risk prediction corresponding to the combined feature is obtained, including: inputting the combined feature into an imaging omics model based on the prediction model, and outputting the classification result of the risk prediction corresponding to the combined feature; the imaging omics model is obtained after training based on the input target imaging omics feature and the intercept corresponding to the target imaging omics feature.

In some embodiments, the imaging omics model is obtained after training based on the input target imaging omics features and the intercept corresponding to the target imaging omics features, including: obtaining a training set and a validation set of the target imaging omics features and the intercept corresponding to the target imaging omics features, training and validating multiple different machine learning algorithm models according to the training set and the validation set, and selecting the algorithm model with the smallest difference in AUC indicators between the training results and the validation results as the imaging omics model.

The interpretation unit 24 is used to construct a nomogram according to the combined features, and to perform probability visualization of the classification results according to the nomogram to obtain the risk probability, and to interpret the classification results according to the risk description corresponding to the risk probability.

In some embodiments, a nomogram is constructed based on the combined features, and the classification results are probabilistically visualized based on the nomogram to obtain the risk probability, including: respectively obtaining the scores of the target imaging genomics features and the independent risk factors in the nomogram, correspondingly obtaining the target imaging genomics feature scores and the independent risk factor scores, obtaining the total score of the target imaging genomics feature scores and the independent risk factor scores, and obtaining the corresponding risk probability based on the total score.

In some embodiments, obtaining a target radiomics feature score includes:

The target radiomics feature is weighted according to its coefficient to obtain the target radiomics feature score.

In some embodiments, the target radiomics feature score is expressed as:

Radiomics score＝a1*F1+a2*F2+...+ai*Fi+b;

Among them, a1 to ai are the coefficients of the target imaging features, F1 to Fi are the target imaging features; the number of target imaging features is i; and b is the intercept.

In this embodiment, the target imaging genomics feature extraction unit 21 is used to obtain multi-axial CT images of renal tissue to comprehensively detect renal tissue, thereby avoiding the omission of useful CT image information, thereby improving the accuracy of kidney stone risk prediction; and, the independent risk factor extraction unit 22 uses a univariate analysis model to first obtain the independent risk of kidney stones from clinical variables, and then controls the cross-influencing factors affecting the independent risk based on the multifactor analysis model to obtain more accurate independent risk factors, thereby further improving the accuracy of kidney stone risk prediction; then, the classification result acquisition unit 23 combines the target imaging genomics features and independent risk factors of the CT image to perform risk prediction, and uses the interpretation unit 24 to construct a nomogram and obtain the risk probability, and interprets the classification results according to the risk description corresponding to the risk probability, thereby improving the interpretability of kidney stone risk prediction.

Those skilled in the art will appreciate that the embodiments of the present application may also provide computer program products. Therefore, the present application may adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program codes.

The present application is described with reference to the flowchart and/or block diagram of the method, device (system) and computer program product according to the embodiment of the present application. It should be understood that each process and/or box in the flowchart and/or block diagram, and the combination of the process and/or box in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one process or multiple processes in the flowchart and/or one box or multiple boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims (10)

1. A method for predicting kidney stone risk based on CT radiomics, comprising: Collect multi-axial CT images of multiple patients respectively, delineate regions of interest on the CT images, obtain region of interest results, and extract target imaging omics features from the region of interest results; Clinical variables of multiple patients were collected respectively, and the independent risk of kidney stones in clinical variable images was obtained based on the univariate analysis model. Then, the cross-influencing factors of the independent risks were controlled based on the multivariate analysis model to obtain independent risk factors. Combining the target radiomics feature with the independent risk factor to obtain a combined feature, and obtaining a classification result of the risk prediction corresponding to the combined feature based on a prediction model; A nomogram is constructed according to the combined features, and the classification results are probability visualized according to the nomogram to obtain the risk probability, and the classification results are explained according to the risk description corresponding to the risk probability. 2. The method for predicting kidney stone risk based on CT radiomics according to claim 1, characterized in that the independent risk of kidney stones of clinical variable images is obtained based on the univariate analysis model, and then the cross-influencing factors of the independent risk are controlled based on the multivariate analysis model to obtain independent risk factors, including: The clinical variables are input into the univariate analysis model in sequence, and the first significant difference index of kidney stones corresponding to the univariate images is output respectively. The corresponding clinical variables of the first significant difference index that meets the preset index threshold are input into the multivariate analysis model, the cross-influencing factors between the single factors are controlled, and the second significant difference index is output. According to the second significant difference index, independent risk factors are selected from the clinical variables. 3. The method for predicting kidney stone risk based on CT radiomics according to claim 1, wherein the step of constructing a nomogram according to the combined features and performing probability visualization of the classification results according to the nomogram to obtain the risk probability comprises: The scores of the target radiomics feature and the independent risk factor in the nomogram are obtained respectively, the target radiomics feature score and the independent risk factor score are obtained correspondingly, the total score of the target radiomics feature score and the independent risk factor score is obtained, and the corresponding risk probability is obtained according to the total score. 4. The method for predicting kidney stone risk based on CT radiomics according to claim 3, wherein obtaining the target radiomics feature score comprises: According to the coefficient of the target radiomics feature, weighting the target radiomics feature to obtain a target radiomics feature score; Wherein, the target radiomics feature score is expressed as: Radiomics SCorea1*F1+a2*F2+...+ai*Fi+b; Among them, a1 to ai are the coefficients of the target imaging features, F1 to Fi are the target imaging features; the number of target imaging features is i; and b is the intercept. 5. The method for predicting kidney stone risk based on CT radiomics according to claim 1, wherein obtaining the classification result of the risk prediction corresponding to the combined feature based on the prediction model comprises: The combined feature is input into an imaging omics model based on the prediction model, and a classification result of risk prediction corresponding to the combined feature is output; the imaging omics model is obtained after training based on the input of the target imaging omics feature and the intercept corresponding to the target imaging omics feature. 6. The method for predicting kidney stone risk based on CT radiomics according to claim 5, wherein the radiomics model is obtained by training according to the input of the target radiomics feature and the intercept corresponding to the target radiomics feature, and comprises: A training set and a validation set of the target imaging features and the intercept corresponding to the target imaging features are obtained, and a plurality of different machine learning algorithm models are trained and validated according to the training set and the validation set, and the algorithm model with the smallest AUC index difference between the training result and the validation result is selected as the imaging model. 7. The method for predicting kidney stone risk based on CT radiomics according to claim 1, wherein extracting target radiomics features from the result of the region of interest comprises: Extract multiple radiomics features of the region of interest, and have two experts extract the same radiomics feature to obtain two groups of extraction results, respectively. Evaluate the intra-group consistency of the two groups of extraction results to obtain the target radiomics feature; the multiple radiomics features include: first-order features, shape features and texture features. 8. The method for predicting kidney stone risk based on CT radiomics according to claim 7, characterized in that the step of evaluating the intra-group consistency of the two groups of extraction results to obtain the target radiomics features comprises: The imaging features with intra-group correlation coefficients not less than a preset threshold are retained to obtain multiple candidate imaging features, and data dimension reduction or parameter compression is performed on the multiple candidate imaging features to screen out multiple target imaging features with the highest correlation with kidney stones. 9. The method for predicting kidney stone risk based on CT radiomics according to claim 1, wherein the step of collecting multi-axial CT images of multiple patients, outlining regions of interest on the CT images, and obtaining region of interest results comprises: Acquiring multi-axial CT images of renal tissues of multiple patients respectively, wherein the CT images include: cross-sectional, coronal and sagittal CT images of renal tissues; The air and fat tissue surrounding the kidney tissue in the CT images of the cross section, the coronal plane and the sagittal plane are simultaneously shielded by a preset image threshold, and the kidney tissue is outlined respectively to obtain a region of interest. 10. A kidney stone risk prediction system based on CT radiomics, characterized by comprising: a target radiomics feature extraction unit, an independent risk factor extraction unit, a classification result acquisition unit and an interpretation unit; wherein, The target imaging genomics feature extraction unit is used to respectively collect multi-axial CT images of multiple patients, outline the region of interest on the CT images, obtain the region of interest results, and extract the target imaging genomics features from the region of interest results; The independent risk factor extraction unit is used to collect clinical variables of multiple patients respectively, obtain independent risks of kidney stones in clinical variable images based on a univariate analysis model, and then control the cross-influencing factors of the independent risks based on a multivariate analysis model to obtain independent risk factors; The classification result acquisition unit is used to combine the target imaging genomics feature with the independent risk factor to obtain a combined feature, and obtain a classification result of the risk prediction corresponding to the combined feature based on a prediction model; The explanation unit is used to construct a nomogram according to the combined features, and to perform probability visualization of the classification results according to the nomogram to obtain the risk probability, and to explain the classification results according to the risk description corresponding to the risk probability.