Method, device, equipment and medium for predicting expression level of kidney cancer HIF-2alpha

Abstract

The invention provides a method, a device, equipment and a medium for predicting the expression level of renal cancer HIF-2alpha. The method comprises the following steps: S10, acquiring clinical pathological data and an abdominal enhanced arterial phase image of a patient with renal clear cell carcinoma; s20, sketching a three-dimensional region of interest of the tumor through the abdominal enhanced arterial phase image; s30, extracting image radiomics characteristics from the tumor three-dimensional region of interest, screening the image radiomics characteristics through Z-score standardization processing, stability test, an mRMR algorithm and an LASSO regression algorithm in sequence to obtain screened characteristics, and calculating to obtain a radiomics score through the screened characteristics; s40, screening out independent clinical predictive factors related to HIF-2alpha expression from the clinical pathology data, wherein the independent clinical predictive factors comprise WHO / ISUP classification and clinical T classification; and inputting the independent clinical prediction factor and the radiomics score into a joint prediction model to obtain a prediction score, and evaluating the probability of HIF-2alpha expression by using the prediction score.

Description

Technical Field

[0001] This invention relates to the field of healthcare informatics technology, and in particular to a method, apparatus, device, and medium for predicting the expression level of HIF-2α in renal cell carcinoma. Background Technology

[0002] Renal cell carcinoma (ccRCC) ranks third in incidence globally among malignant tumors of the urinary system, and its incidence continues to rise. Clinical studies have shown that ccRCC has a highly aggressive biological profile. Approximately 30% of patients already have distant organ metastases at initial diagnosis, and among patients undergoing radical nephrectomy, 40% are at risk of local recurrence. Notably, due to the lack of specific clinical manifestations in early-stage ccRCC, about 60% of patients are diagnosed at an intermediate or advanced stage, posing a challenge to early diagnosis and treatment. Therefore, a systematic analysis of the disease characteristics of ccRCC not only helps in establishing clinical risk prediction models but also provides evidence-based support for clinicians to develop personalized treatment plans and optimize treatment decisions.

[0003] Despite advancements in systemic therapy for advanced renal cell carcinoma, such as anti-angiogenic targeted drugs and immune checkpoint inhibitors (ICIs), tumor metastasis and drug resistance still result in low long-term response rates and limited survival benefits. Therefore, developing novel strategies to predict and improve efficacy is crucial. In this context, hypoxia-inducible factor-2α (HIF-2α) has been identified as a core driver in clear cell renal cell carcinoma (ccRCC). It is continuously activated in the hypoxic tumor microenvironment, driving tumor angiogenesis, proliferation, metastasis, and metabolic reprogramming by upregulating key factors such as VEGF and PDGF, and is significantly associated with poor prognosis. Clinical studies have confirmed that high expression of HIF-2α is associated with resistance to VEGF-TKI drugs and can influence the efficacy of ICIs by remodeling the immunosuppressive microenvironment.

[0004] However, existing methods for assessing HIF-2α expression have significant limitations: current detection methods that rely on tissue biopsy are invasive and cannot be used for preoperative assessment; and due to the high heterogeneity of tumors, small-sample biopsies may not reflect the full picture of the tumor, leading to misdiagnosis; in addition, this method is also difficult to dynamically monitor the expression level. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, equipment and medium for predicting the expression level of HIF-2α in renal cell carcinoma, so as to achieve non-invasive prediction of the expression level of HIF-2α in ccRCC.

[0006] In a first aspect, the present invention provides a method for predicting the expression level of HIF-2α in renal cell carcinoma, comprising the following steps: S10. Obtain clinicopathological data and enhanced arterial phase images of the abdomen from patients with clear cell renal cell carcinoma; S20. Delineate the three-dimensional region of interest of the tumor using the enhanced abdominal arterial phase image; S30. Extract image radiomics features from the three-dimensional region of interest of the tumor, and sequentially filter the image radiomics features through Z-score normalization, stability test, mRMR algorithm and LASSO regression algorithm to obtain the filtered features. Calculate the radiomics score based on the filtered features. S40. Screen out independent clinical predictors related to HIF-2α expression from the clinicopathological data. The independent clinical predictors include WHO / ISUP classification and clinical T stage. The independent clinical predictor and the radiomics score are input into the joint prediction model to obtain a prediction score, and the probability of HIF-2α expression is assessed using the prediction score.

[0007] Secondly, the present invention provides a device for predicting the expression level of HIF-2α in renal cell carcinoma, characterized in that it comprises the following modules: The data acquisition module acquires clinicopathological data and enhanced arterial phase images of the abdomen from patients with clear cell renal cell carcinoma. The image delineation module delineates the three-dimensional region of interest of the tumor using the enhanced arterial phase image of the abdomen; The feature selection module extracts image radiomics features from the three-dimensional region of interest of the tumor, and sequentially filters the image radiomics features through Z-score normalization, stability test, mRMR algorithm and LASSO regression algorithm to obtain the filtered features. The radiomics score is calculated based on the filtered features. The model evaluation module screens independent clinical predictors associated with HIF-2α expression from the clinicopathological data. These independent clinical predictors include WHO / ISUP classification and clinical T stage. The independent clinical predictors and the radiomics score are input into a joint prediction model to obtain a prediction score. The prediction score is then used to evaluate the probability of HIF-2α expression.

[0008] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect.

[0009] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0010] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: A nomogram is a visualization tool built using statistical prediction models that integrates and displays the combined effects of multiple factors on a specific outcome variable. Combining high-dimensional features generated by radiomics with nomogram technology can accurately predict various clinical outcomes, including disease diagnosis, pathological classification, lymph node metastasis, tumor recurrence, prognostic assessment, and tumor gene expression status. Its intuitiveness and good interpretability make it an efficient and convenient decision support tool in clinical practice.

[0011] This invention innovatively integrates radiomics features and clinical indicators to successfully construct a nomogram model for predicting HIF-2α expression levels in ccRCC. HIF-2α, a key driver of ccRCC, directly impacts the efficacy of inhibitor-targeted therapy. This model quantifies the contribution of each indicator through weighted analysis, assisting clinicians in screening potential beneficiaries of HIF-2α inhibitors, significantly reducing gene testing costs while achieving non-invasive assessment. This prediction system based on multidimensional data fusion provides a theoretical basis for personalized treatment planning and has significant clinical translational implications for optimizing medical resource allocation and improving diagnostic and treatment efficiency.

[0012] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0014] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the device in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the clinical application of the nomogram in Embodiment 5 of the present invention. Detailed Implementation

[0015] This application provides a method, apparatus, device, and medium for predicting the HIF-2α expression level in renal cell carcinoma (ccRCC), thereby achieving non-invasive prediction of HIF-2α expression level in ccRCC. Example 1

[0016] This embodiment provides a method for predicting the expression level of HIF-2α in renal cell carcinoma, such as... Figure 1 As shown, it includes the following steps: S10. Obtain clinicopathological data and enhanced arterial phase images of the abdomen from patients with clear cell renal cell carcinoma; S20. Delineate the three-dimensional region of interest of the tumor using the enhanced abdominal arterial phase image; Specifically, S20 is: S21. In the enhanced arterial phase image of the abdomen, delineate the three-dimensional region of interest of the tumor along the edge of the tumor; first, set the window width and window level of all image data to the abdominal window (window width 600HU, window level 40HU), and manually segment the lesion ROI layer by layer along the edge of the tumor in the arterial phase image, focusing attention on the tumor parenchyma, avoiding delineating the necrotic part of the tumor, enhancing blood vessels and other organs and tissues at the edge, and eliminating the influence of non-primary lesion tissue image features on this experiment; S22. Save the delineated 3D region of interest file of the tumor in NIFTI format; S30. Extract image radiomics features from the three-dimensional region of interest of the tumor, and sequentially filter the image radiomics features through Z-score normalization, stability test, mRMR algorithm and LASSO regression algorithm to obtain the filtered features. Calculate the radiomics score based on the filtered features. Specifically, S30 is: S31. Use the open-source software package PyRadiomics to extract radiomics features of the three-dimensional region of interest of the tumor; the radiomics features include several radiomics sub-features, and the types of radiomics sub-features include: geometric features, texture features and histogram features; The geometric shape features are used to describe the three-dimensional shape of the tumor, the texture features are used to describe the second-order and higher-order spatial distribution of intensity, and the histogram features are used to describe the first-order statistical distribution of voxel intensity within the tumor. S32. Perform Z-score standardization on the extracted radiomics features so that the mean value of each radiomics sub-feature is 0 and the standard deviation is 1. S33. Stability test is performed using Spearman correlation coefficient analysis to obtain the Spearman correlation coefficient of each radiomic sub-feature. The radiomic sub-features whose Spearman correlation coefficients are within the set threshold are retained to obtain stable radiomic features. S34. The stable radiomics features are filtered out by using the mRMR algorithm to remove the radiomics sub-features within the stable radiomics features that meet the set conditions, and the remaining radiomics features are obtained. S35. Combine cross-validation (to prevent model overfitting, a 5-fold cross-validation method is used for data validation. By adjusting multiple penalty coefficients, the omics features with non-zero coefficients are retained. These features are identified as the optimal features by the algorithm); the remaining radiomics features after cross-validation are screened using the LASSO regression algorithm to obtain radiomics features with non-zero coefficients as the screened features, and the corresponding coefficients of the screened features are obtained. S36. The radiomics score is calculated using the selected features and their corresponding coefficients; the calculation formula is the sum of the products of each selected feature and its corresponding coefficient. S40. Screen out independent clinical predictors related to HIF-2α expression from the clinicopathological data. The independent clinical predictors include WHO / ISUP classification and clinical T stage. The independent clinical predictor and the radiomics score are input into the joint prediction model to obtain a prediction score, and the probability of HIF-2α expression is evaluated using the prediction score. Specifically, S40 is: S41. The clinicopathological data were analyzed using univariate and multivariate logistic regression algorithms to screen out independent clinical predictors associated with HIF-2α expression; the screened independent clinical predictors included WHO / ISUP classification and clinical T stage. S42. A joint prediction model is established using the random forest algorithm, and the results of the joint prediction model are presented in the form of a nomogram. S43. Input the radiomics score and independent clinical predictors as prediction parameters into the joint prediction model; The predicted scores corresponding to each prediction parameter are obtained by using the Logistic regression algorithm; The nomogram of the joint prediction model displays the prediction scores corresponding to each prediction parameter; the sum of the prediction scores yields the probability of HIF-2α expression.

[0017] like Figure 3 As shown, a patient has a WHO / ISUP grade of 3 (31.7 points), a clinical T stage of 3 (24 points), and a radiomics score (Rad-score) of 0.75 (75 points). The total score is 130.6 points, which corresponds to an approximately 87% probability of high HIF-2α expression.

[0018] Based on the same inventive concept, this application also provides an apparatus corresponding to the method in Embodiment 1, as detailed in Embodiment 2. Example 2

[0019] This embodiment provides a device, such as Figure 2 As shown, it includes the following modules: The data acquisition module acquires clinicopathological data and enhanced arterial phase images of the abdomen from patients with clear cell renal cell carcinoma. The image delineation module delineates the three-dimensional region of interest of the tumor using the enhanced arterial phase image of the abdomen; The image delineation module is specifically as follows: The delineation unit delineates the three-dimensional region of interest of the tumor along the tumor margin in the enhanced arterial phase image of the abdomen. The storage unit saves the outlined three-dimensional region of interest (ROI) file of the tumor in NIFTI format. The feature selection module extracts image radiomics features from the three-dimensional region of interest of the tumor, and sequentially filters the image radiomics features through Z-score normalization, stability test, mRMR algorithm and LASSO regression algorithm to obtain the filtered features. The radiomics score is calculated based on the filtered features. The feature filtering module specifically includes: The feature extraction unit uses the open-source software package PyRadiomics to extract radiomics features of the three-dimensional region of interest of the tumor; the radiomics features include several radiomics sub-features, and the types of radiomics sub-features include: geometric features, texture features, and histogram features; The geometric shape features are used to describe the three-dimensional shape of the tumor, the texture features are used to describe the second-order and higher-order spatial distribution of intensity, and the histogram features are used to describe the first-order statistical distribution of voxel intensity within the tumor. The standardization processing unit performs Z-score standardization on the extracted radiomics features, so that the mean value of each radiomics sub-feature is 0 and the standard deviation is 1. The stability test unit uses Spearman correlation coefficient analysis to perform stability tests, obtains the Spearman correlation coefficient of each radiomic sub-feature, retains the radiomic sub-features whose Spearman correlation coefficients are within a set threshold, and obtains stable radiomic features. The mRMR filtering unit filters the stable radiomics features by using the mRMR algorithm to filter out the radiomics sub-features within the stable radiomics features that meet the set conditions, thus obtaining the remaining radiomics features. The LASSO regression algorithm unit, combined with cross-validation, uses the LASSO regression algorithm to filter the remaining radiomics features, obtaining radiomics features with non-zero coefficients as the filtered features, and obtaining the corresponding coefficients of the filtered features. The scoring calculation unit calculates the radiomics score based on the filtered features and their corresponding coefficients; the calculation formula is the sum of the products of each filtered feature and its corresponding coefficient. The model evaluation module screens independent clinical predictors associated with HIF-2α expression from the clinicopathological data. The independent clinical predictors include WHO / ISUP classification and clinical T stage. The independent clinical predictors and the radiomics score are input into the joint prediction model to obtain a prediction score. The prediction score is then used to evaluate the probability of HIF-2α expression. The model evaluation module specifically includes: The Independent Clinical Predictor Unit uses univariate and multivariate logistic regression algorithms to analyze the clinicopathological data and screen out independent clinical predictors associated with HIF-2α expression; the screened independent clinical predictors include WHO / ISUP classification and clinical T stage; The model building unit establishes a joint prediction model using the random forest algorithm, and the results of the joint prediction model are presented in the form of a no-line graph. The probability prediction unit inputs the radiomics score and independent clinical predictors as prediction parameters into the joint prediction model; The predicted scores corresponding to each prediction parameter are obtained by using the Logistic regression algorithm; The nomogram of the joint prediction model displays the prediction scores corresponding to each prediction parameter; the prediction scores are summed to obtain the probability of HIF-2α expression.

[0020] Since the apparatus described in Embodiment 2 of the present invention is an apparatus used to implement the method of Embodiment 1 of the present invention, those skilled in the art can understand the specific structure and variations of the apparatus based on the method described in Embodiment 1 of the present invention, and therefore will not be described again here. All apparatuses used in the method of Embodiment 1 of the present invention fall within the scope of protection of the present invention.

[0021] Based on the same inventive concept, this application provides an electronic device embodiment corresponding to Embodiment 1, as detailed in Embodiment 3. Example 3

[0022] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement any of the implementation methods in Embodiment 1.

[0023] Since the electronic device described in this embodiment is the device used to implement the method in Embodiment 1 of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in Embodiment 1 of this application. Therefore, how the electronic device implements the method in the embodiment of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiment of this application falls within the scope of protection of this application.

[0024] Based on the same inventive concept, this application provides a storage medium corresponding to Embodiment 1, as detailed in Embodiment 4. Example 4

[0025] This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it can implement any of the implementation methods in Embodiment 1. Example 5

[0026] I. Materials and Methods 1. Research Subjects A retrospective study included 274 patients who presented with ccRCC and were pathologically diagnosed postoperatively. They were randomly assigned to the training set (n=191) and the validation set (n=83) in a 7:3 ratio.

[0027] 1.1 Inclusion Criteria (1) Postoperative pathological diagnosis confirmed ccRCC; (2) Clinical data and postoperative pathology reports are kept intact; (3) Complete a plain CT scan and enhanced CT scan of the kidney within 2 weeks before the operation; (4) The postoperative pathological specimens are well preserved and can be subjected to HIF-2α immunohistochemical detection.

[0028] 1.2 Exclusion Criteria (1) Missing clinical or pathological data; (2) Motion artifacts or insufficient resolution in CT images affect tumor segmentation; (3) Had received immunotherapy or targeted therapy before surgery; (4) Combination with other malignant tumors or other pathological types of kidney malignant tumors; (5) Severe organ dysfunction, such as severe abnormalities in the heart, lungs, liver, and kidneys.

[0029] 2. Clinical and pathological data Patients were randomly assigned to a training set (n=191) and a validation set (n=83) at a ratio of 7:3. Clinical pathological data included gender, age, BMI, past medical history, maximum tumor diameter, tumor location, type of surgery, intraoperative bleeding, WHO / ISUP grade, sarcomatoid degeneration, tumor necrosis, tumor thrombus formation, fat invasion, TNM stage, etc.

[0030] 3. CT Examination Equipment and Methods All patients' abdominal arterial phase images were acquired using a Siemens Sensation 64-slice CT scanner. Patients drank 800-1000 ml of water 10-15 minutes before the examination and fasted for 4 hours prior to the procedure. All patients underwent scanning in a supine position with deep inspiration and breath-holding. Scanning parameters: intelligent milliampere-second (100-350 ms), tube voltage 120 kV, tube current 250-400 mA, matrix 512×512, reconstruction slice thickness 5 mm. After the plain CT scan, iohexol contrast agent (containing 300 mg / ml iodine, Yangtze River Pharmaceutical Group, China) was injected via the median cubital vein at a rate of 3 ml / s, administered at a dose of 1.0-1.5 ml / kg body weight. Venous phase images were acquired 25 seconds after the arterial phase images. The acquired CT images were uploaded to the PACS system.

[0031] 4. Immunohistochemical staining Experimental samples were selected from specimens pathologically confirmed as clear cell renal cell carcinoma. After fixation in 10% neutral buffered formalin for 24 hours, they were routinely embedded in paraffin. 4 μm serial sections were prepared using a rotary microtome. Positively charged slides were attached and baked in a 60°C oven for 2 hours to enhance adhesion. The dewaxing process involved sequential treatment with xylene I and II for 10 minutes each, followed by dehydration with a gradient of ethanol (100%, 95%, 85%) for 5 minutes each, and then complete hydration with PBS buffer (pH 7.4) before use. Antigen retrieval was performed using a high-temperature, high-pressure (HTHP) method. Tissue sections were immersed in preheated EDTA antigen retrieval solution (pH 9.0) and autoclaved for 2 minutes, followed by natural cooling to room temperature to fully expose the HIF-2α antigen epitopes. To eliminate endogenous peroxidase interference, sections were blocked in 3% H2O2 solution at room temperature for 10 minutes, followed by rinsing three times with PBS buffer. The primary antibody used was a rabbit-derived anti-human HIF-2α monoclonal antibody (catalog number 26422-1-AP, Proteintech Group). The optimal working concentration was determined through preliminary experiments to be 1:200 (diluted in PBS containing 1% BSA). After incubation at 4°C in a humidified chamber for 16 hours, the antibody was warmed to room temperature. The secondary antibody was Dako REAL EnVision-labeled goat anti-rabbit IgG polymer (Dako). After incubation at room temperature for 30 minutes, a DAB chromogenic kit was used for the colorimetric reaction, with the colorimetric time strictly controlled between 1 and 3 minutes. Hematoxylin counterstaining and neutral resin mounting were then performed. Results were interpreted using a double-blind method: two senior pathologists independently observed the tissue. In renal cell carcinoma, HIF-2α-positive cells appeared as scattered brownish-yellow granules. Positive staining was mainly located in the tumor cell nuclei, with some perinuclear cytoplasmic staining, most prominent at the tumor infiltration margin and near neovascularization. The positive criterion was the presence of brownish-yellow granules in the tumor cell nuclei. According to international consensus standards, the grading is as follows: negative (-) indicates a positive cell proportion of <5%; low expression (+) indicates a positive rate of ≥50% with weak staining intensity; high expression (++) indicates a positive rate of ≥50% with strong dark brown staining. Samples with inconsistent interpretations were re-examined and confirmed by a third senior pathology expert. Finally, the samples were divided into a high expression group (++) and a low expression group (- / +) for subsequent statistical analysis.

[0032] 5. Image segmentation and feature extraction 5.1 Delineating the ROI For patients included in this invention, enhanced CT arterial phase images (5mm slice thickness) were exported in DICOM format. After standardization and resampling, they were loaded into the open-source image processing platform ITK-SNAP 3.8.0 (http: / / www.itksnap.org / ). The delineation process was jointly completed by two radiologists with 8 and 12 years of experience, respectively. When encountering controversial areas, the delineation strategy was determined by consensus among senior physicians. When delineating the three-dimensional region of interest (ROI) of the tumor, the window width and window level of all image data were preset to the abdominal window (600HU window width, 40HU window level). In the arterial phase image, the lesion ROI was manually segmented layer by layer along the tumor edge, focusing attention on the tumor parenchyma and avoiding delineating necrotic parts of the tumor, enhanced blood vessels, and other peripheral organs and tissues to eliminate the influence of non-primary lesion tissue image features on this experiment. Finally, the file was saved in NIFTI format to preserve the voxel information of the image to the maximum extent.

[0033] 5.2 Extracting Image Radiomics Features Radiomics features were extracted from each Region of Interest (ROI) using the open-source software package PyRadiomics. The features were divided into three groups: geometric features, texture features, and histogram features. Geometric features described the three-dimensional shape of the tumor; texture features described the second- and higher-order spatial distributions of intensity; and histogram features described the first-order statistical distribution of voxel intensity within the tumor. This invention employs multiple methods to extract texture features, including gray-level co-occurrence matrix (GLCM), gray-level dependence matrix (GLDM), gray-level run length matrix (GLRLM), gray-level sizezone matrix (GLSZM), and Neighbouring Gray Tone Difference Matrix (NGTDM), for comprehensive analysis. In the experiments of this invention, a total of 1815 radiomics features were extracted from the arterial phase ROI. Before further feature screening, the extracted features were Z-score standardized to ensure that the mean of each feature was 0 and the standard deviation was 1, thereby eliminating the influence of different dimensions.

[0034] 6. Radiomics Feature Screening and Model Building Because many of the extracted radiomics features contain no diagnostic value or are redundant, potentially leading to model overfitting, screening and dimensionality reduction are necessary before model building. First, Spearman correlation coefficients are used to eliminate highly correlated similar features; features with a Spearman correlation coefficient > 0.9 are considered highly correlated and included in subsequent analysis. Second, consistency checks are performed on the extracted radiomics features, retaining those with good stability for model construction. Enhanced CT arterial phase images of 30 patients were randomly selected from the training set for Region of Interest (ROI) delineation. Inter- and intra-class correlation coefficients (ICCs) of the extracted radiomics features were calculated to assess the consistency of measurement results for the same data by different observers and the repeatability of two measurements by the same observer. Initially, two physicians simultaneously delineated the arterial phase ROIs of the 30 patients and extracted features to observe inter-group consistency; one week later, the same group of physicians delineated the ROIs again and extracted features to assess intra-group consistency. This yields reliable radiomics features unaffected by segmentation variability, retaining features with intra- and inter-group ICCs > 0.75. The Mann-Whitney U test (P < 0.05) is used for initial screening of eligible features. Subsequently, minimum redundancy maximum relevance (mRMR) is used to remove redundant and irrelevant radiomics features. The Least Absolute Shrinkage and Selection operator (LASSO) is then used for further screening of the remaining features. To prevent overfitting, 5-fold cross-validation is employed for data validation. By adjusting multiple penalty coefficients (lambda, λ), features with non-zero coefficients are retained; these are considered optimal by the algorithm. A multivariate logistic regression method is used to build a radiomics model based on the retained features, calculating the radiomics score (Rad-score) for each patient. This score is determined by a linear combination of the retained features and their corresponding coefficients, calculated as the sum of the products of each feature and its coefficient. The final features, after LASSO regression screening, were used to construct an image omics model using nine machine learning algorithms, including LR, SVM, Random Forest, Extra Trees, XG Boost, Light GBM, Gradient Boosting, Ada Boost, and MLP. All machine learning processes were implemented using Python 3.11.4.

[0035] 7. Construction of clinical models and nomogram models Independent clinical predictors associated with HIF-2α expression were screened using univariate and multivariate logistic regression analyses, and a clinical prediction model was constructed. A joint prediction model was established by combining previously selected radiomics features, presented in nomogram form. The nomogram visually displays different parameters and their corresponding scores; the prediction score is obtained by summing the corresponding scores of each parameter, thereby assessing the probability of HIF-2α expression. Receiver operating characteristic (ROC) curves were used to evaluate the predictive power of each model for HIF-2α expression. Evaluation parameters included area under the curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE), and the DeLong test was used to compare the differences in predictive power between models. The Hosmer-Lemeshow (HL) goodness-of-fit test was used to evaluate the calibration of the prediction models, and the results are expressed as calibration curves. Decision curve analysis (DCA) was used to compare the net benefits of the prediction model under different threshold probabilities and to evaluate its clinical application value.

[0036] 8. Statistical methods All statistical analyses were performed using Python 3.9.7 and SPSS software (Version 26.0, IBM). The Shapiro-Wilk test was used to assess whether the measurement data conformed to a normal distribution. The data conformed to a normal distribution. Quantitative data were expressed as mean ± standard deviation, and independent samples t-tests were used for intergroup comparisons. Non-normally distributed quantitative data were expressed as median and interquartile ranges, and Mann-Whitney U tests were used for intergroup comparisons. Categorical data were expressed as percentages (%), and chi-square tests or Fisher's exact tests were used for intergroup comparisons. Odds ratios (ORs) and their 95% confidence intervals (95% CI) for each predictor were calculated using multivariate logistic regression to screen for independent predictors of HIF-2α expression and construct clinical models and nomograms. ROC curves were used to evaluate model efficacy, and the area under the curve (AUC), accuracy, sensitivity, and specificity were calculated. The DeLong test was used to compare the differences in predictive efficacy among the models. The Hosmer-Lemeshow goodness-of-fit test and calibration curves were used to evaluate the calibration of the nomograms. Decision curve analysis (DCA) was used to assess the clinical benefit of each model. The significance level was set at α=0.05, and a two-sided p-value <0.05 was considered statistically significant.

[0037] The technical solutions provided in this application have at least the following technical effects or advantages: This application provides a method, device, equipment, and medium for predicting HIF-2α expression levels in renal cell carcinoma. By integrating CT radiomics features with key clinical parameters, a combined model capable of non-invasively predicting HIF-2α expression levels in clear cell renal cell carcinoma (ccRCC) preoperatively is constructed. Its core advantage lies in providing a more convenient and reproducible assessment tool compared to traditional methods relying on postoperative pathological biopsy. This tool can provide a forward-looking decision-making basis for the precise application of HIF-2α targeted inhibitors, thereby potentially avoiding unnecessary treatment side effects and assisting in achieving truly personalized precision medicine. Based on CT radiomics technology, this invention screens imaging features significantly associated with HIF-2α expression levels and constructs a novel nomogram prediction model in conjunction with clinical indicators, aiming to establish a non-invasive assessment system for HIF-2α expression levels in clear cell renal cell carcinoma. This invention innovatively correlates radiomics features, clinical features, and HIF-2α expression levels, providing a scientific basis for the formulation of individualized treatment plans and efficacy evaluation for ccRCC patients, with the aim of improving patients' clinical efficacy and quality of life.

[0038] II. Results 1. General clinical and pathological data A total of 274 patients pathologically diagnosed with ccRCC were included, comprising 190 males and 84 females, with a mean age of 56.47 ± 11.51 years. They were randomly assigned in a 7:3 ratio to the training set (n=191) and the validation set (n=83). 144 patients showed high HIF-2α expression, and 130 showed low expression. In both the training and validation sets, no statistically significant differences were observed in any clinical characteristics among the ccRCC patient groups (P > 0.05). Patients were divided into high and low expression groups based on HIF-2α expression levels. No statistically significant differences were found between the two groups in terms of age, BMI, sex, hypertension, diabetes, hyperlipidemia, tumor location, intraoperative hemorrhage, sarcomatoid degeneration, clinical N stage, and clinical M stage (P > 0.05). However, statistically significant differences were observed in tumor maximum diameter (P < 0.001), surgical type (P < 0.001), WHO / ISUP grade (P < 0.001), tumor necrosis (P = 0.008), tumor thrombus formation (P < 0.001), fat invasion (P < 0.001), and clinical T stage (P < 0.001) (P < 0.001). Univariate logistic regression analysis showed that the following factors significantly influenced tumor size: maximum tumor diameter (OR = 1.159, 95% CI: 1.020-1.268, P = 0.008), surgical type (OR = 2.304, 95% CI: 1.400-3.662, P = 0.005), WHO / ISUP grade (OR = 3.252, 95% CI: 1.804-5.256, P < 0.001), tumor necrosis (OR = 2.136, 95% CI: 1.248-2.954, P = 0.020), tumor thrombus formation (OR = 2.208, 95% CI: 1.316-3.762, P = 0.008), and fatty invasion (OR = 2.505, 95% CI: 1.508-4.273). P=0.003) and clinical T stage (OR= 2.580, 95% CI: 1.619-3.930, P<0.001) were associated with HIF-2α expression, and the differences were statistically significant (P < 0.05).

[0039] The seven meaningful indicators from the univariate logistic regression analysis were further incorporated into the multivariate analysis. Logistic regression analysis showed that WHO / ISUP classification (OR = 2.808, 95% CI: 1.553-4.841, P = 0.002) and clinical T stage (OR = 2.056, 95% CI: 1.235-3.377, P = 0.016) were independent clinical risk factors for HIF-2α expression, which can be used to construct a clinical model.

[0040] 2. Radiomics Feature Screening A total of 1815 radiomics features were extracted from the ROIs using PyRadiomics. Features with ICC > 0.75 were selected, and the Mann-Whitney U test, Pearson correlation analysis, and mRMR were used for further analysis. The LASSO algorithm was used for screening, and the penalty parameter λ was adjusted through five-fold cross-validation to select radiomics features with non-zero coefficients. Finally, seven radiomics features with the highest statistical significance were selected, and the Rad-score formula was derived based on their corresponding feature coefficients. The Rad-score formula is as follows: Rad-score= 0.29545454545454564 -0.038112 * original shape Sphericity +0.048922 * wavelet HLL ngtdm Complexity +0.014029 * wavelet LLL ngtdm Busyness -0.071105 * wavelet LLH glcm Imc2 -0.032482 * lbp 3D k glszm ZonePercentage -0.005019 * exponential ngtdm Coarseness +0.009697 * wavelet LHL ngtdm Busyness 3. Construction and evaluation of radiomics models The final selected radiomics features were incorporated into nine machine learning algorithms—LR, SVM, Random Forest, Extra Trees, XGBoost, Light GBM, Gradient Boosting, Ada Boost, and MLP—to construct predictive models. ROC analysis was performed on HIF-2α expression prediction for different cohorts. The AUC, accuracy, sensitivity, and specificity of the prediction models from different machine learning algorithms were also analyzed. Results showed that, in the training set, Random Forest, Extra Trees, XGBoost, Light GBM, Gradient Boosting, and Gradient Boosting were the most effective algorithms. Boosting methods all demonstrated good model effectiveness, with the Random Forest model performing best on the training set. Its AUC reached 0.888 (95% CI: 0.836–0.939), accuracy was 0.795, sensitivity was 0.862, and specificity was 0.768. On the validation set, the model achieved an AUC of 0.825 (95% CI: 0.784–0.965), accuracy of 0.778, sensitivity of 0.750, and specificity of 0.789. These results indicate that the Random Forest model is the most ideal radiomics model, exhibiting the best predictive value for HIF-2α expression and demonstrating sustainability and stability.

[0041] 4. Construction and evaluation of clinical models and nomogram models. Previous univariate and multivariate logistic regression analyses identified WHO / ISUP classification (OR=2.808, 95% CI: 1.553-4.841, P=0.002) and clinical T stage (OR=2.056, 95% CI: 1.235-3.377, P=0.016) as independent risk factors for predicting HIF-2α expression. Based on these results, a clinical model was constructed. The AUC values ​​of this model on the training and validation sets were 0.84 (95% CI: 0.785-0.912) and 0.738 (95% CI: 0.588-0.887), respectively. A nomogram model was constructed by combining the radiomics Rad-score with independent clinical predictors such as WHO / ISUP classification and clinical T stage, as shown below. Figure 3 As shown, a patient with a WHO / ISUP classification of 3 (31.7 points), a clinical T stage of 3 (24 points), and a Rad-score of 0.75 (75 points), with a total score of 130.6 points, corresponds to approximately 87% probability of HIF-2α high expression. Immunohistochemical results confirmed that this patient indeed had high HIF-2α expression. This demonstrates the clinical applicability and accuracy of the nomogram prediction model.

[0042] This invention, based on CT radiomics technology, screens imaging features significantly associated with HIF-2α expression levels and constructs a novel nomogram prediction model in conjunction with clinical indicators. The aim is to establish a non-invasive assessment system for HIF-2α expression levels in clear cell renal cell carcinoma (ccRCC). This innovative study correlates radiomics features, clinical characteristics, and HIF-2α expression levels, providing a scientific basis for the development of individualized treatment plans and efficacy evaluation for ccRCC patients, with the goal of improving clinical efficacy and quality of life.

[0043] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting HIF-2α expression levels in renal cell carcinoma, characterized in that, Includes the following steps: S10. Obtain clinicopathological data and enhanced arterial phase images of the abdomen from patients with clear cell renal cell carcinoma; S20. Delineate the three-dimensional region of interest of the tumor using the enhanced abdominal arterial phase image; S30. Extract image radiomics features from the three-dimensional region of interest of the tumor, and sequentially filter the image radiomics features through Z-score normalization, stability test, mRMR algorithm and LASSO regression algorithm to obtain the filtered features. Calculate the radiomics score based on the filtered features. S40. Screen out independent clinical predictors related to HIF-2α expression from the clinicopathological data. The independent clinical predictors include WHO / ISUP classification and clinical T stage. The independent clinical predictor and the radiomics score are input into the joint prediction model to obtain a prediction score, and the probability of HIF-2α expression is assessed using the prediction score.

2. The method for predicting HIF-2α expression level in renal cell carcinoma according to claim 1, characterized in that, Specifically, S20 is: S21. In the enhanced arterial phase image of the abdomen, delineate the three-dimensional region of interest of the tumor along the tumor edge; S22. Save the outlined 3D region of interest file of the tumor in NIFTI format.

3. The method for predicting HIF-2α expression level in renal cell carcinoma according to claim 1, characterized in that, Specifically, S30 is: S31. Use the open-source software package PyRadiomics to extract radiomics features of the three-dimensional region of interest of the tumor; the radiomics features include multiple radiomics sub-features, and the types of radiomics sub-features include: geometric features, texture features and histogram features; The geometric shape features are used to describe the three-dimensional shape of the tumor, the texture features are used to describe the second-order and higher-order spatial distribution of intensity, and the histogram features are used to describe the first-order statistical distribution of voxel intensity within the tumor. S32. Perform Z-score standardization on the extracted radiomics features so that the mean value of each radiomics sub-feature is 0 and the standard deviation is 1. S33. Stability test is performed using Spearman correlation coefficient analysis to obtain the Spearman correlation coefficient of each radiomic sub-feature. The radiomic sub-features whose Spearman correlation coefficients are within the set threshold are retained to obtain stable radiomic features. S34. The stable radiomics features are filtered out by using the mRMR algorithm to remove the radiomics sub-features within the stable radiomics features that meet the set conditions, and the remaining radiomics features are obtained. S35. Combining cross-validation, the remaining radiomics features are screened using the LASSO regression algorithm to obtain radiomics features with non-zero coefficients as the screened features, and the corresponding coefficients of the screened features are obtained. S36. The radiomics score is calculated using the selected features and their corresponding coefficients. The radiomics score is calculated by summing the products of each selected feature and its corresponding coefficient.

4. The method for predicting HIF-2α expression level in renal cell carcinoma according to claim 1, characterized in that, Specifically, S40 is: S41. The clinicopathological data were analyzed using univariate and multivariate logistic regression algorithms to screen out independent clinical predictors associated with HIF-2α expression; the screened independent clinical predictors included WHO / ISUP classification and clinical T stage. S42. A joint prediction model is established using the random forest algorithm, and the results of the joint prediction model are presented in the form of a nomogram. S43. Input the radiomics score and independent clinical predictors as prediction parameters into the joint prediction model; The predicted scores corresponding to each prediction parameter are obtained by using the Logistic regression algorithm; The nomogram of the joint prediction model displays the prediction scores corresponding to each prediction parameter; the sum of the prediction scores yields the probability of HIF-2α expression.

5. A device for predicting HIF-2α expression levels in renal cell carcinoma, characterized in that, Includes the following modules: The data acquisition module acquires clinicopathological data and enhanced arterial phase images of the abdomen from patients with clear cell renal cell carcinoma. The image delineation module delineates the three-dimensional region of interest of the tumor using the enhanced arterial phase image of the abdomen; The feature selection module extracts image radiomics features from the three-dimensional region of interest of the tumor, and sequentially filters the image radiomics features through Z-score normalization, stability test, mRMR algorithm and LASSO regression algorithm to obtain the filtered features. The radiomics score is calculated based on the filtered features. The model evaluation module screens independent clinical predictors associated with HIF-2α expression from the clinicopathological data. These independent clinical predictors include WHO / ISUP classification and clinical T stage. The independent clinical predictors and the radiomics score are input into a joint prediction model to obtain a prediction score. The prediction score is then used to evaluate the probability of HIF-2α expression.

6. The device for predicting HIF-2α expression levels in renal cell carcinoma according to claim 5, characterized in that, The image delineation module is specifically as follows: The delineation unit delineates the three-dimensional region of interest of the tumor along the tumor margin in the enhanced arterial phase image of the abdomen. The storage unit saves the outlined three-dimensional region of interest (ROI) file of the tumor in NIFTI format.

7. The device for predicting HIF-2α expression levels in renal cell carcinoma according to claim 5, characterized in that, The feature filtering module specifically includes: The feature extraction unit uses the open-source software package PyRadiomics to extract radiomics features of the three-dimensional region of interest of the tumor; the radiomics features include several radiomics sub-features, and the types of radiomics sub-features include: geometric features, texture features, and histogram features; The geometric shape features are used to describe the three-dimensional shape of the tumor, the texture features are used to describe the second-order and higher-order spatial distribution of intensity, and the histogram features are used to describe the first-order statistical distribution of voxel intensity within the tumor. The standardization processing unit performs Z-score standardization on the extracted radiomics features, so that the mean value of each radiomics sub-feature is 0 and the standard deviation is 1. The stability test unit uses Spearman correlation coefficient analysis to perform stability tests, obtains the Spearman correlation coefficient of each radiomic sub-feature, retains the radiomic sub-features whose Spearman correlation coefficients are within a set threshold, and obtains stable radiomic features. The mRMR filtering unit filters the stable radiomics features by using the mRMR algorithm to filter out the radiomics sub-features within the stable radiomics features that meet the set conditions, thus obtaining the remaining radiomics features. The LASSO regression algorithm unit, combined with cross-validation, uses the LASSO regression algorithm to filter the remaining radiomics features, obtaining radiomics features with non-zero coefficients as the filtered features, and obtaining the corresponding coefficients of the filtered features. The scoring calculation unit calculates the radiomics score based on the selected features and their corresponding coefficients. The calculation method for the radiomics score is the sum of the products of each selected feature and its corresponding coefficient.

8. The device for predicting HIF-2α expression levels in renal cell carcinoma according to claim 5, characterized in that, The model evaluation module specifically includes: The Independent Clinical Predictor Unit uses univariate and multivariate logistic regression algorithms to analyze the clinicopathological data and screen out independent clinical predictors associated with HIF-2α expression; the screened independent clinical predictors include WHO / ISUP classification and clinical T stage; The model building unit establishes a joint prediction model using the random forest algorithm, and the results of the joint prediction model are presented in the form of a no-line graph. The probability prediction unit inputs the radiomics score and independent clinical predictors as prediction parameters into the joint prediction model; The predicted scores corresponding to each prediction parameter are obtained by using the Logistic regression algorithm; The nomogram of the joint prediction model displays the prediction scores corresponding to each prediction parameter; the prediction scores are summed to obtain the probability of HIF-2α expression.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.

Images