Methods and models for predicting the risk of early postoperative renal function decline in patients with renal cell carcinoma following partial nephrectomy

Abstract

The application provides a kind of prediction method and model of early renal function decline risk of renal cell carcinoma patient after partial nephrectomy.Methods obtain patient clinical parameters by preoperative nomogram or preoperative-intraoperative joint nomogram prediction model, assign each parameter risk score and calculate total score, determine the prediction probability of early postoperative renal function decline and can be visualized output.Preoperative nomogram prediction model parameters include age, history of diabetes, preoperative estimated glomerular filtration rate, kidney score and radiomics score;Intraoperative ischemia time is added to the joint model.Model construction is standardized by CT image, region segmentation, radiology feature extraction and preprocessing, etc., which can assist clinical precise risk assessment and optimize treatment decision.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary prediction technology, and in particular to a method and model for predicting the risk of early renal function decline after partial nephrectomy in patients with renal cell carcinoma. Background Technology

[0002] Partial nephrectomy (PN) is considered the standard treatment for early-stage renal cell carcinoma (RCC). One of its primary goals, besides removing the kidney tumor, is to preserve kidney function. However, a significant portion of patients experience a decline in kidney function post-surgery, which negatively impacts quality of life and overall survival. Early kidney damage increases the risk of chronic kidney disease, end-stage renal disease, and cardiovascular events, leading to increased short-term and long-term mortality. Therefore, early identification of deteriorating kidney function is crucial to preventing further malignant progression of kidney damage.

[0003] Currently, predictors of renal function decline after nephrectomy (PN) primarily include preoperative clinical parameters—such as age, diabetes status, and baseline renal function—and intraoperative variables, including ischemic time, fluid management, and surgical approach. Although several predictive models and risk scores have been proposed, most are derived from single-center datasets, involve limited sample sizes, or lack external validation, limiting their reliability and broad clinical applicability. Therefore, the need for robust and generalizable tools capable of predicting postoperative renal function outcomes remains unmet.

[0004] Radiomics has emerged as a non-invasive quantitative imaging approach that captures macroscopic and microscopic tissue features beyond visual observation. CT-based radiomics has shown great potential in improving the diagnosis, staging, subtype identification, and prognosis of renal cell carcinoma by extracting features related to tumor heterogeneity, texture, and morphology. In addition to tumor features, radiographic features from normal renal parenchyma can reflect tissue integrity, perfusion patterns, and microstructural alterations, providing opportunities to assess functional reserve and predict renal effluent. Despite these advantages, limited research has focused on combining radiographic features from both tumor and normal renal parenchyma to predict early renal function decline after nephrectomy (PN). Furthermore, multicenter evidence in this area remains scarce, and the combined clinical-radiological predictive framework has not been systematically validated.

[0005] Therefore, this invention aims to develop and externally validate a CT radiography-derived nomogram that combines radiographic features of the tumor and ipsilateral renal parenchyma with clinical variables to accurately predict early renal function decline after nephron-sparing surgery in patients with localized renal cancer (CC). Through a multicenter design, rigorous feature selection, and independent validation, this study aims to provide a clinically practical and scalable tool to enhance risk stratification and personalized management in nephron-sparing surgery. This study conforms to STROCSS criteria. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a method for constructing a nomogram to predict early renal function decline after nephron-sparing surgery in patients with localized renal cancer (CC). Through multicenter design, rigorous feature selection and independent validation, a clinically practical and scalable tool is realized to enhance risk stratification and personalized management of nephron-sparing surgery.

[0007] In a first aspect, the present invention provides a method for predicting the risk of early renal function decline after partial nephrectomy in patients with renal cell carcinoma. The method involves obtaining clinical parameters of the patient to be predicted through a prediction model; performing integral allocation on the clinical parameters to obtain risk scores for each clinical parameter; calculating the total risk score based on the risk scores of each clinical parameter; determining the predicted probability of early renal function decline after partial nephrectomy based on the total risk score and visually outputting the result.

[0008] The prediction model includes a preoperative nomogram prediction model and / or a combined preoperative and intraoperative nomogram prediction model; the clinical parameters corresponding to the preoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, and radiomics score; the clinical parameters corresponding to the combined preoperative and intraoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, radiomics score, and intraoperative ischemic time.

[0009] The process of constructing the preoperative normogram prediction model includes:

[0010] S11. Screen the patients in the dataset and obtain the preoperative enhanced CT scan images of each patient in the dataset. The enhanced CT scan images include the cortico-medullary phase, the renal angiography phase, and the excretion phase.

[0011] S12. Standardize the CT enhanced scan image, including resampling the image into isotropic voxels and normalizing the intensity according to the requirements of the image biomarker standardization initiative.

[0012] S13. The tumor region and the ipsilateral normal renal parenchyma region are determined by manual segmentation in each phase of the enhanced CT scan image;

[0013] S14. Extract radiographic features from the tumor region and the ipsilateral normal renal parenchyma region respectively to obtain a tumor region feature set and an ipsilateral normal renal parenchyma region feature set. The radiographic features include shape-related features, first-order statistical features, gray-level run-length matrix features, gray-level co-occurrence matrix features, gray-level correlation matrix features, gray-level size region matrix features, and neighborhood gray-level difference matrix features.

[0014] S15. Preprocess the extracted radiological features;

[0015] S16. Based on LASSO logistic regression, the feature subsets with the best prediction performance are selected from the preprocessed tumor region feature set and the ipsilateral normal renal parenchyma region feature set, respectively, to obtain the tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset. The tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset are fused to form a joint radiomics feature set.

[0016] S17. Based on the coefficients determined by the LASSO logistic regression, the features in the joint radiomics feature set are linearly weighted and combined to calculate a comprehensive radiomics score for each patient in the dataset.

[0017] S18. Determine the preoperative clinical variable set through multivariate logistic regression. The preoperative clinical variable set includes the age, diabetes, estimated glomerular filtration rate, and kidney score of the patients in the dataset.

[0018] S19. Using the radiomics score and the preoperative clinical variable set as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained by logistic regression algorithm, and the preoperative nomogram prediction model is obtained after model validation.

[0019] The construction process of the preoperative-intraoperative combined nomogram prediction model includes:

[0020] S21. Obtain the combined radiological features and preoperative clinical variable set through S11 to S18;

[0021] S22. The intraoperative clinical variables for partial nephrectomy in patients with renal cell carcinoma were determined by multivariate logistic regression, wherein the intraoperative clinical variable was intraoperative ischemia time.

[0022] S23. Using the radiomics score, the preoperative clinical variable set, and the intraoperative clinical variable as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained using a logistic regression algorithm. After model validation, the preoperative-intraoperative combined nomogram prediction model is obtained.

[0023] Furthermore, the aforementioned early postoperative renal function decline is defined as: within 3-24 months after partial nephrectomy, the patient's EGFR decreases by ≥25% compared to the preoperative baseline;

[0024] The patient selection criteria for the dataset are: patients with localized renal cell carcinoma who have undergone partial nephrectomy, have completed multi-phase enhanced CT scans including the cortico-medullary phase, nephrography phase, and excretion phase before surgery, and have complete preoperative and postoperative 3-24 month renal function follow-up data.

[0025] The kidney score is a quantitative score based on tumor complexity, used to assess tumor size, anatomical location depth, and distance from key renal structures.

[0026] The image resampling involves resampling the image to an isotropic voxel of 1×1×1 mm³; the intensity normalization is based on the gray value distribution of the CT image to eliminate the variability in radiographic features caused by different scanners, acquisition protocols, and reconstruction kernels.

[0027] Furthermore, S13 specifically involves: in the ITK-SNAP software, at least two radiologists manually segment the regions of interest (ROIs) of each phase of the standardized CT enhanced scan images to determine the tumor region and the ipsilateral normal renal parenchyma region, and the segmentation results are reviewed and verified by a third senior radiologist.

[0028] Furthermore, in S14, extracting radiological features from the tumor region and the ipsilateral normal renal parenchyma region specifically includes:

[0029] In each phase of CT images, the renal tumor region and the ipsilateral normal renal parenchyma region are segmented to obtain the corresponding region of interest for the tumor and the region of interest for the normal renal parenchyma.

[0030] An initial radiomics feature set was extracted from images of the tumor region of interest and the normal renal parenchyma region of interest at each stage.

[0031] Furthermore, the preprocessing in S15 includes repetitive filtering, residual feature standardization, and multi-step dimensionality reduction:

[0032] The repetitive filtering refers to: calculating the intra-group correlation coefficients within and between observers, and removing features with intra-group correlation coefficients lower than 0.80;

[0033] The standardization of residual features refers to the standardization of the residual features after the repeatability filtering using Z-score normalization.

[0034] The multi-step dimensionality reduction process includes: calculating the Pearson correlation coefficient among the remaining features and removing redundant features with an absolute correlation coefficient > 0.90; screening features related to postoperative renal function decline through univariate logistic regression; determining the optimal regularization parameter through LASSO regression with 10 cross-validations and retaining features with non-zero coefficients at the optimal regularization parameter.

[0035] Furthermore, the model validation involves using a predetermined number of iterations of Bootstrap resampling, combined with ROC curves, calibration curves, and Hosmer-Lemeshow tests to evaluate model stability; including:

[0036] Internal validation: Validation is performed through a validation queue at an independent internal medical center;

[0037] External validation: The universality of the model was validated through validation cohorts of at least 10 independent external medical centers and public datasets;

[0038] Clinical efficacy validation: Through decision curve analysis, within a threshold probability range of 0-1.0, the net clinical benefit of the model compared to the "full treatment" and "no treatment" strategies was evaluated.

[0039] Secondly, the present invention provides a method for predicting the risk of early renal function decline after partial nephrectomy in patients with renal cell carcinoma. The method involves obtaining clinical parameters of the patient to be predicted through a prediction model; performing integral allocation on the clinical parameters to obtain risk scores for each clinical parameter; calculating the total risk score based on the risk scores of each clinical parameter; determining the predicted probability of early renal function decline after partial nephrectomy based on the total risk score and visualizing the output.

[0040] The prediction model includes a preoperative nomogram prediction model and / or a combined preoperative and intraoperative nomogram prediction model; the clinical parameters corresponding to the preoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, and radiomics score; the clinical parameters corresponding to the combined preoperative and intraoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, radiomics score, and intraoperative ischemic time.

[0041] The process of constructing the preoperative normogram prediction model includes:

[0042] S11. Screen the patients in the dataset and obtain the preoperative enhanced CT scan images of each patient in the dataset. The enhanced CT scan images include the cortico-medullary phase, the renal angiography phase, and the excretion phase.

[0043] S12. Standardize the CT enhanced scan image, including resampling the image into isotropic voxels and normalizing the intensity according to the requirements of the image biomarker standardization initiative.

[0044] S13. The tumor region and the ipsilateral normal renal parenchyma region are determined by manual segmentation in each phase of the enhanced CT scan image;

[0045] S14. Extract radiographic features from the tumor region and the ipsilateral normal renal parenchyma region respectively to obtain a tumor region feature set and an ipsilateral normal renal parenchyma region feature set. The radiographic features include shape-related features, first-order statistical features, gray-level run-length matrix features, gray-level co-occurrence matrix features, gray-level correlation matrix features, gray-level size region matrix features, and neighborhood gray-level difference matrix features.

[0046] S15. Preprocess the extracted radiological features;

[0047] S16. Based on LASSO logistic regression, the feature subsets with the best prediction performance are selected from the preprocessed tumor region feature set and the ipsilateral normal renal parenchyma region feature set, respectively, to obtain the tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset. The tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset are fused to form a joint radiomics feature set.

[0048] S17. Based on the coefficients determined by the LASSO logistic regression, the features in the joint radiomics feature set are linearly weighted and combined to calculate a comprehensive radiomics score for each patient in the dataset.

[0049] S18. Determine the preoperative clinical variable set through multivariate logistic regression. The preoperative clinical variable set includes the age, diabetes, estimated glomerular filtration rate, and kidney score of the patients in the dataset.

[0050] S19. Using the radiomics score and the preoperative clinical variable set as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained by logistic regression algorithm, and the preoperative nomogram prediction model is obtained after model validation.

[0051] The construction process of the preoperative-intraoperative combined nomogram prediction model includes:

[0052] S21. Obtain the combined radiological features and preoperative clinical variable set through S11 to S18;

[0053] S22. The intraoperative clinical variables for partial nephrectomy in patients with renal cell carcinoma were determined by multivariate logistic regression, wherein the intraoperative clinical variable was intraoperative ischemia time.

[0054] S23. Using the radiomics score, the preoperative clinical variable set, and the intraoperative clinical variable as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained using a logistic regression algorithm. After model validation, the preoperative-intraoperative combined nomogram prediction model is obtained.

[0055] Furthermore, the image resampling involves resampling the image to an isotropic voxel of 1×1×1 mm³; the intensity normalization is based on the gray value distribution of the CT image to eliminate the variability in radiographic features caused by different scanners, acquisition protocols, and reconstruction kernels.

[0056] Furthermore, S13 specifically involves: in the ITK-SNAP software, at least two radiologists manually segment the regions of interest (ROIs) of each phase of the standardized CT enhanced scan images to determine the tumor region and the ipsilateral normal renal parenchyma region, and the segmentation results are reviewed and verified by a third senior radiologist.

[0057] Furthermore, in S14, extracting radiographic features from the tumor region and the ipsilateral normal renal parenchyma region specifically includes:

[0058] In each phase of CT images, the renal tumor region and the ipsilateral normal renal parenchyma region are segmented to obtain the corresponding region of interest for the tumor and the region of interest for the normal renal parenchyma.

[0059] An initial radiomics feature set was extracted from images of the tumor region of interest and the normal renal parenchyma region of interest at each stage.

[0060] Furthermore, the preprocessing in S15 includes repetitive filtering, residual feature standardization, and multi-step dimensionality reduction:

[0061] The repetitive filtering refers to: calculating the intra-group correlation coefficients within and between observers, and removing features with intra-group correlation coefficients lower than 0.80;

[0062] The standardization of residual features refers to the standardization of the residual features after the repeatability filtering using Z-score normalization.

[0063] The multi-step dimensionality reduction process includes: calculating the Pearson correlation coefficient among the remaining features and removing redundant features with an absolute correlation coefficient > 0.90; screening features related to postoperative renal function decline through univariate logistic regression; determining the optimal regularization parameter through LASSO regression with 10 cross-validations and retaining features with non-zero coefficients at the optimal regularization parameter.

[0064] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0065] (1) Two complementary models are provided: the preoperative nomogram prediction model can be used for preoperative consultation and surgical planning; the preoperative-intraoperative combined nomogram prediction model can be used for real-time risk reassessment after obtaining intraoperative information (such as ischemic time), guide postoperative management, and meet the needs of different stages of diagnosis and treatment.

[0066] (2) It demonstrates the feasibility of using conventional preoperative CT images for functional prediction, and provides a methodological paradigm for developing other non-invasive, low-cost risk prediction tools.

[0067] (2) Dual-region radiographic feature extraction: Radiographic features are extracted from the tumor region and the ipsilateral normal renal parenchyma region, rather than only from the tumor region. This can capture more comprehensive tissue features related to renal function (such as tumor heterogeneity, integrity and perfusion pattern of normal renal parenchyma) and provide richer quantitative evidence for predicting postoperative renal function decline.

[0068] (3) Utilization of multi-phase CT images: Make full use of multi-phase enhanced CT (cortical-medullian phase, nephrography phase and excretion phase) to obtain more comprehensive tumor biology and renal perfusion information, and further enhance the predictive value of features.

[0069] (4) Feature selection method: Repeated filtering, high correlation removal and LASSO regression are used to select features and construct radiological features with high predictive power.

[0070] (5) Clinical-radiological integration: Combining radiological features with clinical factors (age, diabetes, preoperative EGFR, renal score, ischemia time) to make up for the limitations of a single data dimension, constructing preoperative nomograms and preoperative-intraoperative nomograms, and improving the accuracy of the model in predicting the risk of postoperative renal function decline.

[0071] (6) Multicenter validation: External validation was conducted on 10 independent medical centers and a public dataset to ensure the universality and clinical applicability of the model.

[0072] (7) Clinical decision support: Nomograph has good discrimination, calibration and clinical applicability, and can guide individualized surgical decisions and postoperative management.

[0073] 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

[0074] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0075] Figure 1 This is a schematic diagram illustrating the composition of the prediction model of this invention;

[0076] Figure 2 This is a flowchart illustrating the construction process of the prediction model of this invention;

[0077] Figure 3 This is a flowchart illustrating the research process of an embodiment of the present invention;

[0078] Figure 4 This is a table of clinical characteristics of 1440 patients with localized renal cell carcinoma in the training and validation queues of this invention.

[0079] Figure 5 This is a summary table of radiomics features for radiomics scoring in an embodiment of the present invention;

[0080] Figure 6 This is a schematic diagram of three radiological features established using Logistic regression analysis in an embodiment of the present invention;

[0081] Figure 7 This is a summary table of the performance of radiomics features in the training and validation queues according to embodiments of the present invention;

[0082] Figure 8 To explore independent predictors of renal function decline using univariate and multivariate logistic regression analysis in this embodiment of the invention;

[0083] Figure 9 This diagram illustrates the construction and validation of a preoperative radiological-clinical nomogram according to an embodiment of the present invention. A in the diagram is a visualization of the preoperative nomogram, integrating five preoperative factors: age, diabetes, preoperative EGFR, kidney score, and radiological characteristics. B is the ROC curve of the preoperative nomogram in the training cohort, showing its discriminative performance. C is the calibration curve of the preoperative nomogram in the training cohort, validating the consistency between the model's predicted values ​​and actual observed values. D is the decision curve analysis (DCA) of the preoperative nomogram in the training cohort, showing the net clinical benefit within the probability threshold range of 0-1.0.

[0084] Figure 10 This is a schematic diagram of the Bootstrap internal validation of the preoperative nomogram in an embodiment of the present invention; the ROC curve generated by 1000 iterations of Bootstrap resampling;

[0085] Figure 11 This figure shows the external validation results of preoperative radiological-clinical nomograms according to an embodiment of the present invention. In the figure, A, D, and G are the ROC curves of external validation cohorts 1, 2, and 3, respectively, showing the discrimination performance of the nomogram. In the figure, B, E, and H are the calibration curves of external validation cohorts 1, 2, and 3, respectively. The consistency between the model predictions and the actual observed values ​​is verified by the Hosmer-Lemeshow goodness-of-fit test (all P values ​​> 0.05). C, F, and I are the decision curve analysis (DCA) of external validation cohorts 1, 2, and 3, respectively, showing the net clinical benefit of the nomogram in the probability threshold range of 0-1.0.

[0086] Figure 12 This invention includes a nomogram and a power table of five preoperative features in differentiating renal function.

[0087] Figure 13This invention relates to the construction and internal validation of a preoperative-intraoperative combined radiological-clinical nomogram. Figure A shows the visualization of the preoperative-intraoperative combined radiological-clinical nomogram; B and E are the ROC curves of the preoperative nomogram in the training cohort and the internal validation cohort, respectively; C and F are the calibration curves for the training cohort and the internal validation cohort, respectively; and D and G are the decision curve analyses for the training cohort and the internal validation cohort, respectively.

[0088] Figure 14 This invention includes a nomogram and a power table showing the effectiveness of six preoperative and intraoperative features in differentiating renal function.

[0089] Figure 15 This is an external validation result diagram of the preoperative-intraoperative combined radiological-clinical nomogram in an embodiment of the present invention; A, D, and G in the figure are the ROC curves of external validation cohorts 1, 2, and 3, respectively; B, E, and H in the figure are the calibration curves of external validation cohorts 1, 2, and 3, respectively; C, F, and I are the decision curve analysis (DCA) of external validation cohorts 1, 2, and 3, respectively.

[0090] Figure 16 This is a subgroup analysis of the predictive performance of preoperative and preoperative-intraoperative radiological-clinical nomograms in different clinical subgroups according to an embodiment of the present invention.

[0091] Figure 17 This is a nomogram showing the predictive performance of an embodiment of the present invention compared to existing models in predicting renal function decline after partial nephrectomy.

[0092] Figure 18 The following are examples of clinical applications of radiomics-clinical nomograms according to embodiments of the present invention. In the figure, A is an example of preoperative nomogram application, and B is an example of preoperative-intraoperative nomogram application.

[0093] Figure 19 This is a descriptive diagram of the region of interest (ROI) in CT images of normal kidney tissue and kidney tumors, as an embodiment of the present invention;

[0094] Figure 20 The ICC distribution of intratumoral radiomics features under different CT phases in this embodiment of the invention;

[0095] Figure 21 This is an ICC distribution table of radiomics features of normal renal tissue under different CT phases in embodiments of the present invention;

[0096] Figure 22 This is a schematic diagram of the allocation coefficients of the preoperative nomogram prediction model according to an embodiment of the present invention;

[0097] Figure 23 This is a schematic diagram of the allocation coefficients of the preoperative-intraoperative combined nomogram prediction model in an embodiment of the present invention. Detailed Implementation

[0098] The overall approach of the technical solution in this application is as follows: Addressing the shortcomings of existing tools for predicting early renal function decline after partial nephrectomy for renal cell carcinoma, such as single-center modeling, limited sample size, and lack of external validation, this application proposes a method that combines multi-phase CT radiographic features of the tumor and ipsilateral normal renal parenchyma (using rigorous feature screening to construct a radiographic signature) with preoperative clinical factors (age, diabetes, preoperative EGFR, and renal score) and intraoperative clinical factors (intraoperative ischemia time). Preoperative nomograms and preoperative-intraoperative nomograms are constructed separately, and external validation is performed using multi-center cohorts and public datasets to achieve accurate prediction of early postoperative renal function decline, providing a reliable tool for individualized surgical decisions and follow-up strategy optimization.

[0099] Example 1

[0100] This embodiment provides a method for predicting the risk of early renal function decline after partial nephrectomy in patients with renal cell carcinoma. In this invention, early renal function decline is defined as: within 3-24 months after partial nephrectomy, the patient's EGFR decreases by ≥25% compared to the preoperative baseline.

[0101] Although there is no universally accepted standard for “early” renal function decline after PN in the literature, in some studies, a threshold of ≥25% is considered a clinically significant worsening associated with the progression of chronic kidney disease.9,15 The 3–24 month time window was pre-selected to exclude transient perioperative fluctuations and to be consistent with a guideline-based follow-up schedule. EGFR was assessed at 3, 6, 12, 18, and 24 months post-operation. These follow-up time points were uniformly applied to all participating centers. When more than one EGFR measurement was available within a given time interval, the result closest to the predetermined month was used for analysis.

[0102] The prediction method involves obtaining clinical parameters of the patient to be predicted through a prediction model; performing integral allocation on the clinical parameters to obtain risk scores for each clinical parameter; calculating the total risk score based on the risk scores of each clinical parameter; and determining the predicted probability of early renal function decline after partial nephrectomy based on the total risk score and visualizing the output.

[0103] Among them, such as Figure 1 As shown, the prediction model includes a preoperative nomogram prediction model and / or a combined preoperative and intraoperative nomogram prediction model; the clinical parameters corresponding to the preoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, and radiomics score; the clinical parameters corresponding to the combined preoperative and intraoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, radiomics score, and intraoperative ischemic time;

[0104] like Figure 2 As shown, the construction process of the preoperative normogram prediction model includes:

[0105] S11. Screen the patients in the dataset and obtain the preoperative enhanced CT scan images of each patient in the dataset. The enhanced CT scan images include the cortico-medullary phase, the renal angiography phase, and the excretion phase.

[0106] The patient selection criteria for the dataset were: all enrolled patients obtained written informed consent. RCC patients who received PN in the hospital between January 2016 and June 2023 were randomly assigned to the training cohort and the internal validation cohort at a ratio of 7:3.

[0107] Patients who received RCC PN at five centers between January 2016 and June 2023 were included in the external validation cohort 1.

[0108] To further enhance external validation and increase institutional heterogeneity, five additional independent external centers were incorporated, forming external validation cohort 2.

[0109] In addition, imaging and clinical data of 79 patients with renal cell carcinoma were obtained from the public KITS23 online database (https: / / kits-challenge.org / kits23 / ). These patients had complete preoperative and postoperative renal function data, as well as preoperative enhanced CT images of the kidneys. They were designated as external validation cohort 3.

[0110] The inclusion criteria for the patients in the dataset are as follows:

[0111] (1) Postoperative pathological histological examination of localized renal cell carcinoma (T1-T3, N0, M0) conforms to AJCC 8th edition;

[0112] (2) PN occurred between January 2016 and June 2023;

[0113] (3) Preoperative contrast-enhanced CT scan images of the kidneys or abdomen;

[0114] (4) Effective postoperative follow-up and clinical re-examination;

[0115] (5) Obtain written informed consent from patients for inclusion in the study.

[0116] Exclusion is made if the patient meets any of the following criteria: (1) no cortical, nephrography or excretory CT sequence is available; (2) poor quality CT images are not suitable for analysis; (3) incomplete clinicopathological data; (4) missing renal function data during postoperative follow-up; (5) renal function assessment cannot be performed within two weeks prior to surgery.

[0117] Clinical variables of patients in the dataset were collected, including demographic and clinicopathological characteristics: age, sex, body mass index (BMI), Eastern Cooperative Oncology Group (ECOG) score, diabetes, smoking, alcohol consumption, hypertension, hyperlipidemia, preoperative estimated glomerular filtration rate (EGFR), kidney score, ischemic time, robot-assisted partial nephrectomy or laparoscopic partial nephrectomy approach, surgeon experience, duration of anesthesia, surgical approach, intraoperative colloid volume, intraoperative lens volume, CT stage, histological subtype, WHO / ISUP grade, and tumor size.

[0118] According to the 2025 NCCN RCC guidelines, post-nephrectomy follow-up includes regular monitoring of patients through face-to-face visits and telephone communication. The follow-up strategy primarily focuses on renal function assessment and laboratory tests, supplemented by chest, abdominal, and pelvic CT scans. Abdominal MRI and contrast-enhanced ultrasound may also be performed if deemed necessary. For patients unsuitable for contrast agents, MRI or ultrasound is considered the preferred imaging modality. Additional renal biopsy or further intervention may be required if imaging or clinical findings suggest tumor recurrence or residual disease.

[0119] Smoking is defined as consuming at least one cigarette per day for at least six months. Alcohol consumption is defined as drinking alcohol at least once a week, with each instance containing at least 30 ml of alcohol, for at least six months. Diabetes is defined as a written diagnosis of diabetes recorded in the patient's medical record prior to surgery. Patients with only impaired glucose tolerance or laboratory abnormalities but without a confirmed diagnosis are not classified as having diabetes. Hypertension is defined as a medically confirmed diagnosis based on previous medical records, excluding cases determined solely by elevated blood pressure readings or medication use.

[0120] S12. Standardize the CT enhanced scan image, including resampling the image into isotropic voxels and normalizing the intensity according to the requirements of the image biomarker standardization initiative.

[0121] CT scans at participating centers are performed on different scanner models, using different acquisition protocols and reconstruction kernels, which can introduce variability in radiographic features between scanners. To reduce these differences, all images are resampled and intensity normalized according to the Image Biomarker Standardization Initiative (IBSI) 16 recommendation before feature extraction. The image resampling involves resampling the images to isotropic voxels of 1×1×1 mm³; the intensity normalization is based on the grayscale distribution of the CT images to eliminate radiographic feature variability caused by different scanners, acquisition protocols, and reconstruction kernels.

[0122] S13. The tumor region and the ipsilateral normal renal parenchyma region are determined by manual segmentation in each phase of the enhanced CT scan image;

[0123] Manual segmentation was performed by importing all CT images into ITK-SNAP software (version 3.6.0). Two experienced radiologists (with 8 and 10 years of experience in abdominal imaging, respectively) drew regions of interest (ROIs) on axial images one by one, disregarding clinical and pathological findings. For each patient, two sets of ROIs were generated: (1) Tumor ROIs were manually drawn at each stage of the enhanced CT scan (cortical-medullary, nephrography, and excretory phases) to cover the entire visible renal tumor. (2) Normal renal parenchymal ROIs were located in the ipsilateral kidney, avoiding major blood vessels, the collecting system, and artifacts, serving as reference tissue. Figure 19 As shown, it illustrates the description of ROIs in CT images of normal kidney tissue and kidney tumors. All segments were reviewed and verified by a third senior radiologist to resolve discrepancies.

[0124] Manual segmentation was employed to ensure accurate depiction of both tumor and normal renal parenchyma, as accurate contours are crucial for the stability and reproducibility of features in radiological analysis. Although manual delineation is very time-consuming, it excludes perirenal fat, blood vessels, and collection system structures that could hinder quantitative feature extraction, thus providing a reliable reference standard for subsequent model development.

[0125] S14. Extract radiographic features from the tumor region and the ipsilateral normal renal parenchyma region respectively to obtain a tumor region feature set and an ipsilateral normal renal parenchyma region feature set. The radiographic features include shape-related features, first-order statistical features, gray-level run-length matrix features, gray-level co-occurrence matrix features, gray-level correlation matrix features, gray-level size region matrix features, and neighborhood gray-level difference matrix features.

[0126] Specifically, following the Image Biomarker Standardization Initiative (IBSI) guideline 16, radiographic features were extracted from the defined Region of Interest (ROI) using the Python (version 3.7.3) package Pyradiomics version 3.0. Two basic image filtering methods, Laplacian of Gaussian (LOG) and wavelet transform, were used to process the original CT images to obtain their radiographic features.

[0127] Extracting radiographic features from the tumor region and the ipsilateral normal renal parenchyma region specifically includes:

[0128] In each phase of CT images, the renal tumor region and the ipsilateral normal renal parenchyma region are segmented to obtain the corresponding region of interest for the tumor and the region of interest for the normal renal parenchyma.

[0129] An initial radiomics feature set was extracted from images of the tumor region of interest and the normal renal parenchyma region of interest at each stage.

[0130] S15. Preprocess the extracted radiological features; the preprocessing includes repetitive filtering, residual feature standardization, and multi-step dimensionality reduction:

[0131] The repeatability filtering refers to: in order to ensure repeatability and reduce observer bias, randomly selecting a subset of 50 cases to calculate the intra-group correlation coefficients (ICCs) within and between observers, and removing features with ICCs lower than 0.80; Figure 20 and Figure 21 The detailed distribution of computer centers across each feature category is listed.

[0132] The standardization of residual features refers to the standardization of the residual features after the repeatability filtering using Z-score normalization to eliminate scale differences.

[0133] The multi-step dimensionality reduction process includes: calculating the Pearson correlation coefficient among the remaining features, eliminating redundant features with an absolute correlation coefficient > 0.90 to minimize redundancy; screening features associated with postoperative renal function decline using univariate logistic regression, retaining features with smaller P-values ​​and / or higher AUCs; then, performing 10 cross-validations using LASSO logistic regression to determine the optimal regularization parameter (λ) within a range of 100 automatically generated logarithms. The optimal regularization parameter (λ_min) is determined according to the minimum criterion, and radiological features with non-zero coefficients at λ_min are retained to construct the final radiological features.

[0134] S16. Based on LASSO logistic regression, the feature subsets with the best prediction performance are selected from the preprocessed tumor region feature set and the ipsilateral normal renal parenchyma region feature set, respectively, to obtain the tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset. The tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset are fused to form a joint radiomics feature set.

[0135] Feature selection was performed using Lasso logistic regression to develop radiological features that retained the most information. Three independent logistic regression models were developed based on (1) radiological features of the tumor ROIS (intratumoral features), (2) radiological features of the normal renal parenchyma ROIS (renal features), and (3) a combined feature set of fused two regions (combined features). For each model, a radiomics score (RAD-SCORE) was calculated for each patient in the training and validation cohorts using a linear combination of selected features weighted by their respective lasso coefficients. Receiver operating characteristic (ROC) curve analysis was applied to assess the performance of each radiological feature in predicting early postoperative renal function decline. The area under the ROC curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for the training, internal validation, and external validation cohorts.

[0136] S17. Based on the coefficients determined by the LASSO logistic regression, the features in the joint radiomics feature set are linearly weighted and combined to calculate a comprehensive radiomics score for each patient in the dataset.

[0137] S18. Determine the preoperative clinical variable set through multivariate logistic regression. The preoperative clinical variable set includes the age, diabetes, estimated glomerular filtration rate, and kidney score of the patients in the dataset.

[0138] The kidney score is a quantitative score based on tumor complexity, used to assess tumor size, anatomical location depth, and distance from key kidney structures;

[0139] S19. Using the radiomics score and the preoperative clinical variable set as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained by logistic regression algorithm, and the preoperative nomogram prediction model is obtained after model validation.

[0140] The construction process of the preoperative-intraoperative combined nomogram prediction model includes:

[0141] S21. Obtain the combined radiological features and preoperative clinical variable set through S11 to S18;

[0142] S22. The intraoperative clinical variables for partial nephrectomy in patients with renal cell carcinoma were determined by multivariate logistic regression, wherein the intraoperative clinical variable was intraoperative ischemia time.

[0143] S23. Using the radiomics score, the preoperative clinical variable set, and the intraoperative clinical variable as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained using a logistic regression algorithm. After model validation, the preoperative-intraoperative combined nomogram prediction model is obtained.

[0144] Model validation involved evaluating the performance of each nomogram in the training cohort, internal validation cohort, and external validation cohorts 1–3. Patient discrimination ability was assessed using ROC curves and AUC. Evaluation was performed using calibration curves and the Hosmer-Lemeshow goodness-of-fit test. Clinical utility was further assessed through decision curve analysis (DCA), which estimated the net benefit of the nomogram compared to “all treatment” and “no treatment” strategies across a range of threshold probabilities. The total score generated by the nomogram for each patient was used to calculate the individual predicted probability of early renal function decline.

[0145] All statistical analyses in this invention were performed using R software (Version 4.1.0, R Foundation for Statistic Computing, Vienna, Austria) and SPSS (Version 26.0, IBM Corp., Armonk, NY, USA). Continuous variables were expressed as mean ± standard deviation (SD) or median and compared as appropriate using Student's t-test or Mann-Whitney U test. Categorical variables were expressed as frequencies and percentages and compared using chi-square test or Fisher's exact test. Univariate logistic regression was used for the first time to identify potential predictors of early postoperative renal function decline. Variables with p-values ​​<0.05 were included in the multivariate logistic regression model to identify independent risk factors. Odds ratios (ORs) and corresponding 95% confidence intervals (CIs) are reported. The model's discriminative power was assessed using AUC, and comparisons between models or between individual predictors were assessed using the Delong test. To account for multiple pairwise comparisons across models, the p-values ​​of the Delong test were adjusted using the Benjamini-Hochberg false discovery rate (FDR) procedure. To assess potential model overfitting and provide performance estimates for internal validation, 1000 iterations of guided resampling were performed on the preoperative nomogram prediction model and the combined preoperative-intraoperative nomogram prediction model in the training cohort. In each iteration, the model was refined, the ROC curve was recalculated, and a Bootstrap ROC curve was constructed using the average true positive rate to obtain the Bootstrap validation AUC value for 95% CIS. A p-value <0.05 was considered statistically significant. All tests were two-sided.

[0146] In this embodiment, the research flowchart is as follows: Figure 3As shown. A total of 1440 patients with localized renal cell carcinoma were included in the entire training, internal validation, and external validation process. Figure 4 Three external validation cohorts were summarized. The training cohort included 486 patients, the internal validation cohort included 277 patients, and external validation cohort 1 included 256 patients (center 1 [n=47], center 2 [n=65], center 3 [n=63], center 4 [n=50], center 5 [n=31]). External validation cohort 2 included 342 patients (center 6 [n=60], center 7 [n=71], center 8 [n=86], center 9 [n=52], center 10 [n=73]). External validation cohort 3 included 79 patients. In the training cohort, 87 patients (17.90%) experienced early renal function decline after PN. The internal validation cohort consisted of 46 patients, while the incidence of postoperative renal function decline in external validation cohorts 1-3 was 16.61% (16.80%), 16.80% (43 patients), 17.54% (60 patients), and 18.98% (15 patients), respectively.

[0147] A total of 3948 radiographic features were extracted from two tissue types: the tumor region and the ipsilateral normal renal parenchyma region. These were extracted from cortical-medullary, nephroscopic, and excretory phase CT scans, with 1316 ROIs per phase for each tissue type. After rigorous quality control, features with inter-observer / inter-observer ICCs <0.8 were excluded. The intratumoral ROIS retained 2603 phase-stratified radiographic features (cortical-medullary: 835; nephroscopic: 865; excretory: 903), while the normal renal ROIS retained 2665 radiographic features (cortical-medullary: 857; nephroscopic: 923; excretory: 885). Using Lasso regression, 12 radiographic features significantly correlated with postoperative renal function changes were selected from the tumor region, and 8 were selected from the ipsilateral normal renal parenchyma region. 18 features were selected from the combined analysis of the two tissue types (see [link to Lasso regression analysis]). Figure 5 Subsequently, logistic regression analysis was used to establish three radiological features: renal radiological features (see...). Figure 6 (A) and (B) in the text), intratumoral radiological features (see [reference]). Figure 6 (C) and (D) in the text) and combined radiological features (see Figure 6 (E) and (F) in the text.

[0148] ROC analysis showed that the AUC values ​​for renal radiographic signs in the training group, internal validation group, and external validation groups 1–3 were 0.733 (0.712–0.815), 0.712 (0.684–0.865), 0.703 (0.682–0.872), 0.732 (0.623–0.823), and 0.725 (0.698–0.887), respectively. The AUC values ​​for intratumoral radiographic signals were 0.790 (0.745–0.885), 0.715 (0.693–0.886), 0.725 (0.704–0.893), 0.740 (0.652–0.831), and 0.767 (0.721–0.900). The combined radiographic features showed good performance, with AUC values ​​of 0.845 (0.809-0.876), 0.741 (0.685-0.791), 0.822 (0.769-0.867), 0.761 (0.672-0.836), and 0.802 (0.697-0.883), respectively. Figure 7 As shown, the sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of all three features are further detailed. These findings indicate that combined radiological features are superior to renal and intratumoral features in predicting postoperative renal function decline, making it a promising tool worthy of further clinical validation and application.

[0149] Univariate analysis showed that age, BMI, diabetes mellitus, hypertension, EGFR, renal score, ischemic time, intraoperative colloid volume, and radiographic characteristics were significantly associated with early renal function decline after renal cell carcinoma surgery (P<0.05). Multivariate analysis showed that age (P<0.001), diabetes mellitus (P<0.001), pre-EGFR (P<0.001), renal score (P<0.001), ischemic time (P<0.001), and radiographic characteristics (P<0.001) were independent predictors of early renal function decline after RCC surgery. Figure 8 As shown.

[0150] Preoperative nomograms of early renal function decline after PN in RCC patients were based on logistic regression analysis of five independent preoperative factors (age, diabetes, pre-EGFR, renal score, and radiological characteristics) in the training cohort. Figure 9The ROC analysis showed that the AUC of the preoperative nomogram in the training cohort was 0.952 (95% CI, 0.916–0.960), significantly better than age (AUC 0.686, 95% CI, 0.643–0.727, Delong test, Z=7.397, P<0.001), diabetes (AUC 0.694, 95% CI, 0.650–0.734, Delong test, Z=9.465, P<0.001), pre-EGFR (AUC 0.753, 95% CI, 0.712–0.791, Delong test, Z=6.145, P<0.001), and renal score (AUC 0.771, 95% CI, 0.732–0.808, Delong test, Z=4.559, P<0.001). Internal validation of the Bootstrap model confirmed good model stability, with an average Bootstrap AUC of 0.948 (95% CI, 0.927–0.972) generated in the training queue. Figure 10 As shown in (A). The Hosmer-Lemeshow goodness-of-fit test indicates that the nomogram shows good calibration (χ²=5.282, P=0.727), as... Figure 9 As shown in (C). To further evaluate the clinical applicability of nomogram, DCA was performed at different probability thresholds to quantify net benefit. The "all" curve represents all patients who received clinical intervention, while the "no" curve represents no intervention. The red curve represents the decision curve of the nomogram model. At the probability threshold of 0-1.0, the net benefit of nomogram is higher than that of the "all" and "no" curves, indicating its high clinical applicability within this range, such as... Figure 9 As shown in (D).

[0151] In the internal validation cohort, the AUC of nomogram was 0.909 (95% CI, 0.869–0.941), significantly better than age (AUC 0.753, 95% CI, 0.698–0.803, Delong test, Z=3.998, P<0.001), diabetes (AUC 0.689, 95% CI, 0.631–0.743, Delong test, Z=5.638, P<0.001), pre-EGFR (AUC 0.705, 95% CI, 0.647–0.758, Delong test, Z=6.795, P<0.001), and renal score (AUC 0.711, 95% CI, 0.654–0.764, Delong test, Z=5.225, P<0.001). The Hosmer-Lemeshow test further confirmed the well calibration (χ2=6.155, P=0.630), as... Figure 9As shown in (F). DCA shows that, in the internal validation cohort, within the probability threshold range of 0–1.0, the clinical-radiological nomogram achieved better results in guiding clinical decision-making, as shown in (F). Figure 9 As shown in (G).

[0152] In external validation cohort 1, ROC analysis showed that the AUC of the nomogram was 0.931 (95% CI, 0.892–0.959). Figure 11 (A) and Figure 12 As shown. Furthermore, goodness-of-fit tests confirmed the good calibration of the nomograms in external validation cohort 1 (χ²=1.695, P=0.989), as... Figure 11 As shown in (B). DCA analysis showed that, in external validation cohort 1, the clinical radiology nomogram also provided better decision support in the 0-1.0 probability threshold range, as Figure 11 As shown in (C). In external validation cohorts 2 and 3, the AUCs obtained from Roc analysis were 0.926 (95% CI, 0.862–0.967) and 0.914 (95% CI, 0.829–0.965), respectively. Hosmer-Lemeshow goodness-of-fit tests (χ²=2.955, P=0.937 and χ²=4.576, P=0.802) and DCA also showed performance consistent with the cohorts described above, such as Figure 11 As shown in (D) to (I).

[0153] Logistic regression analysis was applied based on six preoperative-intraoperative factors: age, diabetes, pre-EGFR, renal score, ischemic time, and radiological characteristics in the training cohort, such as... Figure 13As shown in (A), a preoperative-intraoperative nomogram was established to predict early renal function decline after PN in RCC patients. ROC analysis showed that the AUC of the preoperative-intraoperative nomogram for the training cohort was 0.962 (95% CI, 0.941–0.977). Significantly superior to age (AUC 0.686, 95% CI, 0.643–0.727, Delong test, Z=8.678, P<0.001), diabetes (AUC 0.694, 95% CI, 0.650–0.734, Delong test, Z=10.071, P<0.001), pre-EGFR (AUC 0.753, 95% CI, 0.712–0.791, Delong test, Z=6.762, P<0.001), renal score (AUC 0.771, 95% CI, 0.732–0.808, Delong test, Z=6.790, P<0.001), and ischemic time (AUC 0.789, 95% CI, 0.750–0.825, Delong test, Z=6.352) The preoperative-intraoperative combined nomogram prediction model showed strong internal robustness, with a mean bootstrapping AUC of 0.960 (95% CI, 0.947–0.990) in the training cohort. Figure 10 As shown in (B). The Hosmer-Lemeshow test showed that the nomogram had good calibrability (χ2=7.066, P=0.530), as Figure 13 As shown in (C). In the training cohort, DCA demonstrated high clinical applicability of the nomogram within the probability threshold of 0–1.0, as... Figure 13 As shown in (D).

[0154] In the internal validation cohort, the preoperative-intraoperative nomogram showed good discriminative performance, with an AUC of 0.926 (95% CI, 0.888–0.954), significantly outperforming all individual predictive indices, such as... Figure 13 (E) shown and Figure 14 As shown in the figure. Calibration analysis shows a good agreement between the predicted and observed results, as indicated by the Hosmer-Lemeshow goodness-of-fit test (χ²=2.608, P=0.957). Figure 13 As shown in (F). Furthermore, compared to all-treatment and no-treatment strategies, DCA demonstrated a consistently higher net clinical benefit across a broad range of threshold probabilities, such as... Figure 13 As shown in (G).

[0155] In the three external validation cohorts, the nomogram maintained robust predictive performance: AUCs of 0.947 (95% CI, 0.913–0.971) for external validation cohort 1, 0.947 (95% CI, 0.888–0.980) for external validation cohort 2, and 0.933 (95% CI, 0.854–0.977) for external validation cohort 3. Figure 15 (A), (D), (G) and Figure 14 As shown. The calibration curves also show a satisfactory model fit, supported by non-significant Hosmer-Lemeshow statistics in external validation cohort 1 (χ²=4.401, P=0.819), external validation cohort 2 (χ²=2.954, P=0.937), and external validation cohort 3 (χ²=6.625, P=0.578), as shown. Figure 15 As shown in (B), (E), and (H). DCA demonstrated strong clinical utility in all external cohorts, with good net benefit in the 0-1.0 threshold range, as... Figure 15 As shown in (C), (F), and (I).

[0156] Subgroup analyses were performed to assess the stability of model performance across clinically relevant levels in the training cohort, such as Figure 16As shown in the figure. Preoperative line plot AUCs were 0.942 (95% CI, 0.915-0.969) and 0.858 (95% CI, 0.795-0.920), respectively; AUCs for tumors ≤4cm and >4cm were 0.944 (95% CI, 0.915-0.972) and 0.966 (95% CI, 0.937-0.988), respectively; AUCs for pre-baseline EGFR ≤90 and >90 ml / min / 1.73m² were 0.934 (95% CI, 0.900-0.963) and 0.949 (95% CI, 0.915-0.978), respectively; and AUCs for RAPN and LPN were 0.966 (95% CI, 0.940-0.988) and 0.858 (95% CI, 0.795-0.920), respectively. The preoperative-intraoperative nomogram showed comparable or slightly high discrimination between the same layers. The AUCs for patients ≤65 years and >65 years were 0.969 (95% CI, 0.946-0.989) and 0.879 (95% CI, 0.820-0.930), respectively. The AUCs for tumors ≤4cm and >4cm were 0.952 (95% CI, 0.924-0.977) and 0.981 (95% CI, 0.960-0.996), respectively. The AUCs for pre-baseline EGFR ≤90 and >90 ml / min / 1.73m2 were 0.954 (95% CI, 0.924-0.979) and 0.968 (95% CI, 0.939-0.990), and 0.971 (95% CI), respectively.

[0157] To determine the discriminative power of the nomogram prediction model of this invention within existing literature, a comparative benchmark analysis was performed against previously reported clinical and renal measurement-based prediction models, such as... Figure 17As shown. Early models—including those proposed by Martini et al., Bertolo et al., and Mari et al.—relied primarily on demographic factors, baseline renal function, comorbidity profiles, or tumor complexity scoring systems, typically yielding AUC or C-index values ​​in the range of 0.710–0.820, used to predict postoperative renal function decline or CKD progression. In contrast, the two radiomics-enhanced nomograms developed in this invention exhibit significantly superior discriminative power. The preoperative nomogram prediction model produced an AUC of 0.952 in the training cohort and 0.909–0.931 in multiple external validation cohorts, while the combined preoperative-intraoperative nomogram prediction model further improved performance, producing an AUC of 0.962 in the training cohort and 0.926–0.947 in the validation cohort. Furthermore, both sensitivity and specificity exceeded those previously reported for tools used only in clinical settings, highlighting enhanced classification reliability. In summary, these results highlight the incremental predictive value of combining radiological features extracted from tumors and ipsilateral renal parenchyma with conventional clinicopathological variables, resulting in substantial improvements in model generalizability and superior risk stratification compared to existing methods.

[0158] To demonstrate the clinical applicability of the proposed nomogram model, two representative hypothetical cases were evaluated. One case involved a 75-year-old diabetic patient with a preoperative EGFR of 120 mL / min / 1.73 mcg. 2 The renal score was 7, and the radiomics score was 10. The allocation coefficients of the preoperative nomogram prediction model in the example are as follows: Figure 22 As shown, the following point allocations were used on the preoperative nomogram prediction model: age 35 points; diabetes 32.5 points; pre-EGFR 25 points; renal score 21 points; radiological signature 55 points. Therefore, the total score was 169.5, corresponding to a predicted probability of 0.54 for early renal function decline. From a clinical perspective, a risk of 0.54 indicates a moderately increased likelihood of early renal function decline compared to the average postoperative patient. In these patients, surgeons can maintain guideline-based follow-up intervals but adopt a more cautious perioperative strategy, including prioritizing nephron-sparing techniques where technically feasible and carefully optimizing modifiable risk factors (such as blood pressure, glycemic control, and perioperative hydration). In the first year postoperatively, clinicians can consider adding one or more ad hoc EGFR measurements between routine visits to detect early, clinically relevant declines, such as... Figure 18 As shown in (A).

[0159] For the same patient, the allocation coefficients of the preoperative-intraoperative combined nomogram prediction model in the examples are as follows: Figure 23As shown, when the intraoperative ischemic time of 30 minutes was included in the preoperative-intraoperative combined nomogram prediction model, the assigned scores were: age 34 points; diabetes 30 points; pre-EGFR 25 points; renal score 19 points; radiological signature 55 points; ischemic time 55 points. The total score increased to 214 points, corresponding to a prediction probability of 0.85. Conversely, a prediction risk of 0.85 represents a very high probability of early renal function decline. For such patients, it is reasonable to implement a more intensive renal protection and monitoring strategy, including close monitoring of EGFR at guideline-recommended time points (3, 6, 12, 18, and 24 months), additional early examinations as needed, early consideration of nephrology referral when a downward trend is observed, strict avoidance of nephrotoxic drugs, and more aggressive management of cardiovascular and metabolic risk factors such as Figure 18 As shown in (B).

[0160] Decision curve analysis showed that, within a wide threshold probability range of 0 to 1.0, both nomogram models provided greater net benefits than the full treatment and no treatment strategies. Within this framework, a predicted risk in the range of 0.5 can be considered moderately elevated, requiring careful optimization and slightly closer monitoring, while a risk in the range of 0.8–0.9 requires significantly enhanced postoperative monitoring and renal protection management.

[0161] This invention is the first to construct and validate a nomogram that combines CT radiographic features of tumor and normal kidney tissue with preoperative and intraoperative clinical factors for high-precision prediction of early renal function decline after nephrectomy (PN). Logistic regression analysis was used to establish a preoperative nomogram prediction model based solely on preoperative variables and a combined preoperative-intraoperative nomogram prediction model that comprehensively considers preoperative and intraoperative factors. ROC analysis showed that the nomogram outperformed individual clinical and surgical variables in predictive performance. The Hosmer-Lemeshow trial demonstrated good calibrability of the model. Furthermore, DCA provides supporting evidence for the clinical application of nomograms, demonstrating their feasibility in promoting beneficial clinical decision-making.

[0162] Example 2

[0163] Based on the same inventive concept, this application also provides an apparatus corresponding to the method in Embodiment 1, as detailed in Embodiment 2.

[0164] This embodiment provides a method for predicting the risk of early renal function decline after partial nephrectomy in patients with renal cell carcinoma. The method involves obtaining clinical parameters of the patient through a prediction model; assigning risk scores to the clinical parameters; calculating the total risk score based on the risk scores of each clinical parameter; and determining and visualizing the predicted probability of early renal function decline after partial nephrectomy based on the total risk score.

[0165] The prediction model includes a preoperative nomogram prediction model and / or a combined preoperative and intraoperative nomogram prediction model; the clinical parameters corresponding to the preoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, and radiomics score; the clinical parameters corresponding to the combined preoperative and intraoperative nomogram prediction model include age, history of diabetes, estimated preoperative glomerular filtration rate, kidney score, radiomics score, and intraoperative ischemic time.

[0166] The process of constructing the preoperative normogram prediction model includes:

[0167] S11. Screen the patients in the dataset and obtain the preoperative enhanced CT scan images of each patient in the dataset. The enhanced CT scan images include the cortico-medullary phase, the renal angiography phase, and the excretion phase.

[0168] S12. Standardize the CT enhanced scan image, including resampling the image into isotropic voxels and normalizing the intensity according to the requirements of the image biomarker standardization initiative.

[0169] S13. The tumor region and the ipsilateral normal renal parenchyma region are determined by manual segmentation in each phase of the enhanced CT scan image;

[0170] S14. Extract radiographic features from the tumor region and the ipsilateral normal renal parenchyma region respectively to obtain a tumor region feature set and an ipsilateral normal renal parenchyma region feature set. The radiographic features include shape-related features, first-order statistical features, gray-level run-length matrix features, gray-level co-occurrence matrix features, gray-level correlation matrix features, gray-level size region matrix features, and neighborhood gray-level difference matrix features.

[0171] S15. Preprocess the extracted radiological features;

[0172] S16. Based on LASSO logistic regression, the feature subsets with the best prediction performance are selected from the preprocessed tumor region feature set and the ipsilateral normal renal parenchyma region feature set, respectively, to obtain the tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset. The tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset are fused to form a joint radiomics feature set.

[0173] S17. Based on the coefficients determined by the LASSO logistic regression, the features in the joint radiomics feature set are linearly weighted and combined to calculate a comprehensive radiomics score for each patient in the dataset.

[0174] S18. Determine the preoperative clinical variable set through multivariate logistic regression. The preoperative clinical variable set includes the age, diabetes, estimated glomerular filtration rate, and kidney score of the patients in the dataset.

[0175] S19. Using the radiomics score and the preoperative clinical variable set as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained by logistic regression algorithm, and the preoperative nomogram prediction model is obtained after model validation.

[0176] The construction process of the preoperative-intraoperative combined nomogram prediction model includes:

[0177] S21. Obtain the combined radiological features and preoperative clinical variable set through S11 to S18;

[0178] S22. The intraoperative clinical variables for partial nephrectomy in patients with renal cell carcinoma were determined by multivariate logistic regression, wherein the intraoperative clinical variable was intraoperative ischemia time.

[0179] S23. Using the radiomics score, the preoperative clinical variable set, and the intraoperative clinical variable as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained using a logistic regression algorithm. After model validation, the preoperative-intraoperative combined nomogram prediction model is obtained.

[0180] Furthermore, the image resampling involves resampling the image to an isotropic voxel of 1×1×1 mm³; the intensity normalization is based on the gray value distribution of the CT image to eliminate the variability in radiographic features caused by different scanners, acquisition protocols, and reconstruction kernels.

[0181] Furthermore, S13 specifically involves: in the ITK-SNAP software, at least two radiologists manually segment the regions of interest (ROIs) of each phase of the standardized CT enhanced scan images to determine the tumor region and the ipsilateral normal renal parenchyma region, and the segmentation results are reviewed and verified by a third senior radiologist.

[0182] Furthermore, in S14, extracting radiological features from the tumor region and the ipsilateral normal renal parenchyma region specifically includes:

[0183] In each phase of CT images, the renal tumor region and the ipsilateral normal renal parenchyma region are segmented to obtain the corresponding region of interest for the tumor and the region of interest for the normal renal parenchyma.

[0184] An initial radiomics feature set was extracted from images of the tumor region of interest and the normal renal parenchyma region of interest at each stage.

[0185] Furthermore, the preprocessing in S15 includes repetitive filtering, residual feature standardization, and multi-step dimensionality reduction:

[0186] The repetitive filtering refers to: calculating the intra-group correlation coefficients within and between observers, and removing features with intra-group correlation coefficients lower than 0.80;

[0187] The standardization of residual features refers to the standardization of the residual features after the repeatability filtering using Z-score normalization.

[0188] The multi-step dimensionality reduction process includes: calculating the Pearson correlation coefficient among the remaining features and removing redundant features with an absolute correlation coefficient > 0.90; screening features related to postoperative renal function decline through univariate logistic regression; determining the optimal regularization parameter through LASSO regression with 10 cross-validations and retaining features with non-zero coefficients at the optimal regularization parameter.

[0189] Since the model described in Embodiment 2 of this invention is an apparatus used to implement the method of Embodiment 1 of this 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 this invention, and therefore will not be repeated here. All apparatuses used in the method of Embodiment 1 of this invention fall within the scope of protection of this invention.

[0190] 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 predictive model for the risk of early renal function decline after partial nephrectomy in patients with renal cell carcinoma, characterized in that: Used to obtain clinical parameters of the patient to be predicted; the clinical parameters are integrated and assigned to obtain the risk score of each clinical parameter; Calculate the total risk score based on the risk scores of each clinical parameter; The predicted probability of early renal function decline after partial nephrectomy in this patient is determined and visualized based on the total risk score. The early postoperative renal function decline is defined as a decrease in EGFR of ≥25% from preoperative baseline within 3-24 months after partial nephrectomy. The prediction model includes a preoperative nomogram prediction model and / or a combined preoperative and intraoperative nomogram prediction model. The clinical parameters corresponding to the preoperative nomogram prediction model include age, history of diabetes, preoperative estimated glomerular filtration rate, renal score, and radiomics score. The clinical parameters corresponding to the combined preoperative and intraoperative nomogram prediction model include age, history of diabetes, preoperative estimated glomerular filtration rate, renal score, radiomics score, and intraoperative ischemic time. The process of constructing the preoperative normogram prediction model includes: S11. Screen the patients in the dataset and obtain the preoperative enhanced CT scan images of each patient in the dataset. The enhanced CT scan images include the cortico-medullary phase, the renal angiography phase, and the excretion phase. S12. Standardize the CT enhanced scan image, including resampling the image into isotropic voxels and normalizing the intensity according to the requirements of the image biomarker standardization initiative. The image resampling involves resampling the image to an isotropic voxel of 1×1×1 mm³; the intensity normalization is based on the gray value distribution of the CT image to eliminate the variability in radiographic features caused by different scanners, acquisition protocols, and reconstruction kernels. S13. The tumor region and the ipsilateral normal renal parenchyma region are determined by manual segmentation in each phase of the enhanced CT scan image; S14. Extract radiographic features from the tumor region and the ipsilateral normal renal parenchyma region respectively to obtain a tumor region feature set and an ipsilateral normal renal parenchyma region feature set. The radiographic features include shape-related features, first-order statistical features, gray-level run-length matrix features, gray-level co-occurrence matrix features, gray-level correlation matrix features, gray-level size region matrix features, and neighborhood gray-level difference matrix features. S15. Preprocess the extracted radiological features; S16. Based on LASSO logistic regression, the feature subsets with the best prediction performance are selected from the preprocessed tumor region feature set and the ipsilateral normal renal parenchyma region feature set, respectively, to obtain the tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset. The tumor region feature subset and the ipsilateral normal renal parenchyma region feature subset are fused to form a joint radiomics feature set. S17. Based on the coefficients determined by the LASSO logistic regression, the features in the joint radiomics feature set are linearly weighted and combined to calculate a comprehensive radiomics score for each patient in the dataset. S18. Determine the preoperative clinical variable set through multivariate logistic regression. The preoperative clinical variable set includes the age, diabetes, estimated glomerular filtration rate, and kidney score of the patients in the dataset. S19. Using the radiomics score and the preoperative clinical variable set as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained by logistic regression algorithm, and the preoperative nomogram prediction model is obtained after model validation. The construction process of the preoperative-intraoperative combined nomogram prediction model includes: S21. Obtain the combined radiological features and preoperative clinical variable set through S11 to S18; S22. The intraoperative clinical variables for partial nephrectomy in patients with renal cell carcinoma were determined by multivariate logistic regression, wherein the intraoperative clinical variable was intraoperative ischemia time. S23. Using the radiomics score, the preoperative clinical variable set, and the intraoperative clinical variable as feature variables, and the label of whether the early postoperative renal function declines as the target variable, the model is trained using a logistic regression algorithm. After model validation, the preoperative-intraoperative combined nomogram prediction model is obtained.

2. The prediction model according to claim 1, characterized in that: Specifically, S13 involves at least two radiologists manually segmenting the regions of interest (ROIs) of each phase of the standardized CT enhanced scan images in ITK-SNAP software to determine the tumor region and the ipsilateral normal renal parenchyma region, and the segmentation results are reviewed and verified by a third senior radiologist. In step S14, extracting radiological features from the tumor region and the ipsilateral normal renal parenchyma region specifically includes: In each phase of CT images, the renal tumor region and the ipsilateral normal renal parenchyma region are segmented to obtain the corresponding region of interest for the tumor and the region of interest for the normal renal parenchyma. An initial set of radiomics features was extracted from images of the tumor region of interest and the normal renal parenchyma region of interest at each stage. The preprocessing in S15 includes repetitive filtering, residual feature normalization, and multi-step dimensionality reduction. The repetitive filtering refers to: calculating the intra-group correlation coefficients within and between observers, and removing features with intra-group correlation coefficients lower than 0.80; The standardization of residual features refers to the standardization of the residual features after the repeatability filtering using Z-score normalization. The multi-step dimensionality reduction process includes: calculating the Pearson correlation coefficient among the remaining features and removing redundant features with an absolute correlation coefficient > 0.90; screening features related to postoperative renal function decline through univariate logistic regression; determining the optimal regularization parameter through LASSO regression with 10 cross-validations and retaining features with non-zero coefficients at the optimal regularization parameter.

Images