A prediction method and prediction model for MIT family translocation renal cancer

Abstract

The present application relates to a kind of prediction method and prediction model of MIT family translocation renal carcinoma, for the problem of insufficient accuracy of existing renal carcinoma subtype prediction method, the method is by obtaining the initial image feature set and clinical characteristics (including age and AJCC clinical stage) of patient, tumor and tumor adjacent kidney tissue image feature set are obtained by screening, corresponding score is obtained by weighted calculation, then MIT-RCC probability is output by prediction model in combination with clinical characteristics.Prediction model is nomogram model, the present application can assist clinician to quickly and accurately identify MIT family translocation renal carcinoma, improve diagnostic efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary prediction technology, and in particular to a prediction method and prediction model for MIT family translocation renal cell carcinoma. Background Technology

[0002] Renal cell carcinoma (RCC) is the most common primary malignant tumor of the kidney, accounting for approximately 85-90% of all kidney malignancies. It ranks third among urinary tract malignancies, and its incidence and mortality rates are increasing annually. Global statistics show that over 400,000 new cases of kidney cancer are diagnosed each year, and nearly 180,000 patients die from the disease. The majority of these patients are over 60 years old, and two-thirds are male. Microphthalmia Transcription Factor (MIT) family translocation renal cell carcinoma (tRCC) is a rare subtype of kidney cancer characterized by chromosomal translocations involving the fusion of transcription factor E3 (TFE3) or EB (TFEB) genes (located at chromosomal loci Xp11.2 and 6p21, respectively) with various partner genes. Because tRCCs with TFE3 and TFEB gene fusions (XP11.2 translocation / TFE3 gene fusion-related renal cell carcinoma and t(6;11)(p21;q12) / TFEB gene fusion-related renal cell carcinoma) share many clinical, histopathological, and genetic similarities, the 2013 International Society of Urological Pathology (ISUP) Vancouver classification grouped these two subtypes into a single category called MIT Family Translocation Renal Cell Carcinoma (MIT-RCC). The 2012 ISUP consensus meeting revised the 2004 classification of renal tumors, resulting in the latest 2016 WHO classification of renal tumors, in which MIT-RCC was designated as a new clinically significant subtype of renal cell carcinoma.The most common types of renal cell carcinoma are clear cell renal carcinoma (ccRCC), chromophobe renal cell carcinoma (ChRCC), and papillary renal cell carcinoma (pRCC). Among these, ccRCC is the most common, while MIT-RCC is relatively rare in clinical practice and has a high degree of malignancy. Its imaging features overlap with those of ccRCC and pRCC. Currently, biopsy and other pathological examinations are still the gold standard for preoperative diagnosis of renal cell carcinoma. However, renal tissue is fragile, and complications such as renal hemorrhage and needle tract metastasis are very likely to occur after biopsy. Therefore, routine preoperative biopsy is generally not recommended for patients with renal tumors.

[0003] In 2012, Dutch scholar Lambin proposed the concept of radiomics. Radiomics refers to the rapid extraction of quantitative features from imaging modalities such as computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and ultrasound (US). Using statistical or machine learning methods, the most valuable radiomic features are screened out and then correlated with lesion features to build predictive models, which helps clinicians make treatment decisions. The principle behind radiomics is that medical images contain information reflecting the potential pathophysiology of diseases, and quantitative image analysis can reveal these relationships. Radiomics can be combined with other data such as genetic and clinical information to help with diagnosis, treatment assessment, and prognosis prediction. Its main advantages include: (1) almost all cancer patients undergo digital imaging, providing a large source of potential radiomics data; (2) radiomics provides non-invasive assessment of the temporal and spatial changes of tumors; and (3) radiomics is routinely performed throughout the treatment process, thus enabling continuous monitoring. The radiomics workflow includes steps such as image data acquisition, region of interest (ROI) segmentation, feature extraction, feature selection, model building, and evaluation.

[0004] Currently, CT scans are widely used for preoperative examination and clinical staging of patients with recurrent tumors (RCC). However, a radiomics model combining CT texture and non-texture features has not yet been developed to predict MIT-RCC. Therefore, this study aims to develop and compare radiomics models based on CT image features to achieve individualized prediction of MIT-RCC. Radiomics technology compensates for the shortcomings of traditional imaging, transforming traditional visual image information into quantifiable objective data for in-depth analysis. Simultaneously, by combining high-throughput extraction and analysis of imaging and clinical data, comprehensive tumor information can be extracted, thereby achieving objective prediction of tumor classification. Accurate preoperative imaging diagnosis is crucial for helping clinicians develop scientific treatment plans, providing important support for selecting appropriate surgical methods or perioperative treatment plans for MIT-RCC patients. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a prediction method and prediction model for MIT family translocation renal cell carcinoma. By integrating CT radiomics features (tumor and adjacent kidney tissue) with clinical data, a nomograph model for predicting MIT family translocation renal cell carcinoma (MIT-RCC) is constructed to achieve accurate and non-invasive prediction of MIT-RCC.

[0006] In a first aspect, the present invention provides a method for predicting MIT family translocation renal cell carcinoma, comprising the following steps: Obtain the initial radiomics feature set and clinical features of the patient to be predicted; the clinical features include age and AJCC clinical stage. The initial radiomics feature set is screened to obtain the screened tumor radiomics feature set and the screened peritumoral kidney tissue radiomics and clinical features set; The kidney tumor radiomics score and the adjacent kidney tissue radiomics score of the patient to be predicted were determined by weighted calculation based on the screened tumor radiomics feature set and the screened adjacent kidney tissue radiomics feature set. The predictive model draws a vertical line on each corresponding coordinate axis based on the renal tumor radiomics score and the radiomics score of the adjacent kidney tissue, age, and AJCC clinical stage to determine the individual scores. All individual scores are added together, the total score is found on the cumulative score coordinate axis, and a descending line is drawn to assess the probability of the patient having MIT-RCC before outputting the results. The construction process of the prediction model includes: Data Acquisition: Patients were screened from a multi-center medical database to obtain CT images and clinicopathological data of the patients in the dataset; the clinicopathological data included age, gender, body mass index, tumor location, history of diabetes, history of hypertension, AJCC clinical stage, and pathological tumor subtype; Data preprocessing: The CT images are normalized in grayscale and slice thickness is made consistent to obtain preprocessed CT images; Radiomics feature extraction: The preprocessed CT images were segmented into tumor ROI and adjacent kidney tissue ROI; radiomics features of the tumor ROI and adjacent kidney tissue ROI were extracted, including shape features, first-order histogram features, gray-level run length matrix features, gray-level co-occurrence matrix features, gray-level dependence matrix features, and adjacent gray-level difference matrix. Screening of radiomics features: Radiomics features with excellent reproducibility and significant correlation with MTT-RCC are screened to generate a final feature set, which is used as the input to the model; Feature weight determination: Through univariate and multivariate logistic regression analysis, the clinical features that are significantly associated with MIT-RCC are determined from the clinicopathological data, and the association strength between tumor radiomics features, adjacent kidney tissue radiomics features, age and AJCC stage and MIT-RCC is quantified; a preset weight is assigned to each feature through regression coefficients; Model construction: A nomograph model is constructed based on the preset weights, and the weight allocation is optimized through cross-validation to ensure generalization ability; Model training and validation: Central hierarchical validation was adopted, dividing the final feature set into a training set and an independent validation set. The training set and the independent validation set were respectively input into the nomograph model, and evaluated by ROC curve, calibration curve and decision curve to confirm the probability prediction accuracy and clinical applicability of the nomograph model; the model goodness of fit was verified by Hosmer-Lemeshow test to obtain the prediction model.

[0007] Secondly, this invention provides a predictive model for MIT family translocation renal cell carcinoma, comprising: The kidney tumor radiomics scoring input module is used to receive the kidney tumor radiomics scores of the patients to be predicted. The radiomics scoring input module for adjacent kidney tissue is used to receive the radiomics scores of adjacent kidney tissue of the patient to be predicted. The clinical feature input module is used to receive clinical features, including age and AJCC clinical stage. The individual line plotting module plots a vertical line on each corresponding coordinate axis based on the renal tumor radiomics score, the radiomics score of adjacent kidney tissue, age, and AJCC clinical stage to determine the individual score. The calculation module adds up all the individual scores and finds the total score on the cumulative score axis; The total score line drawing module draws a descending line based on the total score. The probability assessment module is used to assess and output the probability of a patient being diagnosed with MIT family translocation renal cell carcinoma. The construction process of the prediction model includes: Data Acquisition: Patients were screened from a multi-center medical database to obtain CT images and clinicopathological data of the patients in the dataset; the clinicopathological data included age, gender, body mass index, tumor location, history of diabetes, history of hypertension, AJCC clinical stage, and pathological tumor subtype; Data preprocessing: The CT images are normalized in grayscale and slice thickness is made consistent to obtain preprocessed CT images; Radiomics feature extraction: The preprocessed CT images were segmented into tumor ROI and adjacent kidney tissue ROI; radiomics features of the tumor ROI and adjacent kidney tissue ROI were extracted, including shape features, first-order histogram features, gray-level run length matrix features, gray-level co-occurrence matrix features, gray-level dependence matrix features, and adjacent gray-level difference matrix. Screening of radiomics features: Radiomics features with excellent reproducibility and significant correlation with MTT-RCC are screened to generate a final feature set, which is used as the input to the model; Feature weight determination: Through univariate and multivariate logistic regression analysis, the clinical features that are significantly associated with MIT-RCC are determined from the clinicopathological data, and the association strength between tumor radiomics features, adjacent kidney tissue radiomics features, age and AJCC stage and MIT-RCC is quantified; a preset weight is assigned to each feature through regression coefficients; Model construction: A nomograph model is constructed based on the preset weights, and the weight allocation is optimized through cross-validation to ensure generalization ability; Model training and validation: Central hierarchical validation was adopted, dividing the final feature set into a training set and an independent validation set. The training set and the independent validation set were respectively input into the nomograph model, and evaluated by ROC curve, calibration curve and decision curve to confirm the probability prediction accuracy and clinical applicability of the nomograph model; the model goodness of fit was verified by Hosmer-Lemeshow test to obtain the prediction model.

[0008] Furthermore, in the final feature set, The tumor ROI has 22 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.2.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.2.0 mm. 3D grayscale run length matrix. Run length non-uniformity; log.σ.3.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.5.0 mm. 3D grayscale dependency matrix_dependency variance; Original grayscale dependency matrix and dependency variance; Original grayscale dependency matrix with low grayscale emphasis; wavelet.HHH_grayscale dependency matrix_dependency variance; Wavelet.HHH_Grayscale region matrix_Large area low grayscale emphasis; Wavelet.HLH_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.HLL_Grayscale Dependency Matrix_High Dependency High Grayscale Emphasis; Wavelet.HLL_grayscale region matrix_large area low grayscale emphasis; Wavelet.LHH_Gray-level Co-occurrence Matrix_Difference Entropy; wavelet.LHL_grayscale dependency matrix_dependency variance; wavelet.LLH_grayscale run length matrix_run variance; Wavelet.LLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.LLL_Gray-level size region matrix_Gray-level non-uniformity normalization; The adjacent kidney tissue ROI possessed 24 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix with large area high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix and its non-uniformity. log.σ.5.0 mm. 3D_grayscale co-occurrence matrix_maximum probability; log.σ.5.0 mm. 3D_grayscale size region matrix_large area emphasis; log.σ.5.0 mm. 3D grayscale region matrix with region variance; Original first-order feature 10 percentile; Original first-order feature minimum; Original grayscale region matrix - non-uniformity of region size; Wavelet.HHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HHH_Grayscale region matrix_Large area high grayscale emphasis; wavelet.HHH_grayscale region matrix_region variance; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.HLL_First-order feature_energy; Wavelet.HLL_First-order feature_Total energy; Wavelet.LHH_First-order feature_energy; Wavelet.LHH_First-order feature_Total energy; Wavelet.LHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.LHH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.LHL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.LLL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Grayscale Region Matrix_Region Variance

[0009] Furthermore, the specific process of radiomics feature extraction is as follows: In ITK-SNAP software, three radiologists with more than 5 years of diagnostic experience and at least the title of attending physician manually delineate the tumor region in the preprocessed CT plain scan and arterial phase images to obtain the tumor ROI and the adjacent kidney tissue ROI. The region of interest is manually drawn on each cross-sectional image of the renal cell carcinoma to obtain the region of interest. The segmentation results are reviewed and verified by two radiologists with more than 10 years of experience in diagnosing urinary tract tumors and at least the title of associate senior physician. If the opinions of the two physicians are inconsistent, another radiologist with 20 years of experience in diagnosing imaging and at least the title of senior physician participates in the joint discussion to determine the location of the region of interest. The tumor ROI and the adjacent kidney tissue ROI are fused to form a three-dimensional region of interest.

[0010] Furthermore, the screening of radiomics features includes: calculating the intragroup correlation coefficient (ICC) to assess the repeatability of features; screening features with high repeatability based on a threshold of ICC > 0.8; and using the Least Absolute Shrinkage and Selection Operator (LASSO) regression algorithm to further screen features that are significantly correlated with MIT-RCC.

[0011] Furthermore, the patient screening criteria for the dataset are that they simultaneously meet the following conditions: they have not received any treatment before the CT examination; the CT images have no obvious image noise or obvious artifacts; the pathological diagnosis is RCC; they have complete clinical pathological data; and they can obtain plain CT scans and arterial phase images of the kidneys from a Picture Archiving and Communication System (PACS). All CT images were obtained from multi-slice spiral CT or higher-level equipment, and the scanning parameters met the following requirements: tube voltage of 120 kVp, tube current automatically adjusted but not less than 100 mAs; slice thickness of 5 mm; contrast-enhanced scans required the use of iodine-based contrast agents, with injection dosage and flow rate standardized according to the patient's weight, and images from both plain and arterial phases were included to ensure that the imaging data contained basic tumor density information and blood supply characteristics.

[0012] One or more technical solutions provided by this invention have at least the following technical effects: (1) By integrating the radiomics features of the tumor and the adjacent kidney tissue, a multidimensional assessment of MIT-RCC can be achieved, breaking through the limitations of traditional tumor region analysis alone and improving the comprehensiveness and accuracy of subtype identification.

[0013] (2) The high repeatability feature screening criterion of ICC>0.8 is combined with the LASSO regression algorithm to effectively remove noisy features and retain key indicators that are highly correlated with MIT-RCC, so as to ensure the robustness and specificity of the model prediction.

[0014] (3) Integrate clinical characteristics such as age and AJCC clinical stage with dual-region radiomics scores to construct a nomograph prediction model, realize quantitative risk assessment, and provide an intuitive and visual tool for clinical decision-making.

[0015] (4) Through a three-level ROI delineation mechanism of three attending physicians for initial screening, two associate chief physicians for review, and senior physicians for arbitration, the accuracy and consistency of the region of interest location are ensured, and the error of manual annotation is reduced.

[0016] (5) Strictly standardize CT scan parameters (120kVp tube voltage, 5mm slice thickness, etc.) and the use of iodine contrast agent to ensure the uniformity of image data quality and improve the universality of model cross-center application.

[0017] (6) It realizes the full-process automated analysis from image data to MIT-RCC probability, assists clinicians to complete accurate classification within 10 minutes, significantly shortens the diagnosis time and improves the efficiency of diagnosis and treatment.

[0018] (7) Establish a feature extraction system that includes plain scan and arterial scan images to fully capture the basic density and blood supply characteristics of the tumor, providing richer imaging evidence for differential diagnosis.

[0019] 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

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 This is a flowchart illustrating the usage process of the prediction model of this invention; Figure 2 This is a flowchart illustrating the construction process of the prediction model of this invention; Figure 3 This is a flowchart illustrating the research process of an embodiment of the present invention; Figure 4 This is a schematic diagram of the ROI delineation of a CT plain scan image according to an embodiment of the present invention. Figure 5 A schematic diagram of the ROI delineation for CT arterial phase images according to an embodiment of the present invention: Figure 6 This is a schematic diagram of the radiomics research framework according to an embodiment of the present invention; Figure 7 This is a summary table of the clinical characteristics of patients in the training and validation sets of this invention. Figure 8 This is a summary table of the image omics feature selection results in the embodiments of the present invention; Figure 9 This is a table showing the results of using Logistic regression analysis to predict clinical factors in MIT family translocation renal cell carcinoma in an embodiment of the present invention. Figure 10 This is a schematic diagram illustrating the construction status of radiomics features and the radiomics-clinical nomograph in an embodiment of the present invention; Figure 11 This is a schematic diagram showing the renal CT imaging assessment status of a patient with AJCC clinical stage III according to an embodiment of the present invention; Figure 12 This is a table evaluating the predictive performance of the model in an embodiment of the present invention.

[0022] Figure 13This is a schematic diagram of the verification status of the nomograph in an embodiment of the present invention.

[0023] Figure 14 This is a block diagram illustrating the principle of the prediction model of this invention. Detailed Implementation

[0024] The technical solution in this application embodiment follows the following general approach: Addressing the challenges of preoperative diagnosis of MIT family translocation renal cell carcinoma (MIT-RCC) and the overlap of traditional imaging findings with other renal cell carcinoma subtypes, a multicenter retrospective study was conducted on 746 patients with renal cell carcinoma (RCC), including 77 MIT-RCC cases. Using CT plain and arterial phase images, three-dimensional regions of interest (ROIs) were segmented into the tumor and surrounding kidney tissue. Radiomic features were extracted and screened using LASSO regression. Combined with clinical factors such as age and AJCC clinical stage, a nomograph prediction model integrating tumor radiomics scores (Tumor_Radscores), kidney tissue radiomics scores (Kidney_Radscores), and clinical indicators was constructed. This model enables accurate preoperative prediction of MIT-RCC, providing a reference for individualized risk assessment and precision treatment. Example 1

[0025] This embodiment provides a method for predicting MIT family translocation renal cell carcinoma, such as... Figure 1 As shown, it includes the following steps: Obtain the initial radiomics feature set and clinical features of the patient to be predicted; the clinical features include age and AJCC clinical stage. The initial radiomics feature set is screened to obtain the screened tumor radiomics feature set and the screened peritumoral kidney tissue radiomics and clinical features set; The kidney tumor radiomics score and the adjacent kidney tissue radiomics score of the patient to be predicted were determined by weighted calculation based on the screened tumor radiomics feature set and the screened adjacent kidney tissue radiomics feature set. The predictive model draws a vertical line on each corresponding coordinate axis based on the renal tumor radiomics score and the radiomics score of the adjacent kidney tissue, age, and AJCC clinical stage to determine the individual scores. All individual scores are added together, the total score is found on the cumulative score coordinate axis, and a descending line is drawn to assess the probability of the patient having MIT-RCC before outputting the results. Among them, such as Figure 2 As shown, the construction process of the prediction model includes: Data Acquisition: Patients were screened from a multi-center medical database to obtain CT images and clinicopathological data of the patients in the dataset; the clinicopathological data included age, gender, body mass index, tumor location, history of diabetes, history of hypertension, AJCC clinical stage, and pathological tumor subtype; like Figure 3 As shown in Figure 3, the study retrospectively collected data from 908 patients with recurrent tumors (RCC) admitted to hospitals A, B, C, D, and E between January 2009 and December 2023. 162 patients were excluded based on the exclusion criteria (40 of whom had received anti-tumor treatment before CT examination; 59 due to substandard image quality or missing data; 32 with incomplete clinicopathological data; and 31 with concurrent malignant tumors in other organs). Finally, 746 RCC patients were included in the analysis, including 77 cases of MIT-RCC and 659 cases of other types of RCC. The study cohort was divided by institution of origin into a training cohort of 539 patients (hospitals A and B) and a validation cohort of 207 patients (hospitals C, D, and E).

[0026] Patients in the dataset were selected based on the following criteria: no prior treatment prior to the CT scan; CT images free of significant noise or artifacts; pathological diagnosis of RCC; complete clinicopathological data; and access to plain and arterial phase CT images of the kidneys from a Picture Archiving and Communication System (PACS). Patients meeting any of the following criteria were ineligible for inclusion: patients who had received treatment prior to the CT scan; patients with poor / missing CT images; patients with incomplete clinicopathological data due to other reasons; or patients with concurrent malignant tumors in other organs.

[0027] All CT images were obtained from a multi-slice (64-slice) spiral CT scanner or higher. Scanning parameters met the following requirements: tube voltage 120 kVp, tube current automatically adjusted but not less than 100 mAs; slice thickness 5 mm; patients were placed in a supine position, and both plain and enhanced abdominal scans were performed, ranging from the diaphragm level to the anterior superior iliac spine level. Iohexol, a non-ionic contrast agent, was injected via the antecubital vein using a high-pressure injector. The arterial phase was 25-28 seconds post-injection, the venous phase was 65-70 seconds post-injection, and the delayed phase was 170 seconds post-injection. The plain and arterial phases provide baseline tumor density and blood supply characteristics. Choosing these two phases reduces redundant information while ensuring sufficient data volume, improving model efficiency and stability. Plain and arterial phase images of the kidneys were acquired from a Picture Archiving and Communication System (PACS), and features from these images were extracted to construct the model.

[0028] The spatial resolution of the image must meet the requirements for clinical diagnosis (e.g., a matrix of 512×512). The image noise level should be controlled within the allowable range (standard deviation not exceeding 20 HU) to avoid affecting diagnosis and radiomics feature extraction due to excessive noise.

[0029] Contrast-enhanced scans require the use of iodine-based contrast agents. The injection dosage and flow rate are standardized according to the patient's weight. Images from both the plain and arterial phases are included to ensure that the imaging data contains information on the tumor's basic density and blood supply characteristics.

[0030] The clinicopathological data were obtained from the pathology department after the patient's biopsy or postoperatively and were reviewed by two experienced pathologists to ensure the accuracy and consistency of the diagnosis. The patient's clinical information (such as comorbidities and demographic data) was extracted from the medical record. Tumor location was determined based on preoperative imaging and intraoperative findings. AJCC clinical staging was performed according to the latest version of the AJCC staging manual.

[0031] Data preprocessing: Grayscale values ​​and slice thickness are normalized and consistent in the CT images to obtain preprocessed CT images, reducing cross-center data differences. Data preprocessing requires complete image data, including the original DICOM format files from plain and contrast-enhanced scans.

[0032] Radiomics feature extraction: The preprocessed CT images were segmented into tumor ROI and adjacent kidney tissue ROI; radiomics features of the tumor ROI and adjacent kidney tissue ROI were extracted, including shape features, first-order histogram features, gray-level run length matrix features, gray-level co-occurrence matrix features, gray-level dependence matrix features, and adjacent gray-level difference matrix.

[0033] In radiomics, ROI / VOI segmentation is a core component, and its repeatability and accuracy are crucial. In this study, a manual segmentation method was used for ROI segmentation. Specifically, in ITK-SNAP software, three radiologists with more than 5 years of diagnostic experience and at least the title of attending physician manually delineated the tumor region on preprocessed CT plain and arterial phase images to obtain the tumor ROI and the adjacent kidney tissue ROI. The region of interest was manually drawn on each cross-sectional image of the renal cell carcinoma to obtain the region of interest. The segmentation results were reviewed and verified by two radiologists with more than 10 years of experience in diagnosing urinary tract tumors and at least the title of associate senior physician. When the opinions of the two physicians differed, another radiologist with 20 years of experience in diagnosing radiological tumors and at least the title of senior physician participated in the discussion to determine the location of the region of interest. The tumor ROI and the adjacent kidney tissue ROI were then fused to form a three-dimensional region of interest.

[0034] like Figure 4The image shown is a schematic diagram of ROI delineation on a CT scan. A and D are CT scan images of a patient with MIT-RCC. The red area in B is the tumor ROI. C is the tumor's three-dimensional region of interest (VOI) formed by software fusion after delineating ROIs layer by layer. The blue area in E is the kidney ROI adjacent to the tumor. F is the kidney's three-dimensional VOI formed by software fusion after delineating ROIs layer by layer.

[0035] like Figure 5 The image shown is a schematic diagram of ROI delineation in CT arterial phase images. A and D are CT arterial phase images of MIT-RCC patients; the red area in B is the tumor ROI; C is the tumor stereoscopic region of interest (VOI) formed by software fusion after delineating ROIs layer by layer; E is the kidney ROI adjacent to the tumor in the blue area; and F is the kidney stereoscopic VOI formed by software fusion after delineating ROIs layer by layer.

[0036] Then, radiomics features were extracted using the Python (version 3.7.3) package PyRadiomics version 3.0. In addition to applying two basic image filters (i.e., Laplacian of Gaussian (LoG) and wavelet images), radiomics features were also obtained from the original images. The extracted features include: (1) Shape features: such as volume, surface area and perimeter; (2) First-order histogram features: such as mean, standard deviation, skewness and kurtosis; (3) Gray-Level Run-Length Matrix (GLRLM) features: such as matrix entropy, contrast and uniformity; (4) Gray-Level Co-occurrence Matrix (GLCM) features; (5) Gray-Level Dependence Matrix (GLDM) features: such as contrast, correlation, energy and entropy; (6) Gray-Level Size Zone Matrix (GLSZM) features: such as size distribution and gray-level distribution; (7) Neighboring Gray-Tone Difference Matrix (NGTDM) features: such as roughness, contrast, busyness, complexity and intensity difference.

[0037] Radiomics Feature Screening: Radiomics features with excellent reproducibility and significant association with MTT-RCC were screened to generate a final feature set, which served as the input to the model. The screening aimed to eliminate scale differences and ensure comparability; all features were first standardized using Z-score transformation. Features with low reproducibility were then excluded from further analysis. Intraclass Correlation Coefficient (ICC) (using R package "psych" version 2.4.3) was used to analyze inter- and intra-observer reproducibility. In this study, ICC > 0.8 was used as the threshold for feature screening for subsequent analysis. Least Absolute Shrinkage and Selection Operator (LASSO) regression (using R packages "glmnet" versions 4.1-7 and "pROC" version 1.18.0) was used to screen radiomics features with excellent reproducibility and significant association with MTT-RCC. Finally, a radiomics model based on renal tumors and adjacent renal tissue was constructed using logistic regression analysis.

[0038] Feature weight determination: Through univariate and multivariate logistic regression analysis, clinical features significantly associated with MIT-RCC were identified from the clinicopathological data, and the association strength between tumor radiomics features, adjacent kidney tissue radiomics features, age, and AJCC stage and MIT-RCC was quantified. Pre-defined weights were assigned to each feature based on regression coefficients. Specifically, univariate logistic regression analysis was used to assess the correlation between radiomics features, clinical features, and MIT-RCC, followed by multivariate logistic regression analysis to screen for independent predictors of MIT-RCC. Model construction: A nomograph model is constructed based on the preset weights, and the weight allocation is optimized through cross-validation to ensure generalization ability; Model Training and Validation: Central hierarchical validation was employed, dividing the final feature set into a training set and an independent validation set. These sets were then input into the nomomotor model, and the model's probabilistic predictive accuracy and clinical applicability were evaluated using ROC curves, calibration curves, and decision curves. The model's goodness of fit was verified using the Hosmer-Lemeshow test, yielding the predicted model. Specifically, the diagnostic efficacy of nomomotor was assessed using accuracy (ACC), sensitivity (SEN), and specificity (SPE), while receiver operating characteristic (ROC) curves were plotted and the area under the curve (AUC) was calculated. The DeLong test was used to compare the AUC differences between the combined imaging-clinical nomomotor model and different predictive factors. The calibration of nomomotor was further evaluated using the Hosmer-Lemeshow test and calibration curves. Decision curve analysis (DCA) was performed to assess the clinical utility of nomomotor. The radiomics-clinical nomomap was constructed and validated using R packages, including “Hmisc” version 5.0-1, “car” version 3.1-2, “rms” version 6.6-0, “pROC” version 1.18-0, “survival” version 3.5-5, “survminer” version 0.4.9 and “rmda” version 1.6.

[0039] Statistical analysis was performed using SPSS 26.0 (IBM, Armonk, NY, USA) and R 4.1.0 (R Foundation for Statistical Computation, Vienna, Austria). Chi-square tests or Fisher's exact tests were used for comparisons of categorical variables, while independent samples t-tests or Mann-Whitney U tests were used for comparisons of continuous variables. The significance level was set at α = 0.05 (two-tailed).

[0040] like Figure 6As shown, the research process of this invention is as follows: Based on manually segmented regions of interest (ROIs) of the tumor and its surrounding kidney tissue, imaging features of the kidney tumor and adjacent kidney tissue for each patient were extracted from plain and arterial phase sequences. 1316 features were extracted from each ROI. Features with intra- or inter-observer ICC values ​​below 0.8 were removed from the analysis due to their low reproducibility. Subsequently, feature selection was performed using the Least Absolute Contraction and Selection Operator (LASSO), and a radiomics feature model was constructed using the selected features. Based on this, the radiomics features were combined with clinical factors and integrated into a nomogram to achieve personalized assessment.

[0041] The following is an analysis of the research results of the examples. Clinicopathological data: The general clinicopathological characteristics of patients in the training set (539 cases) and validation set (207 cases) were analyzed to explore the differences between MIT-RCC and other types of renal cell carcinoma. Results showed that MIT-RCC patients exhibited a significant trend towards younger age; the mean age of the MIT group in the training set was 12.01 years younger than that of the non-MIT group (44.19±11.75 vs 56.20±8.55, P<0.001), and the difference remained statistically significant in the validation set (49.56±12.87 vs 54.96±10.83, P=0.008). There were no significant differences between the two groups in terms of BMI, gender, tumor location, and the prevalence of metabolic diseases (P>0.05), but there were differences in the clinical stage of AJCC (P<0.001). In the training set, the proportion of stage III-IV patients in the MIT group reached 58.5% (31 / 53), which was 3.3 times higher than that in the non-MIT group, and this trend continued in the validation set (20.8% vs 8.2%). Studies suggest that the occurrence of MIT is strongly correlated with young onset and advanced clinical stage, which may be related to its unique biological behavior, providing an important basis for early clinical identification.

[0042] Radiomics Feature Selection: Radiomic features of renal tumors and adjacent renal tissues for each patient were extracted from plain and arterial phase sequences. 1316 features were extracted from each region of interest (ROI). Features with intra- or inter-observer ICC values ​​below 0.8, indicating low reproducibility, were excluded from the analysis. The number of radiomics features in adjacent renal tissues was reduced to 855, while the number of features within the tumor was reduced to 931. Subsequently, 24 radiomics features were obtained from the adjacent renal ROI and 22 features from the tumor ROI using LASSO regression.

[0043] Furthermore, such as Figure 8 As shown in Table 2, the tumor ROI in the final feature set has 22 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.2.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.2.0 mm. 3D grayscale run length matrix. Run length non-uniformity; log.σ.3.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.5.0 mm. 3D grayscale dependency matrix_dependency variance; Original grayscale dependency matrix and dependency variance; Original grayscale dependency matrix with low grayscale emphasis; wavelet.HHH_grayscale dependency matrix_dependency variance; Wavelet.HHH_Grayscale region matrix_Large area low grayscale emphasis; Wavelet.HLH_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.HLL_Grayscale Dependency Matrix_High Dependency High Grayscale Emphasis; Wavelet.HLL_grayscale region matrix_large area low grayscale emphasis; Wavelet.LHH_Gray-level Co-occurrence Matrix_Difference Entropy; wavelet.LHL_grayscale dependency matrix_dependency variance; wavelet.LLH_grayscale run length matrix_run variance; Wavelet.LLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.LLL_Gray-level size region matrix_Gray-level non-uniformity normalization; The adjacent kidney tissue ROI possessed 24 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix with large area high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix and its non-uniformity. log.σ.5.0 mm. 3D_grayscale co-occurrence matrix_maximum probability; log.σ.5.0 mm. 3D_grayscale size region matrix_large area emphasis; log.σ.5.0 mm. 3D grayscale region matrix with region variance; Original first-order feature 10 percentile; Original first-order feature minimum; Original grayscale region matrix - non-uniformity of region size; Wavelet.HHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HHH_Grayscale region matrix_Large area high grayscale emphasis; wavelet.HHH_grayscale region matrix_region variance; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.HLL_First-order feature_energy; Wavelet.HLL_First-order feature_Total energy; Wavelet.LHH_First-order feature_energy; Wavelet.LHH_First-order feature_Total energy; Wavelet.LHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.LHH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.LHL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.LLL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Grayscale Region Matrix_Region Variance

[0044] Construction of Nomographs: An example is the use of univariate and multivariate logistic regression analysis to analyze clinical factors to predict the risk of MIT-RCC, such as... Figure 9As shown in the figure, univariate and multivariate logistic regression analyses revealed that age, clinical stage of AJCC, and radiomics characteristics were important factors in predicting MIT-RCC. In univariate analysis, age (OR=0.907, 95% CI: 0.881-0.934, P<0.001), AJCC clinical stage II (OR=3.432, 95% CI: 1.332-8.843, P=0.011), III (OR=7.942, 95% CI: 4.039-15.618, P<0.001), and IV (OR=18.827, 95% CI: 4.584-77.319, P<0.001), as well as tumor radiomics (OR=0.768, 95% CI: 0.694-0.850, P<0.001) and renal radiomics characteristics (OR=1.919, 95% CI: 1.656-2.224), showed significant differences. All of these were significantly associated with MIT-RCC (P < 0.001). In multivariate analysis, age (OR = 0.937, 95% CI: 0.910–0.964, P < 0.001), AJCC clinical stage III (OR = 10.293, 95% CI: 3.389–29.156, P < 0.001) and IV (OR = 5.926, 95% CI: 1.937–18.125, P < 0.001), as well as tumor radiomics (OR = 0.802, 95% CI: 0.710–0.907, P < 0.001) and peritumoral radiomics features (OR = 1.636, 95% CI: 1.381–1.938, P < 0.001) remained independent predictors. Other factors such as BMI, gender, tumor location, hypertension, and diabetes did not show significant correlations in univariate and multivariate analyses (P > 0.05).

[0045] In summary, younger age, higher AJCC stage, tumor radiomics features, and peritumoral renal radiomics features are important predictors of MIT-RCC. A nomograph (E in Figure 10) was constructed by integrating age, AJCC clinical stage, tumor radiomics features, and peritumoral renal radiomics features. Tenfold cross-validation was used to adjust for renal tumors (…). Figure 10 B) and adjacent kidney tissue (in the middle) Figure 10 The parameter lambda for the D) feature. Referring to the coefficient curve generated from the optimal log(lambda) sequence, the tumor ( Figure 10 A) and kidney ( Figure 10 The coefficients of feature C in the model contain 22 and 24 non-zero values, respectively. Figure 10 The E) in the prediction of MIT-RCC Nomograph.

[0046] To use this model, the following steps must be followed: (1) Determine the patient’s renal tumor radiomics score, peri-tumor renal tissue radiomics score, age, and AJCC clinical stage, and then draw a vertical line on each of the corresponding coordinate axes to determine their respective scores; (2) Add these scores together; (3) Find the total score on the cumulative score coordinate axis and then draw a descending line to assess the probability that the patient has MIT-RCC.

[0047] For example, a patient with a total score of 162.5, a Tumor_radiomics score of 6.5 (out of 70), a Kidney_radiomics score of 7.8 (out of 42.5), an age of 31 years (out of 39), and an AJCC clinical stage of stage III (out of 11), predicted a MIT-RCC probability of approximately 0.9. The final immunohistochemical results showed XP11.2 translocation / TFE3 gene fusion-related renal cell carcinoma. Figure 11 As shown in AD. Figure 11 Renal CT imaging evaluation of a 31-year-old male patient (AJCC clinical stage III) showed: A was the plain scan, and B was the arterial phase contrast-enhanced scan. The Nomograph (D) comprehensive scoring system yielded scores of 39 (age), 11 (AJCC clinical stage III), 42.5 (radiomamic score of adjacent kidney tissue), and 70 (tumor radiomics score), for a total score of 162.5, suggesting a probability of MIT-RCC of 0.9. Final pathological examination (C) confirmed XP11.2 translocation / TFE3 gene fusion-related renal cell carcinoma.

[0048] Model Validation: This invention comprehensively evaluated the predictive performance of the nomograph model, which combines four key factors: age, AJCC clinical stage, Tumor radiomics score, and Kidney radiomics score. Results show that the nomograph model exhibits superior predictive ability on both the training and validation sets, with its overall performance significantly outperforming single-factor models. Compared to traditional clinical models based solely on age or AJCC clinical stage, the nomograph model enhances its ability to identify MIT-RCC by integrating clinical and radiomics features. This result demonstrates that multi-factor joint modeling can more accurately capture the complex characteristics of the disease, providing a more reliable basis for clinical decision-making and laying the foundation for future personalized treatment strategies.

[0049] In the training set, the nomotu model achieved an AUC of 0.952 (95% CI: 0.927–0.971), significantly outperforming single-factor models, including age (AUC = 0.768, z = 3.959, P < 0.001), AJCC clinical stage (AUC = 0.706, z = 5.665, P < 0.001), renal radiomics (AUC = 0.851, z = 3.937, P < 0.001), and tumor radiomics (AUC = 0.745, z = 4.254, P < 0.001). In the validation set, the AUC of the nomotu model was 0.914 (95% CI: 0.879–0.942), which was significantly higher than age (AUC = 0.722, z = 4.513, P < 0.001), AJCC clinical stage (AUC = 0.683, z = 5.442, P < 0.001), renal radiomics (AUC = 0.807, z = 4.485, P < 0.001), and tumor radiomics (AUC = 0.730, z = 5.295, P < 0.001) (see [link to validation set]). Figure 12 (Table 4). In Table 4 In the training set, a) compares the nomota model with age in predicting MIT-RCC. b) compares the nomota model with AJCC clinical staging in predicting MIT-RCC. c) compares the nomota model with renal radiomics in predicting MIT-RCC. d) compares the nomota model with tumor radiomics in predicting MIT-RCC.

[0050] In the validation set, a) compares the nomota model with age in predicting MIT-RCC. b) compares the nomota model with AJCC clinical stage in predicting MIT-RCC. c) compares the nomota model with renal radiomics in predicting MIT-RCC. d) compares the nomota model with tumor radiomics in predicting MIT-RCC.

[0051] These results demonstrate that the Nomotu model has significantly higher predictive performance in identifying MIT-RCC, outperforming other univariate models.

[0052] Furthermore, the Hosmer-Lemeshow test in the training set showed that the nomograph model (χ2=2.920, p=0.939) had a good fit. Figure 13 (C) In the validation set, the Hosmer-Lemeshow test showed that the nomograph model (χ2=3.923, P=0.864) had a good fit. Figure 13(D in the text). The decision curves of the Nomotu model on the training and validation sets are as follows: Figure 13 As shown in EF, this indicates that the radiomics nomograph model has good clinical applicability.

[0053] Figure 13 In the figures, A and B represent receiver operating characteristic (ROC) curve analyses of the four models. C and D represent Hosmer-Lemeshow goodness-of-fit tests used to evaluate model calibration in the training group (χ² = 2.920, p = 0.939) and the validation group (χ² = 3.923, p = 0.864). Decision curve analyses in E and F indicate that nomogenographs can facilitate clinical decision-making within a fairly large risk threshold. Example 2

[0054] Based on the same inventive concept, this application also provides an apparatus corresponding to the method in Embodiment 1, as detailed in Embodiment 2.

[0055] like Figure 14 As shown, the predictive model for MIT family translocation renal cell carcinoma in this embodiment includes: The kidney tumor radiomics scoring input module is used to receive the kidney tumor radiomics scores of the patients to be predicted. The radiomics scoring input module for adjacent kidney tissue is used to receive the radiomics scores of adjacent kidney tissue of the patient to be predicted. A clinical feature input module is used to receive clinical features, including age and AJCC clinical stage; The single-item line drawing module draws a vertical line on each corresponding coordinate axis based on the radiomics score of the kidney tumor and the radiomics score of the adjacent kidney tissue, age, and AJCC clinical stage to determine the individual score; The calculation module adds up all the individual scores and finds the total score on the cumulative score axis; The total score line drawing module draws a descending line based on the total score. The probability assessment module is used to assess and output the probability of a patient being diagnosed with MIT family translocation renal cell carcinoma. Among them, such as Figure 2 As shown, the construction process of the prediction model includes: Data Acquisition: Patients were screened from a multi-center medical database to obtain CT images and clinicopathological data of the patients in the dataset; the clinicopathological data included age, gender, body mass index, tumor location, history of diabetes, history of hypertension, AJCC clinical stage, and pathological tumor subtype; Data preprocessing: The CT images are normalized in grayscale and slice thickness is made consistent to obtain preprocessed CT images; Radiomics feature extraction: The preprocessed CT images were segmented into tumor ROI and adjacent kidney tissue ROI; radiomics features of the tumor ROI and adjacent kidney tissue ROI were extracted, including shape features, first-order histogram features, gray-level run length matrix features, gray-level co-occurrence matrix features, gray-level dependence matrix features, and adjacent gray-level difference matrix. Screening of radiomics features: Radiomics features with excellent reproducibility and significant correlation with MTT-RCC are screened to generate a final feature set, which is used as the input to the model; Feature weight determination: Through univariate and multivariate logistic regression analysis, the clinical features that are significantly associated with MIT-RCC are determined from the clinicopathological data, and the association strength between tumor radiomics features, adjacent kidney tissue radiomics features, age and AJCC stage and MIT-RCC is quantified; a preset weight is assigned to each feature through regression coefficients; Model construction: A nomograph model is constructed based on the preset weights, and the weight allocation is optimized through cross-validation to ensure generalization ability; Model training and validation: Central hierarchical validation was adopted, dividing the final feature set into a training set and an independent validation set. The training set and the independent validation set were respectively input into the nomograph model, and evaluated by ROC curve, calibration curve and decision curve to confirm the probability prediction accuracy and clinical applicability of the nomograph model; the model goodness of fit was verified by Hosmer-Lemeshow test to obtain the prediction model.

[0056] like Figure 8 As shown, in the final feature set, the tumor ROI has 22 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.2.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.2.0 mm. 3D grayscale run length matrix. Run length non-uniformity; log.σ.3.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.5.0 mm. 3D grayscale dependency matrix_dependency variance; Original grayscale dependency matrix and dependency variance; Original grayscale dependency matrix with low grayscale emphasis; wavelet.HHH_grayscale dependency matrix_dependency variance; Wavelet.HHH_Grayscale region matrix_Large area low grayscale emphasis; Wavelet.HLH_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.HLL_Grayscale Dependency Matrix_High Dependency High Grayscale Emphasis; Wavelet.HLL_grayscale region matrix_large area low grayscale emphasis; Wavelet.LHH_Gray-level Co-occurrence Matrix_Difference Entropy; wavelet.LHL_grayscale dependency matrix_dependency variance; wavelet.LLH_grayscale run length matrix_run variance; Wavelet.LLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.LLL_Gray-level size region matrix_Gray-level non-uniformity normalization; The adjacent kidney tissue ROI possessed 24 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix with large area high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix and its non-uniformity. log.σ.5.0 mm. 3D_grayscale co-occurrence matrix_maximum probability; log.σ.5.0 mm. 3D_grayscale size region matrix_large area emphasis; log.σ.5.0 mm. 3D grayscale region matrix with region variance; Original first-order feature 10 percentile; Original first-order feature minimum; Original grayscale region matrix - non-uniformity of region size; Wavelet.HHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HHH_Grayscale region matrix_Large area high grayscale emphasis; wavelet.HHH_grayscale region matrix_region variance; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.HLL_First-order feature_energy; Wavelet.HLL_First-order feature_Total energy; Wavelet.LHH_First-order feature_energy; Wavelet.LHH_First-order feature_Total energy; Wavelet.LHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.LHH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.LHL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.LLL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Grayscale Region Matrix_Region Variance

[0057] Furthermore, the specific process of radiomics feature extraction is as follows: In ITK-SNAP software, three radiologists with more than 5 years of diagnostic experience and at least the title of attending physician manually delineate the tumor region in the preprocessed CT plain scan and arterial phase images to obtain the tumor ROI and the adjacent kidney tissue ROI. The region of interest is manually drawn on each cross-sectional image of the renal cell carcinoma to obtain the region of interest. The segmentation results are reviewed and verified by two radiologists with more than 10 years of experience in diagnosing urinary tract tumors and at least the title of associate senior physician. If the opinions of the two physicians are inconsistent, another radiologist with 20 years of experience in diagnosing imaging and at least the title of senior physician participates in the joint discussion to determine the location of the region of interest. The tumor ROI and the adjacent kidney tissue ROI are fused to form a three-dimensional region of interest.

[0058] The screening of radiomics features includes: calculating the intragroup correlation coefficient (ICC) to assess the repeatability of features; screening features with high repeatability based on an ICC > 0.8 threshold; and further screening features significantly associated with MIT-RCC using the Least Absolute Shrinkage and Selection Operator (LASSO) regression algorithm.

[0059] The patient selection criteria for the dataset are as follows: no prior treatment was received before the CT examination; the CT images have no obvious image noise or artifacts; the pathological diagnosis is RCC; complete clinical pathological data is available; and plain CT scans and arterial phase images of the kidneys can be obtained from a Picture Archiving and Communication System (PACS). All CT images were obtained from multi-slice spiral CT or higher-level equipment, and the scanning parameters met the following requirements: tube voltage of 120 kVp, tube current automatically adjusted but not less than 100 mAs; slice thickness of 5 mm; contrast-enhanced scans required the use of iodine-based contrast agents, with injection dosage and flow rate standardized according to the patient's weight, and images from both plain and arterial phases were included to ensure that the imaging data contained basic tumor density information and blood supply characteristics.

[0060] 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.

[0061] 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 MIT family translocation renal cell carcinoma, characterized in that: Includes the following steps: Obtain the initial radiomics feature set and clinical features of the patient to be predicted; the clinical features include age and AJCC clinical stage. The initial radiomics feature set is screened to obtain the screened tumor radiomics feature set and the screened peritumoral kidney tissue radiomics and clinical features set; The kidney tumor radiomics score and the adjacent kidney tissue radiomics score of the patient to be predicted were determined by weighted calculation based on the screened tumor radiomics feature set and the screened adjacent kidney tissue radiomics feature set. The predictive model draws a vertical line on each corresponding coordinate axis based on the renal tumor radiomics score and the radiomics score of the adjacent kidney tissue, age, and AJCC clinical stage to determine the individual scores. All individual scores are added together, the total score is found on the cumulative score coordinate axis, and a descending line is drawn to assess the probability of the patient having MIT-RCC before outputting the results. The construction process of the prediction model includes: Data Acquisition: Patients were screened from a multi-center medical database to obtain CT images and clinicopathological data of the patients in the dataset; the clinicopathological data included age, gender, body mass index, tumor location, history of diabetes, history of hypertension, AJCC clinical stage, and pathological tumor subtype; Data preprocessing: The CT images are normalized in grayscale and slice thickness is made consistent to obtain preprocessed CT images; Radiomics feature extraction: The preprocessed CT images were segmented into tumor ROI and adjacent kidney tissue ROI; radiomics features of the tumor ROI and adjacent kidney tissue ROI were extracted, including shape features, first-order histogram features, gray-level run length matrix features, gray-level co-occurrence matrix features, gray-level dependence matrix features, and adjacent gray-level difference matrix. Screening of radiomics features: Radiomics features with excellent reproducibility and significant correlation with MTT-RCC are screened to generate a final feature set, which is used as the input to the model; Feature weight determination: Through univariate and multivariate logistic regression analysis, the clinical features that are significantly associated with MIT-RCC are determined from the clinicopathological data, and the association strength between tumor radiomics features, adjacent kidney tissue radiomics features, age and AJCC stage and MIT-RCC is quantified; a preset weight is assigned to each feature through regression coefficients; Model construction: A nomograph model is constructed based on the preset weights, and the weight allocation is optimized through cross-validation to ensure generalization ability; Model training and validation: Central hierarchical validation was adopted, dividing the final feature set into a training set and an independent validation set. The training set and the independent validation set were respectively input into the nomograph model, and evaluated by ROC curve, calibration curve and decision curve to confirm the probability prediction accuracy and clinical applicability of the nomograph model; the model goodness of fit was verified by Hosmer-Lemeshow test to obtain the prediction model.

2. The method according to claim 1, characterized in that: The final feature set, The tumor ROI has 22 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.2.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.2.0 mm. 3D grayscale run length matrix. Run length non-uniformity; log.σ.3.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.5.0 mm. 3D grayscale dependency matrix_dependency variance; Original grayscale dependency matrix and dependency variance; Original grayscale dependency matrix with low grayscale emphasis; wavelet.HHH_grayscale dependency matrix_dependency variance; Wavelet.HHH_Grayscale region matrix_Large area low grayscale emphasis; Wavelet.HLH_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.HLL_Grayscale Dependency Matrix_High Dependency High Grayscale Emphasis; Wavelet.HLL_grayscale region matrix_large area low grayscale emphasis; Wavelet.LHH_Gray-level Co-occurrence Matrix_Difference Entropy; wavelet.LHL_grayscale dependency matrix_dependency variance; wavelet.LLH_grayscale run length matrix_run variance; Wavelet.LLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.LLL_Gray-level size region matrix_Gray-level non-uniformity normalization; The adjacent kidney tissue ROI possessed 24 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix with large area high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix and its non-uniformity. log.σ.5.0 mm. 3D_grayscale co-occurrence matrix_maximum probability; log.σ.5.0 mm. 3D_grayscale size region matrix_large area emphasis; log.σ.5.0 mm. 3D grayscale region matrix with region variance; Original first-order feature 10 percentile; Original first-order feature minimum; Original grayscale region matrix - non-uniformity of region size; Wavelet.HHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HHH_Grayscale region matrix_Large area high grayscale emphasis; wavelet.HHH_grayscale region matrix_region variance; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.HLL_First-order feature_energy; Wavelet.HLL_First-order feature_Total energy; Wavelet.LHH_First-order feature_energy; Wavelet.LHH_First-order feature_Total energy; Wavelet.LHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.LHH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.LHL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.LLL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Grayscale Region Matrix_Region Variance 3. The method according to claim 1, characterized in that: The specific process of radiomics feature extraction is as follows: In ITK-SNAP software, three radiologists with more than 5 years of diagnostic experience and at least the title of attending physician manually delineate the tumor region in the preprocessed CT plain scan and arterial phase images to obtain the tumor ROI and the adjacent kidney tissue ROI. The region of interest is manually drawn on each cross-sectional image of the renal cell carcinoma to obtain the region of interest. The segmentation results are reviewed and verified by two radiologists with more than 10 years of experience in diagnosing urinary tract tumors and at least the title of associate senior physician. If the opinions of the two physicians are inconsistent, another radiologist with 20 years of experience in diagnosing imaging and at least the title of senior physician participates in the discussion to determine the location of the region of interest. The tumor ROI and the adjacent kidney tissue ROI are fused to form a three-dimensional region of interest.

4. The method according to claim 1, characterized in that: The screening of radiomics features includes: calculating intragroup correlation coefficients to assess feature repeatability; screening features with high repeatability based on an intragroup correlation coefficient > 0.8; and further screening features significantly associated with MIT-RCC using a minimum absolute shrinkage and selection operator regression algorithm.

5. The method according to claim 1, characterized in that: The patient selection criteria for the dataset are as follows: no prior treatment was received before the CT examination; the CT images have no obvious image noise or artifacts; the pathological diagnosis is RCC; complete clinical pathological data is available; and plain CT scans and arterial phase images of the kidneys can be obtained from the image storage and transmission system. All CT images were obtained from multi-slice spiral CT or higher-level equipment, and the scanning parameters met the following requirements: tube voltage of 120kVp, tube current automatically adjusted but not less than 100 mAs; slice thickness of 5 mm; contrast-enhanced scans required the use of iodine-based contrast agents, with injection dosage and flow rate standardized according to the patient's weight, and images from both plain and arterial phases were included to ensure that the image data contained basic tumor density information and blood supply characteristics.

6. A predictive model for MIT family translocation renal cell carcinoma, characterized in that: include: The kidney tumor radiomics scoring input module is used to receive the kidney tumor radiomics scores of the patients to be predicted. The radiomics scoring input module for adjacent kidney tissue is used to receive the radiomics scores of adjacent kidney tissue of the patient to be predicted. The clinical feature input module is used to receive clinical features, including age and AJCC clinical stage. The single-item line drawing module draws a vertical line on each corresponding coordinate axis based on the radiomics score of the kidney tumor and the radiomics score of the adjacent kidney tissue, age, and AJCC clinical stage to determine the individual score; The calculation module adds up all the individual scores and finds the total score on the cumulative score axis; The total score line drawing module draws a descending line based on the total score. The probability assessment module is used to assess and output the probability of a patient being diagnosed with MIT family translocation renal cell carcinoma. The construction process of the prediction model includes: Data Acquisition: Patients were screened from a multi-center medical database to obtain CT images and clinicopathological data of the patients in the dataset; the clinicopathological data included age, gender, body mass index, tumor location, history of diabetes, history of hypertension, AJCC clinical stage, and pathological tumor subtype; Data preprocessing: The CT images are normalized in grayscale and slice thickness is made consistent to obtain preprocessed CT images; Radiomics feature extraction: The preprocessed CT images were segmented into tumor ROI and adjacent kidney tissue ROI; radiomics features of the tumor ROI and adjacent kidney tissue ROI were extracted, including shape features, first-order histogram features, gray-level run length matrix features, gray-level co-occurrence matrix features, gray-level dependence matrix features, and adjacent gray-level difference matrix. Screening of radiomics features: Radiomics features with excellent reproducibility and significant correlation with MTT-RCC are screened to generate a final feature set, which is used as the input to the model; Feature weight determination: Through univariate and multivariate logistic regression analysis, the clinical features that are significantly associated with MIT-RCC are determined from the clinicopathological data, and the association strength between tumor radiomics features, adjacent kidney tissue radiomics features, age and AJCC stage and MIT-RCC is quantified; a preset weight is assigned to each feature through regression coefficients; Model construction: A nomograph model is constructed based on the preset weights, and the weight allocation is optimized through cross-validation to ensure generalization ability; Model training and validation: Central hierarchical validation was adopted, dividing the final feature set into a training set and an independent validation set. The training set and the independent validation set were respectively input into the nomograph model, and evaluated by ROC curve, calibration curve and decision curve to confirm the probability prediction accuracy and clinical applicability of the nomograph model; the model goodness of fit was verified by Hosmer-Lemeshow test to obtain the prediction model.

7. The prediction model according to claim 6, characterized in that: The final feature set, The tumor ROI has 22 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.2.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.2.0 mm. 3D grayscale run length matrix. Run length non-uniformity; log.σ.3.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D grayscale dependency matrix_dependency variance; log.σ.4.0 mm. 3D_grayscale dependency matrix_low grayscale emphasis; log.σ.5.0 mm. 3D grayscale dependency matrix_dependency variance; Original grayscale dependency matrix and dependency variance; Original grayscale dependency matrix with low grayscale emphasis; wavelet.HHH_grayscale dependency matrix_dependency variance; Wavelet.HHH_Grayscale region matrix_Large area low grayscale emphasis; Wavelet.HLH_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.HLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.HLL_Grayscale Dependency Matrix_High Dependency High Grayscale Emphasis; Wavelet.HLL_grayscale region matrix_large area low grayscale emphasis; Wavelet.LHH_Gray-level Co-occurrence Matrix_Difference Entropy; wavelet.LHL_grayscale dependency matrix_dependency variance; wavelet.LLH_grayscale run length matrix_run variance; Wavelet.LLL_Gray-level Co-occurrence Matrix_Difference Entropy; Wavelet.LLL_Gray-level size region matrix_Gray-level non-uniformity normalization; The adjacent kidney tissue ROI possessed 24 radiomics features, specifically including: log.σ.1.0 mm. 3D_grayscale dependency matrix_high dependency high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix with large area high grayscale emphasis; log.σ.1.0 mm. 3D grayscale region matrix and its non-uniformity. log.σ.5.0 mm. 3D_grayscale co-occurrence matrix_maximum probability; log.σ.5.0 mm. 3D_grayscale size region matrix_large area emphasis; log.σ.5.0 mm. 3D grayscale region matrix with region variance; Original first-order feature 10 percentile; Original first-order feature minimum; Original grayscale region matrix - non-uniformity of region size; Wavelet.HHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HHH_Grayscale region matrix_Large area high grayscale emphasis; wavelet.HHH_grayscale region matrix_region variance; Wavelet.HLH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.HLH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.HLL_First-order feature_energy; Wavelet.HLL_First-order feature_Total energy; Wavelet.LHH_First-order feature_energy; Wavelet.LHH_First-order feature_Total energy; Wavelet.LHH_Gray-level Dependency Matrix_High Dependency High Gray-level Emphasis; Wavelet.LHH_Grayscale Region Matrix_Large Area Low Grayscale Emphasis; Wavelet.LHL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Gray-level Dependency Matrix_Dependence Non-uniformity; Wavelet.LLL_Grayscale Region Matrix_Large Area High Grayscale Emphasis; Wavelet.LLL_Grayscale Region Matrix_Region Variance 8. The prediction model according to claim 6, characterized in that: The specific process of radiomics feature extraction is as follows: In ITK-SNAP software, three radiologists with more than 5 years of diagnostic experience and at least the title of attending physician manually delineate the tumor region in the preprocessed CT plain scan and arterial phase images to obtain the tumor ROI and the adjacent kidney tissue ROI. The region of interest is manually drawn on each cross-sectional image of the renal cell carcinoma to obtain the region of interest. The segmentation results are reviewed and verified by two radiologists with more than 10 years of experience in diagnosing urinary tract tumors and at least the title of associate senior physician. If the opinions of the two physicians are inconsistent, another radiologist with 20 years of experience in diagnosing imaging and at least the title of senior physician participates in the discussion to determine the location of the region of interest. The tumor ROI and the adjacent kidney tissue ROI are fused to form a three-dimensional region of interest.

9. The prediction model according to claim 6, characterized in that: The screening of radiomics features includes: calculating intragroup correlation coefficients to assess feature repeatability; screening features with high repeatability based on an intragroup correlation coefficient > 0.8; and further screening features significantly associated with MIT-RCC using a minimum absolute shrinkage and selection operator regression algorithm.

10. The prediction model according to claim 6, characterized in that: The patient selection criteria for the dataset are as follows: no prior treatment was received before the CT examination; the CT images have no obvious image noise or artifacts; the pathological diagnosis is RCC; complete clinical pathological data is available; and plain CT scans and arterial phase images of the kidneys can be obtained from the image storage and transmission system. All CT images were obtained from multi-slice spiral CT or higher-level equipment, and the scanning parameters met the following requirements: tube voltage of 120kVp, tube current automatically adjusted but not less than 100 mAs; slice thickness of 5 mm; contrast-enhanced scans required the use of iodine-based contrast agents, with injection dosage and flow rate standardized according to the patient's weight, and images from both plain and arterial phases were included to ensure that the image data contained basic tumor density information and blood supply characteristics.

Images