Exploration of Prognostic Prediction Models for Renal Cell Carcinoma using Diffusion Relaxation Correlation Spectroscopic Imaging

Exploration of Prognostic Prediction Models for Renal Cell Carcinoma using Diffusion Relaxation Correlation Spectroscopic Imaging | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Exploration of Prognostic Prediction Models for Renal Cell Carcinoma using Diffusion Relaxation Correlation Spectroscopic Imaging Mengying Zhu, Yuansheng Luo, Xiaobin Wei, Jingli Chen, Shihang Pan, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7287776/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background The prognosis of renal cell carcinoma (RCC) varies greatly, and a highly efficient prognostic strategy is crucial for treatment selection. This study aim to evaluate the feasibility of predictive models based on diffusion relaxation correlation spectroscopic imaging (DR-CSI) in distinguishing RCC patients with different clinical outcomes. Methods One hundred and twenty-seven RCC patients who underwent DR-CSI were enrolled, including 48 patients as cohort 1 for development and 79 postoperative follow-up patients as cohort 2 for validation. DR-CSI results were analyzed using spectral equipartition method combined with various feature selection methods and classifiers, from 2*2 to 9*9. The Kruskal‒Wallis (KW), analysis of variance (ANOVA), recursive feature elimination (RFE), and Relief methods were used for subregion selection. The classifiers including Gaussian process (GP), support vector machine (SVM), linear discriminant analysis (LDA), and logistic regression via Lasso (LR Lasso). Clinicopathological and conventional MR parameters were obtained. Diagnostic performance was evaluated using AUC and compared with DeLong’s test. Kaplan‒Meier method and multivariable analysis were used for evaluating the performance of prediction model. Results DR-CSI-based equipartition models demonstrated excellent interobserver agreement (ICC: 0.86–0.99). The equipartition method (6*6) combined with KW feature selection and GP classifier achieved the highest diagnostic performance in distinguishing patients with metastatic RCC from those without metastatic RCC, with an AUC of 0.87, significantly outperforming clinicopathological and conventional MR parameters (vs. age, P < 0.001; vs. tumor diameter, P = 0.002; vs. WHO/ISUP grade, P < 0.001; vs. ADC, P = 0.001; vs. T2 value, P = < 0.001). The 6*6 model could effectively predict the recurrence of patients in cohort 2 (P = 0.005), whereas the other models could not. Additionally, the 6*6 model might serve as an independent predictive factor for recurrence in RCC. Conclusions DR-CSI-based 6*6 model combined with KW and GP may assess the aggressiveness of RCC and have great promise in predicting prognostic risk stratification. Renal Cell Carcinoma Diffusion‒Relaxation Correlation Spectroscopic Imaging Prognosis Prediction Model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Background The incidence of renal cell carcinoma (RCC) has increased by approximately 1.4% annually over the past decade( 1 , 2 ). The prognosis of RCC varies and is related to several factors( 3 , 4 ). Postoperative targeted adjuvant therapy can benefit high-risk patients by reducing the likelihood of recurrence, whereas treating patients without a risk of recurrence may lead to an unnecessary economic burden and adverse reactions( 5 ). Therefore, highly efficient predictive models would be greatly valuable for stratification of patient. Clinically, clinicopathological prognostic models, particularly the tumor-node-metastasis (TNM) staging system, are commonly used for predicting the outcomes of RCC patients. However, the TNM stage reflects only the tumor's macroscopic anatomy and cannot determine whether metastasis has occurred in advance. As a result, even patients with the same TNM stage may have different risks of recurrence( 6 , 7 ). Several pathological components, including sarcomatoid/rhabdomyoid morphology, necrosis, and intravascular invasion, correlate with survival outcomes ( 8 , 9 ). Since the biological behavior associated with these components is complex, the analysis of different components is time-consuming and difficult prior to surgery. Additionally, for relatively large tumors, the focal nature of the analysis makes it difficult to represent the overall characteristics of the tumor accurately. Quantitative magnetic resonance imaging (MRI) can compensate for the partial sampling involved in pathological analysis and better reflect the histological information of the tissue( 10 , 11 ). However, conventional MRI parameters are based on single-dimensional parameters. Moreover, the limited spatial resolution of MRI makes accurate characterization of the tissue microstructure and microenvironment using voxel-by-voxel or region of interest (ROI)-averaged quantitative MRI metrics challenging. In recent years, significant advancements have been made in the field of the novel diffusion‒relaxation correlation spectroscopic imaging (DR-CSI), which couples the diffusion and relaxation times and allows assessment of tissue information within a voxel using an inverse Laplace algorithm( 12 ). This imaging method allows independent tissue components to be detected as peaks in the DR-CSI spectra and mapped onto the corresponding MR images. DR-CSI has been applied ex vivo and in vivo, demonstrating potential in the identification of distinct tissue components or compartments while overcoming the limitations of the imaging resolution and providing greater insights into the tissue microenvironment than conventional quantitative metrics( 13 – 18 ). Thus, in theory, compared with other imaging modalities DR-CSI could better differentiate patients with good and poor outcomes through improved classification of prognosis-related pathological information. Therefore, this study aimed to evaluate the feasibility of DR-CSI in assessing the aggressiveness of RCC and to develop and validate a practical model for differentiating RCC patients with different clinical outcomes. Methods This study was approved by the Ethics Committee of xxx (institutional ethics approval number: xxx). Participants Patients with renal tumors who underwent renal MRI and chest CT at xxx hospital from November 2020 to October 2023 were enrolled in this study. All patients were examined on a 3.0 T MR scanner (Magnetom Prisma, Siemens Healthineers, Erlangen, Germany), Details on the MR protocols are presented in Table 1 . Table 1 Protocols of magnetic resonance sequences Parameters T1-weighted imaging T2-weighted imaging T2-weighted imaging DR-CSI Plane Axial Axial Coronal Axial FOV (mm 2 ) 380*308 380*380 400*400 380*283 Respiratory control Breath-hold Trigger Breath-hold Free-breathing TR (ms) 3.97 3650 363 2100 TE (ms) 2.52 92 96 50/80/110/140/180/200 Image matrix 320*182 384*384 320*256 268*200 Bandwidth (Hz/pixel) 1040 723 710 710 Flip angle 9 76 96 90 Acquisition time 15 s 3 min 3 s 15 s 9 min 25 s b-values (s/mm2 ) / / / 0/150/400/800/1200/1500 Note: DR-CSI = diffusion-relaxation correlation spectroscopic imaging; FOV = field of view; TE = echo time; TR = repetition time. The inclusion criteria were as follows: 1) clinical T1-2 stage disease and MR examinations including both routine and DR-CSI sequences prior to any treatment; 2) pathologically confirmed RCC; 3) an interval between the pathologic result and MR examination within 4 weeks; and 4) tumor diameter > 2 cm as measured on axial T2-weighted images. The exclusion criteria were as follows: 1) poor MR image quality due to motion, respiratory or metal artifacts (n = 8); 2) an incomplete DR-CSI scan (n = 4); and 3) inability to identify solid tumor components on the MR images (n = 4). Metastasis was defined as regional lymph node metastasis, postoperative local ipsilateral recurrence, or distant metastasis( 19 ). The follow-up time was calculated from the date of surgery to the date of the outcome of interest (metastasis or death from RCC) or the date of the last follow-up. The last follow-up period was at least 1.5 years. Finally, 140 patients were enrolled in this study, including 24 with metastatic disease and 116 with non-metastasic disease according to the preoperative examination. Postoperative follow-up was performed for the 116 non-metastasic patients, among whom 13 patients who were lost to follow-up were excluded. Among the 103 patients who were followed up postoperatively, 24 non-metastasic patients were matched via propensity score matching (PSM) with 24 preoperative metastatic patients, forming cohort 1 for developing models to distinguish between metastatic and non-metastasic RCC, details for PSM were shown in Supplementary Material 1. The ability of the model to predict RCC recurrence was assessed with the data of the remaining 66 non-metastasic patients and 13 metastatic patients, collectively considered cohort 2. Figure 1 shows the flowchart of this study. For all patients, the RCC subtypes were confirmed by pathology. The histological specimens were obtained through radical or partial nephrectomy, then sliced and stained with hematoxylin and eosin (H&E). The World Health Organization (WHO)/International Society of Urological Pathology (ISUP) grading system was used as the reference standard, although some subtypes are not considered clinically important in the grading system( 20 , 21 ). The patients were classified into 2 groups according to the grades of their RCC subtypes: the low-grade group (WHO/ISUP grades I and II) and the high-grade group (WHO/ISUP grades III and IV). The pathological subtypes included clear cell (ccRCC) and nonccRCC. Image Postprocessing and Quantification Strategy ROIs of the tumor region were manually delineated by two radiologists (with 5 and 16 years of experience in abdominal MRI) using ITK-SNAP (version 3.8.0; http://www.itksnap.org/pmwiki/pmwiki.php ) software( 22 ). Details information were shown in Supplementary Material 2. Image processing and analysis was conducted using MATLAB software version R2022b (The MathWorks, Natick, MA, USA). Details of DR-CSI spectrum fitting were shown in Supplementary Material 3. Data Analysis and Model Development For each equipartition model developed on the basis of the data from cohort 1, the intraclass correlation coefficient (ICC) was calculated to evaluate reproducibility, and subregions exhibiting ICC values < 0.6 were excluded prior to data analysis. We used Feature Explorer software (FAE, v0.5.15), which was developed using the Python programming language (3.7.6) ( https://github.com/salan668/FAE)(23) , to assess the included subregions for each of the equipartition models and establish the optimal prediction model. Subregion Selection First, a computer-generated random dataset was used to construct each model. The dataset was divided into two distinct sets: a training set, comprising 70% of the dataset (n = 34), and the independent test set, comprising the remaining 30% (n = 14). To address imbalances in the training dataset, we upsampled the data by repeating random subsets of data to achieve balance between positive and negative samples. The dataset was normalized using the Z score normalization method. Second, owing to the high dimensionality of the subregion space, we used the Pearson correlation coefficient (PCC) method for dimensionality reduction. Specifically, if the PCC value exceeded 0.99, one of the subregions was randomly removed. Finally, the Kruskal‒Wallis (KW), analysis of variance (ANOVA), recursive feature elimination (RFE), and Relief methods were used for subregion selection( 24 ). Classifiers and Model Evaluation The classifiers performance were assessed by building models on the basis of 4 machine learning (ML) algorithms based on Python code with scikit-learn library ( https://scikit-learn.org/ ), including Gaussian process (GP), support vector machine (SVM), linear discriminant analysis (LDA), and logistic regression via Lasso (LR Lasso). The results were evaluated using a leave-one-out cross-validation (LOOCV) strategy. Specifically, using LOOCV, training sets were created by taking all the samples except one, which was used as the validation set. The area under the curve (AUC) was calculated for each test condition to assess the classifiers performance of the model. The workflow diagram of the data processing is depicted in Fig. 2 . Statistical Analysis Statistical analysis was performed with SPSS version 26.0 (IBM Corp., Armonk, NY, USA) and FAE version 0.5.15 (East China Normal University, Shanghai, China). A P value < 0.05 was considered to indicate statistical significance. Descriptive statistics (means ± standard deviations) are reported for measurement data among the clinical characteristics. For comparisons between groups, the independent-samples t test was applied for normally distributed data, whereas the Mann‒Whitney U test was used for nonnormally distributed data. Counting data among the clinical characteristics are presented as frequencies, and between-group comparisons were conducted using the chi-square test. The follow-up times are summarized as medians and interquartile ranges (IQRs). The ICC values were interpreted as poor, acceptable, moderate, good and excellent for values 0.9, respectively. The diagnostic performance of individual clinicopathological and conventional MRI parameters and each model were compared with DeLong’s test. The ability of the different models to predict postoperative recurrence in patients with RCC was assessed with the Kaplan‒Meier (KM) method. The predictive values of clinicopathological parameters, MRI parameters and the equipartition model for postoperative RCC recurrence were first assessed with univariable logistic regression analysis. Variables with P values < 0.05 in the univariable analyses were included in a multivariable logistic regression model via stepwise forward selection. Odds ratios (ORs) with 95% confidence intervals (95% CIs) were computed to report the results of the logistic regression analyses. Results Clinical and Pathological Information The clinical charateristic of all patients in cohort 1 and cohort 2 are given in Supplementary Table 1. Among the 127 RCC patients included in this study, 102 had ccRCC, 15 had pRCC, and 10 had other subtypes (6 chromophobe renal cell carcinoma, 3 TFE3-rearranged renal cell carcinoma, and 1 each mucinous tubular and spindle cell carcinoma). The mean age was 57.6 ± 13.3 years, and the mean tumor diameter was 5.6 ± 2.4 cm. Among the 127 patients with ccRCC or nonccRCC, 75 had WHO/ISUP grades 1–2 disease and were classified into the low-grade group, whereas 52 had WHO/ISUP grades 3–4 disease and were classified into the high-grade group. Comparisons of the clinicopathological data of patients in cohort 1 before and after PSM are presented in Table 2 . Before PSM, there were significant differences in tumor diameter (P < 0.001) and WHO/ISUP grade (P < 0.001) between metastatic and non-metastasic RCC patients. After PSM, these differences were no longer significant (tumor diameter, P = 0.121; WHO/ISUP grade, P = 0.928). In cohort 2, with a median follow-up of 931 days (IQR: 613–1058), among the 79 patients who survived at the last follow-up, 13 patients experienced recurrence, with a median follow-up of 317 days (IQR: 176–686), and none died. Table 2 Clinicopathological characteristics before and after PSM Before PSM After PSM Metastasis (n = 24) Non-metastasis (n = 90) P value Non-metastasis (n = 24) P value Age(years) 59.2 ± 13.8 56.9 ± 13.6 0.458 58.4 ± 14.1 0.837 Tumor diameter (cm) 7.4 ± 2.4 4.5 ± 2.0 < 0.001 6.4 ± 2.1 0.121 Gender 0.279 0.505 Male 19(79) 61(68) 17(71) Female 5( 21 ) 29(32) 7( 29 ) Pathological Subtype 0.277 0.505 ccRCC 17(71) 75(83) 19(80) Non-ccRCC 7( 29 ) 15( 17 ) 5( 20 ) WHO/ISUP Grade < 0.001 0.928 High-grade 17(71) 26( 29 ) 18(76) Low-grade 7( 29 ) 64(71) 6( 24 ) Note: Data are presented as mean ± standard deviation or number (percentage); PSM = propensity score matching; ccRCC = clear cell renal cell carcinoma. Diagnostic Performance of the Equipartition Models In cohort 1, the agreement between the 2 radiologists was assessed at the level of the subregions in each equipartition model. The overall ICCs from subregions 2*2 to 9*9 were 0.99, 0.95, 0.94, 0.91, 0.90, 0.87, 0.88, and 0.86, respectively, indicating that the ICCs decreased as the number of subregions increased. All equipartition models showed good to excellent interobserver agreement. Thus, all the results are based on the first reader's measurements. Multiple models were developed in the data of cohort 1 using different combinations of feature selection methods and classifiers. The ROC curves of all the equipartition models constructed with different pipelines in distinguishing patients with metastatic RCC from those without metastatic RCC are shown in Fig. 3 , and the AUC values are shown in Supplementary Table 2. For each equipartition method, the highest AUCs were 0.60, 0.71, 0.81, 0.82, 0.87, 0.72, 0.76, and 0.85 for models 2*2, 3*3, 4*4, 5*5, 6*6, 7*7, 8*8, and 9*9, respectively. The subregion selection methods for the above models differed, but except for 3*3, all the other models were developed on the basis of the GP classifier. Among them, the highest AUC values obtained for the 4*4, 5*5, 6*6 and 9*9 models were greater than 0.8, among which the 6*6 model yielded the highest AUC value of 0.87. The 4*4 model was constructed using a pipeline involving KW subregion selection and the GP classifier with 4 subregions; the 5*5 model was constructed using a pipeline involving Relief subregion selection and the GP classifier with 9 subregions; the 6*6 model was constructed using the pipeline of KW subregion selection and the GP classifier with 19 subregions; and the 9*9 model was constructed using a pipeline involving KW feature selection and the GP classifier with 32 subregions. Figure 4 shows the P value of each subregion in all equipartition models, the ROC curves and subregion contributions of the models with AUC value higher than 0.8 are also shown. The diagnostic performance of the 6*6 model in distinguishing between metastatic and non-metastasic RCC patient was significantly superior to that of the 2*2, 3*3 and 7*7 models (P < 0.001; P = 0.031; and P = 0.007, respectively), whereas that of the 4*4, 5*5, 8*8 and 9*9 was not significantly different from that of the 6*6 model (P = 0.086; P = 0.168; P = 0.086; and P = 0.311, respectively). In addition, for the 6*6 model, under each subregion selection method (ANOVA, KW, Relief and RFE), construction with the GP classifier consistently yielded the highest AUCs (0.85, 0.87, 0.79 and 0.81, respectively). Typical examples of metastatic and non-metastasic RCC DR-CSI spectra are shown in Fig. 5 . Comparison of Diagnostic Performance between the Equipartition Models and Other Parameters We compared the diagnostic performance of the equipartition models with that of other groups of parameters, including clinicopathological parameters and conventional MR parameters, and the corresponding P values are shown in Table 3 . The AUCs of the clinical/pathological parameters of age, tumor diameter and WHO/ISUP grade were 0.52, 0.74 and 0.52, respectively. The AUCs of conventional MR parameters, including the ADC and T2 value, were 0.62 and 0.52, respectively. The highest AUCs achieved by the 6*6 and 9*9 models were significantly greater than the AUC values of all clinicopathological and conventional MR parameters (6*6 vs. age, P < 0.001; 6*6 vs. tumor diameter, P = 0.002; 6*6 vs. WHO/ISUP Grade, P < 0.001; 6*6 vs. ADC, P = 0.001; 6*6 vs. T2, P = < 0.001; 9*9 vs. age, P < 0.001; 9*9 vs. tumor diameter, P = 0.013; 9*9 vs. WHO/ISUP Grade, P < 0.001; 9*9 vs. ADC, P = 0.003; 6*6 vs. T2, P = 0.001). However, the performance of the 2*2 model was not significantly different from that of any of the other parameters. The AUCs of the best-performing 4*4, 5*5, 6*6 and 9*9 models were significantly greater than the AUCs of the ADC and T2 value (4*4 vs. ADC, P = 0.025; 5*5 vs. ADC, P = 0.037; 6*6 vs. ADC, P = 0.001; 9*9 vs. ADC, P = 0.003; 4*4 vs. T2, P = 0.022; 5*5 vs. T2, P = 0.008; 6*6 vs. T2, P < 0.001; 9*9 vs. T2, P = 0.001). Table 3 Comparison the diagnostic performance between equipartition model and clinicopathological or conventional MRI parameters (P value) Equipartition models (P value) 2*2 3*3 4*4 5*5 6*6 7*7 8*8 9*9 Clinicopathological parameters Age 0.165 0.030 0.006 0.003 < 0.001 0.039 0.004 < 0.001 Tumor diameter 0.942 0.336 0.131 0.076 0.002 0.308 0.321 0.013 WHO/ISUP Grade 0.109 0.028 0.001 < 0.001 < 0.001 0.005 0.015 < 0.001 Conventional MRI parameters ADC 0.637 0.105 0.025 0.037 0.001 0.122 0.165 0.003 T2 0.223 0.055 0.022 0.008 < 0.001 0.067 0.022 0.001 Note: ADC = apparent diffusion coefficient. Survival Risk Stratification Based on the Diagnostic Models To assess the ability of each model to predict the postoperative recurrence of RCC, an additional 79 patients who successfully completed follow-up were included in cohort 2. The clinical data of the patients in cohort 2 are given in Supplementary Table 3. The data processing followed the same procedures as that used for cohort 1. We further verified the performance of the models in cohort 1 that achieved the highest AUC values > 0.8, including the 4*4, 5*5, 6*6 and 9*9 models, with the optimal cutoff values of each equipartition model, through which all patients were divided into low- and high-risk groups according to the results of the DR-CSI model. Among the 79 follow-up patients, 50, 29, 28 and 12 patients were in the high-risk group according to the 4*4, 5*5, 6*6 and 9*9 models, respectively. The high-risk patients identified with the 6*6 model had a significantly worse outcome than the corresponding low-risk patients (P = 0.005), whereas the differentiation produced by the 4*4, 5*5 and 9*9 models did not result in significant differences in outcome between the two risk group (P = 0.096, P = 0.056 and P = 0.894, respectively). The KM curves of RCC patients at different risk stratifications are shown in Fig. 6 . Moreover, univariable and multivariable logistic regression revealed that the results indicated that the 6*6 model and WHO/ISUP grade were independent factors influencing postoperative recurrence in RCC patients (P = 0.002 and P = 0.001, respectively); the detailed results are shown in Table 4 . Table 4 The value of each parameters in predicting postoperative recurrence of renal cell carcinoma with univariate and multivariate logistic regression Parameters Univariate Multivariate OR+(95% CI) p OR+(95% CI) p 6*6 model 12.60(3.25–48.92) < 0.001 15.21(2.71–85.31) 0.002 WHO/ISUP Grade 16.31(4.06–65.53) < 0.001 19.52(3.47-109.81) 0.001 Tumor diameter 1.73(1.28–2.34) 0.001 - - Age 1.02(0.97–1.07) 0.408 - - ADC 1.00(0.99-1.00) 0.044 - - T2 1.00(0.99-1.00) 0.842 - - Note: OR = Odds Ratios; 95% CI = 95% confidence intervals; ADC = apparent diffusion coefficient. Discussion This study presents evidence supporting the potential use of DR-CSI in evaluating the clinical outcomes of RCC patients. The results demonstrate the feasibility and adaptability of the equipartition method (6*6) combined with KW feature selection and GP classifier for leveraging the spectral information obtained with DR-CSI. Further analysis suggested that models based on DR-CSI data outperformed clinicopathological and conventional MRI parameters in distinguishing patients with different risks of recurrence, and its output may serve as an independent predictive factor of recurrence in RCC. Our findings suggest that DR-CSI-based models could have excellent performance in predicting recurrence in RCC patients and may serve as a valuable tool for stratifying patients on the basis of their clinical outcomes. Additionally, it could play a fundamental role in identifying patients who can benefit from postoperative adjuvant targeted therapy and immunotherapy while avoiding unnecessary treatment for patients who are unlikely to experience recurrence, helping reduce financial burdens, minimize potential adverse effects, and prevent overestimation of the efficacy of adjuvant therapy( 5 ). The effectiveness of DR-CSI largely stems from its reliance on pathological features based on prior knowledge( 12 ). However, we propose that DR-CSI can capture valuable information beyond the traditional pathology assessments of RCC. Traditional pathology assessments, such as tumor grading, provide insights into only certain aspects of tumor histology and are unable to fully integrate general histopathological and pathophysiological features. In contrast, imaging modalities may offer advantages in specific scenarios like component like epithelium, stroma, et al. On the basis of the results of our study, DR-CSI extends the scope of prognostic prediction by providing additional insights into the outcomes of RCC patients. Futhermore, its underlying pathological significance warrants further investigation in future research. Moreover, DR-CSI demonstrated superior performance in predicting recurrence with respect to the traditional ADC and T2 values. We believe that this is because the parameters derived from the monoexponential model primarily reflect signal decay at specific TE or b values, whereas in reality, the relaxation and diffusion processes occur within a more complex multidimensional space. As a result of this complexity, the predictive ability of traditional parameters is worse than that of DR-CSI. Additionally, clinical, anatomical, and geometric factors have been reported to play a role in prognosis prediction( 25 – 27 ). In this study, the DR-CSI-based model also outperformed most clinical, anatomical, and geometric factors in predicting recurrence, further demonstrating the advantages of this imaging modality in clinical application. Importantly, however, our study applied specific inclusion criteria that may have influenced the predictive performance of these features. As a result, further validation in diverse cohorts with different clinical and anatomical characteristics is needed to confirm these findings. The application of DR-CSI spectral information remains an important topic of discussion. Traditionally, individual peaks identified within the DR-CSI spectrum have been linked to tissue microstructure to improve their clinical utility. However, this approach presents significant challenges in clinical practice: obtaining high-quality spectra is not always feasible, distinguishing closely positioned or overlapping peaks is difficult, and in many cases, pathological or physiological evidence to clarify the significance of specific peaks is lacking. To tackle these challenges, multiple strategies have been proposed to partition DR-CSI spectra for clinical use ( 15 , 17 , 18 ). Yet existing methods still yield limited spectral information and lack a validated basis for defining regional boundaries. Subsequently, Luo et al. further explored an equalization method that divides the spectrum into multiple, equal subregions( 28 ). However, it was unclear whether this approach could be universally applied, and the optimal choice of feature selection and classifier in modeling were not fully explored. In this study, we identified that the pipeline integrating KW subregion selection with a GP classifier was a potentially optimal strategy. Compared with other feature selection methods and classifiers, this combination has several notable advantages. Specifically, unlike parametric approaches, the KW-GP pipeline is well suited for capturing nonlinear relationships, thereby improving classification accuracy. The KW test, which is based on data ranks rather than absolute values, is robust to outliers, allowing more reliable identification of discriminative features and minimizing the impact of extrema. Furthermore, the KW method allows direct comparisons among three or more groups simultaneously, leading to highly efficient feature selection in multiclass radiomic analyses( 29 ). Importantly, GP classifiers possess inherent advantages in small-sample learning scenarios, maintaining generalizability while mitigating the risk of overfitting( 30 ). Collectively, these characteristics make the KW-GP pipeline particularly advantageous for the data characteristic and analytical requirements of this study. However, this study had several limitations. First, given the single-center nature of the study, our findings require validation through larger-scale, independent, prospective multicenter studies to assess their generalizability. Second, given the novelty of this technology, the follow-up period in our cohort was relatively limited. Investigating the value of DR-CSI in the long term could be particularly insightful. Third, while the DR-CSI parameters demonstrated significant predictive value, incorporating additional feature parameters into the model should be considered in future research. Conclusion In conclusion, this study highlights the potential of DR-CSI as a novel imaging technique for assessing the aggressiveness of RCC and predicting patient outcomes. Our findings demonstrate that DR-CSI outperforms conventional clinicopathological and MRI parameters in distinguishing between patients with high and low risks of disease recurrence. Abbreviations RCC Renal cell carcinoma TNM Tumor-node-metastasis MRI Magnetic resonance imaging ROI Region of interest DR-CSI Diffusion‒relaxation correlation spectroscopic imaging PSM Propensity score matching H&E Hematoxylin and eosin ICC Intraclass correlation coefficient PCC Pearson correlation coefficient KW Kruskal‒Wallis ANOVA Analysis of variance RFE Recursive feature elimination GP Gaussian process SVM Support vector machine LDA Linear discriminant analysis LR Lasso Logistic regression via Lasso LOOCV Leave-one-out cross-validation AUC The area under the curve Declarations Ethics approval and consent to participate This study was performed in line with the principles of the Declaration of Helsinki. The Institutional Review Board of the Ethics Committee of Renji Hospital School of Medicine, Shanghai Jiao Tong University (institutional ethics approval number: LY2025-058-B) approved the research protocol for this retrospective study and waived the requirement for written informed consent from the patients. Consent for publication Not applicable. Availability of data and materials The data of this article can be obtained with the consent of the corresponding author. Competing interests The authors declare no competing interests. Funding This research was supported by National Natural Science Foundation of China (NSFC 82371912), Shanghai Science and Technology Committee of Shanghai Municipality (grant no. 23Y11903900), and Transverse Project from Renji Hospital, School of Medicine, Shanghai Jiao Tong University (grant no. RJKY22-004). Authors' contributions MZ and YL : Data Curation, Formal analysis, Validation, Writing - Original Draft GW and GL : Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Supervision, Writing - Review & Editing XW, JC and SP : Resources All authors reviewed the manuscript. Acknowledgements None. References Hirsch MS, Signoretti S, Dal Cin P. Adult Ren Cell Carcinoma Surg Pathol Clin. 2015;8(4):587–621. Znaor A, Lortet-Tieulent J, Laversanne M, Jemal A, Bray F. International Variations and Trends in Renal Cell Carcinoma Incidence and Mortality. Eur Urol. 2015;67(3):519–30. Teloken PE, Thompson RH, Tickoo SK, et al. Prognostic Impact of Histological Subtype on Surgically Treated Localized Renal Cell Carcinoma. J Urol. 2009;182(5):2132–6. Leibovich BC, Lohse CM, Crispen PL, et al. Histological Subtype is an Independent Predictor of Outcome for Patients With Renal Cell Carcinoma. J Urol. 2010;183(4):1309–16. Ravaud A, Motzer RJ, Pandha HS, et al. Adjuvant Sunitinib in High-Risk Renal-Cell Carcinoma after Nephrectomy. N Engl J Med. 2016;375(23):2246–54. Yoshida T, Ohe C, Tsuzuki T, et al. Clinical impact of segmental renal vein invasion on recurrence in patients with clinical T1 renal cell carcinoma undergoing partial nephrectomy. Int J Clin Oncol. 2019;25(3):464–71. Shah PH, Moreira DM, Patel VR, et al. Partial Nephrectomy is Associated with Higher Risk of Relapse Compared with Radical Nephrectomy for Clinical Stage T1 Renal Cell Carcinoma Pathologically Up Staged to T3a. J Urol. 2017;198(2):289–96. Cheville JC, Lohse CM, Weaver AL, et al. Sarcomatoid renal cell carcinoma: an examination of underlying histologic subtype and an analysis of associations with patient outcome. Am J Surg Pathol. 2004;28(4):435–41. Klatte T, Said JW, de Martino M, et al. Presence of Tumor Necrosis is Not a Significant Predictor of Survival in Clear Cell Renal Cell Carcinoma: Higher Prognostic Accuracy of Extent Based Rather Than Presence/Absence Classification. J Urol. 2009;181(4):1558–64. Oh J, Lee JM, Park J et al. Hepatocellular Carcinoma: Texture Analysis of Preoperative Computed Tomography Images Can Provide Markers of Tumor Grade and Disease-Free Survival. Korean J Radiol. 2019; 20(4). Zhou H, Vallières M, Bai HX, et al. MRI features predict survival and molecular markers in diffuse lower-grade gliomas. Neurooncology. 2017;19(6):862–70. Kim D, Doyle EK, Wisnowski JL, Kim JH, Haldar JP. Diffusion-relaxation correlation spectroscopic imaging: A multidimensional approach for probing microstructure. Magn Reson Med. 2017;78(6):2236–49. Zhang Z, Wu HH, Priester A, et al. Prostate Microstructure in Prostate Cancer Using 3-T MRI with Diffusion-Relaxation Correlation Spectrum Imaging: Validation with Whole-Mount Digital Histopathology. Radiology. 2020;296(2):348–55. Benjamini D, Priemer DS, Perl DP, Brody DL, Basser PJ. Mapping astrogliosis in the individual human brain using multidimensional MRI. Brain. 2023;146(3):1212–26. Wei X, Zhu L, Zeng Y et al. Detection of prostate cancer using diffusion-relaxation correlation spectrum imaging with support vector machine model – a feasibility study. Cancer Imaging. 2022; 22(1). Liu F, Hu W, Sun Y, et al. Exploration of Interstitial Fibrosis in Chronic Kidney Disease by Diffusion-Relaxation Correlation Spectrum MR Imaging: A Preliminary Study. J Magn Reson Imaging. 2022;58(2):415–26. Dai Y, Hu W, Wu G, et al. Grading Clear Cell Renal Cell Carcinoma Grade Using Diffusion Relaxation Correlated MR Spectroscopic Imaging. J Magn Reson Imaging. 2023;59(2):699–710. Dai Y, Zhu M, Hu W, et al. To characterize small renal cell carcinoma using diffusion relaxation correlation spectroscopic imaging and apparent diffusion coefficient based histogram analysis: a preliminary study. Radiol Med. 2024;129(6):834–44. Elkassem AA, Allen BC, Sharbidre KG, Rais-Bahrami S, Smith AD. Update on the Role of Imaging in Clinical Staging and Restaging of Renal Cell Carcinoma Based on the AJCC 8th Edition, From the AJR Special Series on Cancer Staging. Am J Roentgenol. 2021;217(3):541–55. Xiao Q, Yi X, Guan X, et al. Validation of the World Health Organization/International Society of Urological Pathology grading for Chinese patients with clear cell renal cell carcinoma. Translational Androl Urol. 2020;9(6):2665–74. Moch H. WHO-ISUP-Graduierungssystem für Nierenkarzinome. Pathologe. 2016;37(4):355–60. Yushkevich PA, Gao Y, Gerig G. ITK-SNAP: An interactive tool for semi-automatic segmentation of multi-modality biomedical images. 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)2016; 3342-5. Dimitriadis SI, Song Y, Zhang J et al. FeAture Explorer (FAE): A tool for developing and comparing radiomics models. PLoS ONE. 2020; 15(8). Urbanowicz RJ, Meeker M, La Cava W, Olson RS, Moore JH. Relief-based feature selection: Introduction and review. J Biomed Inform. 2018;85:189–203. Frank I, Blute M, Cheville J, Lohse C, Weaver A, Zincke H. An Outcome Prediction Model for Patients with Clear Cell Renal Cell Carcinoma Treated with Radical Nephrectomy Based on Tumor Stage, Size, Grade and Necrosis: The Ssign Score. J Urol. 2002;168(6):2395–400. Leibovich BC, Blute ML, Cheville JC, et al. Prediction of progression after radical nephrectomy for patients with clear cell renal cell carcinoma. Cancer. 2003;97(7):1663–71. Zisman A, Pantuck AJ, Dorey F, et al. Improved prognostication of renal cell carcinoma using an integrated staging system. J Clin Oncol. 2001;19(6):1649–57. Luo Y, Zhu M, Wei X, et al. Investigation of clear cell renal cell carcinoma grades using diffusion-relaxation correlation spectroscopic imaging with optimized spatial-spectrum analysis. Br J Radiol. 2024;97(1153):135–41. Saeys Y, Inza I, Larrañaga P. A review of feature selection techniques in bioinformatics. Bioinformatics. 2007;23(19):2507–17. Maniruzzaman, Rahman MJ, Benojir A, et al. Statistical characterization and classification of colon microarray gene expression data using multiple machine learning paradigms. Comput Methods Programs Biomed. 2019;176:173–93. Additional Declarations No competing interests reported. Supplementary Files SupplementalMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7287776","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":503238961,"identity":"e4224b2b-b76a-472b-aa1a-d740419c57d5","order_by":0,"name":"Mengying Zhu","email":"","orcid":"","institution":"Renji Hospital","correspondingAuthor":false,"prefix":"","firstName":"Mengying","middleName":"","lastName":"Zhu","suffix":""},{"id":503238962,"identity":"0e386ff7-5b96-4682-be78-6ddf7b6cfe30","order_by":1,"name":"Yuansheng Luo","email":"","orcid":"","institution":"Renji Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuansheng","middleName":"","lastName":"Luo","suffix":""},{"id":503238963,"identity":"1ed6584b-ac35-4478-84f2-728bee3bfbf4","order_by":2,"name":"Xiaobin Wei","email":"","orcid":"","institution":"Renji Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaobin","middleName":"","lastName":"Wei","suffix":""},{"id":503238964,"identity":"8f1915cb-c3cb-48bf-a1a1-ca8c6d5779e4","order_by":3,"name":"Jingli Chen","email":"","orcid":"","institution":"Renji Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jingli","middleName":"","lastName":"Chen","suffix":""},{"id":503238965,"identity":"80c95219-7895-424a-a31d-571f943f6ef5","order_by":4,"name":"Shihang Pan","email":"","orcid":"","institution":"Renji Hospital","correspondingAuthor":false,"prefix":"","firstName":"Shihang","middleName":"","lastName":"Pan","suffix":""},{"id":503238966,"identity":"e20be43c-1385-4952-b0c7-b044255d3d76","order_by":5,"name":"Guangyu Wu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAwklEQVRIiWNgGAWjYBADZjb2xsYHH0jTwnO42XAGafZIpLdJcxCjUD4ix0y6ouIOO5/kwwZpBgY7Od0GAloMb+SYSZ4584yZTTqxwbiAIdnY7AAhLTNyt0k2th0Ga0mewXAgcRtxWv4BtUgebDjMQ4wWeQmQlgagFgnGxmaitBjwvP9s2XAMqIUnsZlxhgERfpFvT0u82VBzOFm+/fjzHx8q7OQIajG4kACmk6FcAsrBtvRDDLUjQu0oGAWjYBSMVAAAiUdBiaeVRrsAAAAASUVORK5CYII=","orcid":"","institution":"Renji Hospital","correspondingAuthor":true,"prefix":"","firstName":"Guangyu","middleName":"","lastName":"Wu","suffix":""},{"id":503238967,"identity":"841ffe25-9ed5-4e1e-968f-e4ddb6cfa9df","order_by":6,"name":"Guiqin Liu","email":"","orcid":"","institution":"Renji Hospital","correspondingAuthor":false,"prefix":"","firstName":"Guiqin","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2025-08-04 06:53:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7287776/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7287776/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":89660441,"identity":"f86e8f80-0ea6-4746-a755-117161506f96","added_by":"auto","created_at":"2025-08-22 11:05:59","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":335113,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the study design.Note: PSM = Propensity Score Matching\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/a652b295ddebccc1a20d92c2.png"},{"id":89660440,"identity":"b3a97229-055e-4260-9ce0-7e77fdc03360","added_by":"auto","created_at":"2025-08-22 11:05:59","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":859219,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of the data processing. Illustration of the processes for diffusion-relaxation correlation spectrum imaging (DR-CSI), equipartition model development combined with various feature selection methods and classifiers, and model performance validation for differentiating RCC patients with different clinical outcomes.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/9549c81a7c0b13fe96d39b68.png"},{"id":89660449,"identity":"67034d32-77c1-4624-ad2a-17d647616c62","added_by":"auto","created_at":"2025-08-22 11:06:00","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":240395,"visible":true,"origin":"","legend":"\u003cp\u003eThe Areas under the curve (AUCs) of testing datasets in each equipartition model using various pipelines, containing four subregion selections and four Machine Learning(ML) algorithms. Subregion selections using (a) analysis of variance (ANOVA), (b) Kruskal–Wallis (KW), (c) Relief, and (d) Recursive Feature Elimination (RFE). Machine Learning(ML) algorithms using Gaussian process(GP), Support Vector Machine(SVM), Linear Discriminant Analysis(LDA), and Logistic Regression Lasso(LR Lasso) classifers.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/4e0cbb1638c5d6469f2351d7.png"},{"id":89661536,"identity":"721109ab-f50a-4652-a1b0-8b44dea974a4","added_by":"auto","created_at":"2025-08-22 11:14:00","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":642528,"visible":true,"origin":"","legend":"\u003cp\u003e(a) The heatmap shows the P value of each subregion in the different equipartition models. Not Available values were represented by NA. The Receiver operating characteristic (ROC) curves and subregions contribution of the models with AUC value higher than 0.8 including (b) 4*4 model, the rank of 4 selected subregions using the pipeline of KW and GP classifier; (c) 5*5 model, the rank of 9 selected subregions using the pipeline of Relief subregion selection and GP classifier; (d) 6*6 model, the rank of 19 selected subregions using the pipeline of KW subregion selection and GP classifier; (e) 9*9 model, the rank of 32 selected subregions using the pipeline of KW subregion selection and GP classifier.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/1887538ae8070c062befe43f.png"},{"id":89660445,"identity":"e3a7b72c-1cf6-4b42-b411-df03d0660d87","added_by":"auto","created_at":"2025-08-22 11:06:00","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":595584,"visible":true,"origin":"","legend":"\u003cp\u003eTypical spectrum mapping of DR-CSI in patients with metastasis and non-metastasis RCCs. (a, c) A 55-year-old patient with pathologically confirmed grade IV pRCC in the left kidney, retroperitoneal lymph node metastases. (b, d) A 50-year-old patient with pathologically confirmed grade III ccRCC in the left kidney, without metastasis at follow-up. (a, b) From left to right: DWI with b1200, T2-weighted image and contrast-enhanced T1-weighted image (excretory phase), (c, d) D-T2 spectra for the solid part of tumor.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/b31823bba73894701b649e24.png"},{"id":89660451,"identity":"7785b9aa-27ae-455a-8d7c-85fafdffed66","added_by":"auto","created_at":"2025-08-22 11:06:00","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":352857,"visible":true,"origin":"","legend":"\u003cp\u003eThe Kaplan-Meier recurrence-free survival curves of patients with RCCs at different risks stratified by the equipartition models. (a) 4*4 model, (b) 5*5 model, (c) 6*6 model and (d) 9*9 model.\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/7b081b460c882b21c4d4ab74.png"},{"id":93104374,"identity":"675ce8b4-e8ff-4eaf-a164-1af399c31a58","added_by":"auto","created_at":"2025-10-09 06:17:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3821587,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/dad49507-6525-433a-923b-fb706f81fc92.pdf"},{"id":89661534,"identity":"c91cd6cf-3019-4918-94b5-e9c9e8e15ae2","added_by":"auto","created_at":"2025-08-22 11:13:59","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":56897,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-7287776/v1/25fa037c0941fc791ee23113.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Exploration of Prognostic Prediction Models for Renal Cell Carcinoma using Diffusion Relaxation Correlation Spectroscopic Imaging","fulltext":[{"header":"Background","content":"\u003cp\u003eThe incidence of renal cell carcinoma (RCC) has increased by approximately 1.4% annually over the past decade(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). The prognosis of RCC varies and is related to several factors(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Postoperative targeted adjuvant therapy can benefit high-risk patients by reducing the likelihood of recurrence, whereas treating patients without a risk of recurrence may lead to an unnecessary economic burden and adverse reactions(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). Therefore, highly efficient predictive models would be greatly valuable for stratification of patient.\u003c/p\u003e\u003cp\u003eClinically, clinicopathological prognostic models, particularly the tumor-node-metastasis (TNM) staging system, are commonly used for predicting the outcomes of RCC patients. However, the TNM stage reflects only the tumor's macroscopic anatomy and cannot determine whether metastasis has occurred in advance. As a result, even patients with the same TNM stage may have different risks of recurrence(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eSeveral pathological components, including sarcomatoid/rhabdomyoid morphology, necrosis, and intravascular invasion, correlate with survival outcomes (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Since the biological behavior associated with these components is complex, the analysis of different components is time-consuming and difficult prior to surgery. Additionally, for relatively large tumors, the focal nature of the analysis makes it difficult to represent the overall characteristics of the tumor accurately.\u003c/p\u003e\u003cp\u003eQuantitative magnetic resonance imaging (MRI) can compensate for the partial sampling involved in pathological analysis and better reflect the histological information of the tissue(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). However, conventional MRI parameters are based on single-dimensional parameters. Moreover, the limited spatial resolution of MRI makes accurate characterization of the tissue microstructure and microenvironment using voxel-by-voxel or region of interest (ROI)-averaged quantitative MRI metrics challenging.\u003c/p\u003e\u003cp\u003eIn recent years, significant advancements have been made in the field of the novel diffusion‒relaxation correlation spectroscopic imaging (DR-CSI), which couples the diffusion and relaxation times and allows assessment of tissue information within a voxel using an inverse Laplace algorithm(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). This imaging method allows independent tissue components to be detected as peaks in the DR-CSI spectra and mapped onto the corresponding MR images. DR-CSI has been applied ex vivo and in vivo, demonstrating potential in the identification of distinct tissue components or compartments while overcoming the limitations of the imaging resolution and providing greater insights into the tissue microenvironment than conventional quantitative metrics(\u003cspan additionalcitationids=\"CR14 CR15 CR16 CR17\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). Thus, in theory, compared with other imaging modalities DR-CSI could better differentiate patients with good and poor outcomes through improved classification of prognosis-related pathological information.\u003c/p\u003e\u003cp\u003eTherefore, this study aimed to evaluate the feasibility of DR-CSI in assessing the aggressiveness of RCC and to develop and validate a practical model for differentiating RCC patients with different clinical outcomes.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e This study was approved by the Ethics Committee of xxx (institutional ethics approval number: xxx).\u003c/p\u003e\u003cp\u003eParticipants\u003c/p\u003e\u003cp\u003ePatients with renal tumors who underwent renal MRI and chest CT at xxx hospital from November 2020 to October 2023 were enrolled in this study. All patients were examined on a 3.0 T MR scanner (Magnetom Prisma, Siemens Healthineers, Erlangen, Germany), Details on the MR protocols are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e Protocols of magnetic resonance sequences\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParameters\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eT1-weighted imaging\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eT2-weighted imaging\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eT2-weighted imaging\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eDR-CSI\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePlane\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eAxial\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAxial\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCoronal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eAxial\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eFOV (mm\u003csup\u003e2\u003c/sup\u003e )\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e380*308\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e380*380\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e400*400\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e380*283\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eRespiratory control\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBreath-hold\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTrigger\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eBreath-hold\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eFree-breathing\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eTR (ms)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.97\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3650\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e363\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e2100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eTE (ms)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.52\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e50/80/110/140/180/200\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eImage matrix\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e320*182\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e384*384\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e320*256\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e268*200\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eBandwidth (Hz/pixel)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1040\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e723\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e710\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e710\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eFlip angle\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e90\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eAcquisition time\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15 s\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3 min 3 s\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e15 s\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e9 min 25 s\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u003cp\u003eb-values (s/mm2 )\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0/150/400/800/1200/1500\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003eNote: DR-CSI\u0026thinsp;=\u0026thinsp;diffusion-relaxation correlation spectroscopic imaging; FOV\u0026thinsp;=\u0026thinsp;field of view; TE\u0026thinsp;=\u0026thinsp;echo time; TR\u0026thinsp;=\u0026thinsp;repetition time.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe inclusion criteria were as follows: 1) clinical T1-2 stage disease and MR examinations including both routine and DR-CSI sequences prior to any treatment; 2) pathologically confirmed RCC; 3) an interval between the pathologic result and MR examination within 4 weeks; and 4) tumor diameter\u0026thinsp;\u0026gt;\u0026thinsp;2 cm as measured on axial T2-weighted images. The exclusion criteria were as follows: 1) poor MR image quality due to motion, respiratory or metal artifacts (n\u0026thinsp;=\u0026thinsp;8); 2) an incomplete DR-CSI scan (n\u0026thinsp;=\u0026thinsp;4); and 3) inability to identify solid tumor components on the MR images (n\u0026thinsp;=\u0026thinsp;4).\u003c/p\u003e\u003cp\u003eMetastasis was defined as regional lymph node metastasis, postoperative local ipsilateral recurrence, or distant metastasis(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). The follow-up time was calculated from the date of surgery to the date of the outcome of interest (metastasis or death from RCC) or the date of the last follow-up. The last follow-up period was at least 1.5 years.\u003c/p\u003e\u003cp\u003eFinally, 140 patients were enrolled in this study, including 24 with metastatic disease and 116 with non-metastasic disease according to the preoperative examination. Postoperative follow-up was performed for the 116 non-metastasic patients, among whom 13 patients who were lost to follow-up were excluded. Among the 103 patients who were followed up postoperatively, 24 non-metastasic patients were matched via propensity score matching (PSM) with 24 preoperative metastatic patients, forming cohort 1 for developing models to distinguish between metastatic and non-metastasic RCC, details for PSM were shown in Supplementary Material 1. The ability of the model to predict RCC recurrence was assessed with the data of the remaining 66 non-metastasic patients and 13 metastatic patients, collectively considered cohort 2. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the flowchart of this study.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFor all patients, the RCC subtypes were confirmed by pathology. The histological specimens were obtained through radical or partial nephrectomy, then sliced and stained with hematoxylin and eosin (H\u0026amp;E). The World Health Organization (WHO)/International Society of Urological Pathology (ISUP) grading system was used as the reference standard, although some subtypes are not considered clinically important in the grading system(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). The patients were classified into 2 groups according to the grades of their RCC subtypes: the low-grade group (WHO/ISUP grades I and II) and the high-grade group (WHO/ISUP grades III and IV). The pathological subtypes included clear cell (ccRCC) and nonccRCC.\u003c/p\u003e\u003cp\u003eImage Postprocessing and Quantification Strategy\u003c/p\u003e\u003cp\u003eROIs of the tumor region were manually delineated by two radiologists (with 5 and 16 years of experience in abdominal MRI) using ITK-SNAP (version 3.8.0; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.itksnap.org/pmwiki/pmwiki.php\u003c/span\u003e\u003cspan address=\"http://www.itksnap.org/pmwiki/pmwiki.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e software(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). Details information were shown in Supplementary Material 2.\u003c/p\u003e\u003cp\u003eImage processing and analysis was conducted using MATLAB software version R2022b (The MathWorks, Natick, MA, USA). Details of DR-CSI spectrum fitting were shown in Supplementary Material 3.\u003c/p\u003e\u003cp\u003eData Analysis and Model Development\u003c/p\u003e\u003cp\u003eFor each equipartition model developed on the basis of the data from cohort 1, the intraclass correlation coefficient (ICC) was calculated to evaluate reproducibility, and subregions exhibiting ICC values\u0026thinsp;\u0026lt;\u0026thinsp;0.6 were excluded prior to data analysis. We used Feature Explorer software (FAE, v0.5.15), which was developed using the Python programming language (3.7.6) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/salan668/FAE)(23)\u003c/span\u003e\u003cspan address=\"https://github.com/salan668/FAE)(23)\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, to assess the included subregions for each of the equipartition models and establish the optimal prediction model.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eSubregion Selection\u003c/h2\u003e\u003cp\u003eFirst, a computer-generated random dataset was used to construct each model. The dataset was divided into two distinct sets: a training set, comprising 70% of the dataset (n\u0026thinsp;=\u0026thinsp;34), and the independent test set, comprising the remaining 30% (n\u0026thinsp;=\u0026thinsp;14). To address imbalances in the training dataset, we upsampled the data by repeating random subsets of data to achieve balance between positive and negative samples. The dataset was normalized using the Z score normalization method. Second, owing to the high dimensionality of the subregion space, we used the Pearson correlation coefficient (PCC) method for dimensionality reduction. Specifically, if the PCC value exceeded 0.99, one of the subregions was randomly removed. Finally, the Kruskal‒Wallis (KW), analysis of variance (ANOVA), recursive feature elimination (RFE), and Relief methods were used for subregion selection(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eClassifiers and Model Evaluation\u003c/h3\u003e\n\u003cp\u003eThe classifiers performance were assessed by building models on the basis of 4 machine learning (ML) algorithms based on Python code with scikit-learn library (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://scikit-learn.org/\u003c/span\u003e\u003cspan address=\"https://scikit-learn.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), including Gaussian process (GP), support vector machine (SVM), linear discriminant analysis (LDA), and logistic regression via Lasso (LR Lasso). The results were evaluated using a leave-one-out cross-validation (LOOCV) strategy. Specifically, using LOOCV, training sets were created by taking all the samples except one, which was used as the validation set. The area under the curve (AUC) was calculated for each test condition to assess the classifiers performance of the model. The workflow diagram of the data processing is depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eStatistical analysis was performed with SPSS version 26.0 (IBM Corp., Armonk, NY, USA) and FAE version 0.5.15 (East China Normal University, Shanghai, China). A P value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered to indicate statistical significance. Descriptive statistics (means\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviations) are reported for measurement data among the clinical characteristics. For comparisons between groups, the independent-samples t test was applied for normally distributed data, whereas the Mann‒Whitney U test was used for nonnormally distributed data. Counting data among the clinical characteristics are presented as frequencies, and between-group comparisons were conducted using the chi-square test. The follow-up times are summarized as medians and interquartile ranges (IQRs). The ICC values were interpreted as poor, acceptable, moderate, good and excellent for values\u0026thinsp;\u0026lt;\u0026thinsp;0.6, 0.6\u0026ndash;0.7, 0.7\u0026ndash;0.8, 0.8\u0026ndash;0.9 and \u0026gt;\u0026thinsp;0.9, respectively.\u003c/p\u003e\u003cp\u003eThe diagnostic performance of individual clinicopathological and conventional MRI parameters and each model were compared with DeLong\u0026rsquo;s test. The ability of the different models to predict postoperative recurrence in patients with RCC was assessed with the Kaplan‒Meier (KM) method. The predictive values of clinicopathological parameters, MRI parameters and the equipartition model for postoperative RCC recurrence were first assessed with univariable logistic regression analysis. Variables with P values\u0026thinsp;\u0026lt;\u0026thinsp;0.05 in the univariable analyses were included in a multivariable logistic regression model via stepwise forward selection. Odds ratios (ORs) with 95% confidence intervals (95% CIs) were computed to report the results of the logistic regression analyses.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eClinical and Pathological Information\u003c/p\u003e\u003cp\u003eThe clinical charateristic of all patients in cohort 1 and cohort 2 are given in Supplementary Table\u0026nbsp;1. Among the 127 RCC patients included in this study, 102 had ccRCC, 15 had pRCC, and 10 had other subtypes (6 chromophobe renal cell carcinoma, 3 TFE3-rearranged renal cell carcinoma, and 1 each mucinous tubular and spindle cell carcinoma). The mean age was 57.6\u0026thinsp;\u0026plusmn;\u0026thinsp;13.3 years, and the mean tumor diameter was 5.6\u0026thinsp;\u0026plusmn;\u0026thinsp;2.4 cm. Among the 127 patients with ccRCC or nonccRCC, 75 had WHO/ISUP grades 1\u0026ndash;2 disease and were classified into the low-grade group, whereas 52 had WHO/ISUP grades 3\u0026ndash;4 disease and were classified into the high-grade group. Comparisons of the clinicopathological data of patients in cohort 1 before and after PSM are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Before PSM, there were significant differences in tumor diameter (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and WHO/ISUP grade (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) between metastatic and non-metastasic RCC patients. After PSM, these differences were no longer significant (tumor diameter, P\u0026thinsp;=\u0026thinsp;0.121; WHO/ISUP grade, P\u0026thinsp;=\u0026thinsp;0.928). In cohort 2, with a median follow-up of 931 days (IQR: 613\u0026ndash;1058), among the 79 patients who survived at the last follow-up, 13 patients experienced recurrence, with a median follow-up of 317 days (IQR: 176\u0026ndash;686), and none died.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e Clinicopathological characteristics before and after PSM\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e\u003cp\u003eBefore PSM\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u003cp\u003eAfter PSM\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMetastasis\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNon-metastasis\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;90)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNon-metastasis\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge(years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e59.2\u0026thinsp;\u0026plusmn;\u0026thinsp;13.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e56.9\u0026thinsp;\u0026plusmn;\u0026thinsp;13.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.458\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e58.4\u0026thinsp;\u0026plusmn;\u0026thinsp;14.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.837\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTumor diameter (cm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7.4\u0026thinsp;\u0026plusmn;\u0026thinsp;2.4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4.5\u0026thinsp;\u0026plusmn;\u0026thinsp;2.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e6.4\u0026thinsp;\u0026plusmn;\u0026thinsp;2.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.121\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.279\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.505\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e19(79)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e61(68)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e17(71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e29(32)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e7(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePathological Subtype\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.277\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.505\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eccRCC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e17(71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e75(83)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e19(80)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNon-ccRCC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWHO/ISUP Grade\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.928\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHigh-grade\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e17(71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e26(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e18(76)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLow-grade\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e64(71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e6(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003eNote: Data are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation or number (percentage); PSM\u0026thinsp;=\u0026thinsp;propensity score matching; ccRCC\u0026thinsp;=\u0026thinsp;clear cell renal cell carcinoma.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eDiagnostic Performance of the Equipartition Models\u003c/p\u003e\u003cp\u003e In cohort 1, the agreement between the 2 radiologists was assessed at the level of the subregions in each equipartition model. The overall ICCs from subregions 2*2 to 9*9 were 0.99, 0.95, 0.94, 0.91, 0.90, 0.87, 0.88, and 0.86, respectively, indicating that the ICCs decreased as the number of subregions increased. All equipartition models showed good to excellent interobserver agreement. Thus, all the results are based on the first reader's measurements.\u003c/p\u003e\u003cp\u003eMultiple models were developed in the data of cohort 1 using different combinations of feature selection methods and classifiers. The ROC curves of all the equipartition models constructed with different pipelines in distinguishing patients with metastatic RCC from those without metastatic RCC are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, and the AUC values are shown in Supplementary Table\u0026nbsp;2. For each equipartition method, the highest AUCs were 0.60, 0.71, 0.81, 0.82, 0.87, 0.72, 0.76, and 0.85 for models 2*2, 3*3, 4*4, 5*5, 6*6, 7*7, 8*8, and 9*9, respectively. The subregion selection methods for the above models differed, but except for 3*3, all the other models were developed on the basis of the GP classifier. Among them, the highest AUC values obtained for the 4*4, 5*5, 6*6 and 9*9 models were greater than 0.8, among which the 6*6 model yielded the highest AUC value of 0.87. The 4*4 model was constructed using a pipeline involving KW subregion selection and the GP classifier with 4 subregions; the 5*5 model was constructed using a pipeline involving Relief subregion selection and the GP classifier with 9 subregions; the 6*6 model was constructed using the pipeline of KW subregion selection and the GP classifier with 19 subregions; and the 9*9 model was constructed using a pipeline involving KW feature selection and the GP classifier with 32 subregions. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows the P value of each subregion in all equipartition models, the ROC curves and subregion contributions of the models with AUC value higher than 0.8 are also shown. The diagnostic performance of the 6*6 model in distinguishing between metastatic and non-metastasic RCC patient was significantly superior to that of the 2*2, 3*3 and 7*7 models (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; P\u0026thinsp;=\u0026thinsp;0.031; and P\u0026thinsp;=\u0026thinsp;0.007, respectively), whereas that of the 4*4, 5*5, 8*8 and 9*9 was not significantly different from that of the 6*6 model (P\u0026thinsp;=\u0026thinsp;0.086; P\u0026thinsp;=\u0026thinsp;0.168; P\u0026thinsp;=\u0026thinsp;0.086; and P\u0026thinsp;=\u0026thinsp;0.311, respectively). In addition, for the 6*6 model, under each subregion selection method (ANOVA, KW, Relief and RFE), construction with the GP classifier consistently yielded the highest AUCs (0.85, 0.87, 0.79 and 0.81, respectively). Typical examples of metastatic and non-metastasic RCC DR-CSI spectra are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eComparison of Diagnostic Performance between the Equipartition Models and Other Parameters\u003c/p\u003e\u003cp\u003eWe compared the diagnostic performance of the equipartition models with that of other groups of parameters, including clinicopathological parameters and conventional MR parameters, and the corresponding P values are shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The AUCs of the clinical/pathological parameters of age, tumor diameter and WHO/ISUP grade were 0.52, 0.74 and 0.52, respectively. The AUCs of conventional MR parameters, including the ADC and T2 value, were 0.62 and 0.52, respectively. The highest AUCs achieved by the 6*6 and 9*9 models were significantly greater than the AUC values of all clinicopathological and conventional MR parameters (6*6 vs. age, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; 6*6 vs. tumor diameter, P\u0026thinsp;=\u0026thinsp;0.002; 6*6 vs. WHO/ISUP Grade, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; 6*6 vs. ADC, P\u0026thinsp;=\u0026thinsp;0.001; 6*6 vs. T2, P\u0026thinsp;=\u0026thinsp;\u0026lt;\u0026thinsp;0.001; 9*9 vs. age, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; 9*9 vs. tumor diameter, P\u0026thinsp;=\u0026thinsp;0.013; 9*9 vs. WHO/ISUP Grade, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; 9*9 vs. ADC, P\u0026thinsp;=\u0026thinsp;0.003; 6*6 vs. T2, P\u0026thinsp;=\u0026thinsp;0.001). However, the performance of the 2*2 model was not significantly different from that of any of the other parameters. The AUCs of the best-performing 4*4, 5*5, 6*6 and 9*9 models were significantly greater than the AUCs of the ADC and T2 value (4*4 vs. ADC, P\u0026thinsp;=\u0026thinsp;0.025; 5*5 vs. ADC, P\u0026thinsp;=\u0026thinsp;0.037; 6*6 vs. ADC, P\u0026thinsp;=\u0026thinsp;0.001; 9*9 vs. ADC, P\u0026thinsp;=\u0026thinsp;0.003; 4*4 vs. T2, P\u0026thinsp;=\u0026thinsp;0.022; 5*5 vs. T2, P\u0026thinsp;=\u0026thinsp;0.008; 6*6 vs. T2, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; 9*9 vs. T2, P\u0026thinsp;=\u0026thinsp;0.001).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e Comparison the diagnostic performance between equipartition model and clinicopathological or conventional MRI parameters (P value)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"9\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"8\" nameend=\"c9\" namest=\"c2\"\u003e\u003cp\u003eEquipartition models (P value)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2*2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3*3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4*4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5*5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e6*6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e7*7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e8*8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e9*9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e\u003cp\u003eClinicopathological\u003c/p\u003e\u003cp\u003eparameters\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.165\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e0.030\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.006\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.003\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e0.039\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e\u003cb\u003e0.004\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTumor diameter\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.942\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.336\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.131\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.076\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.002\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.308\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.321\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e\u003cb\u003e0.013\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWHO/ISUP Grade\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.109\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e0.028\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e0.005\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e\u003cb\u003e0.015\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e\u003cp\u003eConventional\u003c/p\u003e\u003cp\u003eMRI parameters\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eADC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.637\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.105\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.025\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.037\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.122\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.165\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e\u003cb\u003e0.003\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.223\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.055\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.022\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.008\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.067\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e\u003cb\u003e0.022\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e\u003cb\u003e0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e\u003cp\u003eNote: ADC\u0026thinsp;=\u0026thinsp;apparent diffusion coefficient.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eSurvival Risk Stratification Based on the Diagnostic Models\u003c/p\u003e\u003cp\u003eTo assess the ability of each model to predict the postoperative recurrence of RCC, an additional 79 patients who successfully completed follow-up were included in cohort 2. The clinical data of the patients in cohort 2 are given in Supplementary Table\u0026nbsp;3. The data processing followed the same procedures as that used for cohort 1. We further verified the performance of the models in cohort 1 that achieved the highest AUC values\u0026thinsp;\u0026gt;\u0026thinsp;0.8, including the 4*4, 5*5, 6*6 and 9*9 models, with the optimal cutoff values of each equipartition model, through which all patients were divided into low- and high-risk groups according to the results of the DR-CSI model. Among the 79 follow-up patients, 50, 29, 28 and 12 patients were in the high-risk group according to the 4*4, 5*5, 6*6 and 9*9 models, respectively. The high-risk patients identified with the 6*6 model had a significantly worse outcome than the corresponding low-risk patients (P\u0026thinsp;=\u0026thinsp;0.005), whereas the differentiation produced by the 4*4, 5*5 and 9*9 models did not result in significant differences in outcome between the two risk group (P\u0026thinsp;=\u0026thinsp;0.096, P\u0026thinsp;=\u0026thinsp;0.056 and P\u0026thinsp;=\u0026thinsp;0.894, respectively). The KM curves of RCC patients at different risk stratifications are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. Moreover, univariable and multivariable logistic regression revealed that the results indicated that the 6*6 model and WHO/ISUP grade were independent factors influencing postoperative recurrence in RCC patients (P\u0026thinsp;=\u0026thinsp;0.002 and P\u0026thinsp;=\u0026thinsp;0.001, respectively); the detailed results are shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e The value of each parameters in predicting postoperative recurrence of renal cell carcinoma with univariate and multivariate logistic regression\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eParameters\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eUnivariate\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eMultivariate\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOR+(95% CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ep\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eOR+(95% CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ep\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e6*6 model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e12.60(3.25\u0026ndash;48.92)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e15.21(2.71\u0026ndash;85.31)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.002\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWHO/ISUP Grade\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16.31(4.06\u0026ndash;65.53)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e19.52(3.47-109.81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.001\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTumor diameter\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.73(1.28\u0026ndash;2.34)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.02(0.97\u0026ndash;1.07)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.408\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eADC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.00(0.99-1.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.044\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.00(0.99-1.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.842\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e\u003cp\u003eNote: OR\u0026thinsp;=\u0026thinsp;Odds Ratios; 95% CI\u0026thinsp;=\u0026thinsp;95% confidence intervals; ADC\u0026thinsp;=\u0026thinsp;apparent diffusion coefficient.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study presents evidence supporting the potential use of DR-CSI in evaluating the clinical outcomes of RCC patients. The results demonstrate the feasibility and adaptability of the equipartition method (6*6) combined with KW feature selection and GP classifier for leveraging the spectral information obtained with DR-CSI. Further analysis suggested that models based on DR-CSI data outperformed clinicopathological and conventional MRI parameters in distinguishing patients with different risks of recurrence, and its output may serve as an independent predictive factor of recurrence in RCC.\u003c/p\u003e\u003cp\u003eOur findings suggest that DR-CSI-based models could have excellent performance in predicting recurrence in RCC patients and may serve as a valuable tool for stratifying patients on the basis of their clinical outcomes. Additionally, it could play a fundamental role in identifying patients who can benefit from postoperative adjuvant targeted therapy and immunotherapy while avoiding unnecessary treatment for patients who are unlikely to experience recurrence, helping reduce financial burdens, minimize potential adverse effects, and prevent overestimation of the efficacy of adjuvant therapy(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe effectiveness of DR-CSI largely stems from its reliance on pathological features based on prior knowledge(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). However, we propose that DR-CSI can capture valuable information beyond the traditional pathology assessments of RCC. Traditional pathology assessments, such as tumor grading, provide insights into only certain aspects of tumor histology and are unable to fully integrate general histopathological and pathophysiological features. In contrast, imaging modalities may offer advantages in specific scenarios like component like epithelium, stroma, et al. On the basis of the results of our study, DR-CSI extends the scope of prognostic prediction by providing additional insights into the outcomes of RCC patients. Futhermore, its underlying pathological significance warrants further investigation in future research.\u003c/p\u003e\u003cp\u003eMoreover, DR-CSI demonstrated superior performance in predicting recurrence with respect to the traditional ADC and T2 values. We believe that this is because the parameters derived from the monoexponential model primarily reflect signal decay at specific TE or b values, whereas in reality, the relaxation and diffusion processes occur within a more complex multidimensional space. As a result of this complexity, the predictive ability of traditional parameters is worse than that of DR-CSI. Additionally, clinical, anatomical, and geometric factors have been reported to play a role in prognosis prediction(\u003cspan additionalcitationids=\"CR26\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). In this study, the DR-CSI-based model also outperformed most clinical, anatomical, and geometric factors in predicting recurrence, further demonstrating the advantages of this imaging modality in clinical application. Importantly, however, our study applied specific inclusion criteria that may have influenced the predictive performance of these features. As a result, further validation in diverse cohorts with different clinical and anatomical characteristics is needed to confirm these findings.\u003c/p\u003e\u003cp\u003eThe application of DR-CSI spectral information remains an important topic of discussion. Traditionally, individual peaks identified within the DR-CSI spectrum have been linked to tissue microstructure to improve their clinical utility. However, this approach presents significant challenges in clinical practice: obtaining high-quality spectra is not always feasible, distinguishing closely positioned or overlapping peaks is difficult, and in many cases, pathological or physiological evidence to clarify the significance of specific peaks is lacking.\u003c/p\u003e\u003cp\u003eTo tackle these challenges, multiple strategies have been proposed to partition DR-CSI spectra for clinical use (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). Yet existing methods still yield limited spectral information and lack a validated basis for defining regional boundaries. Subsequently, Luo et al. further explored an equalization method that divides the spectrum into multiple, equal subregions(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). However, it was unclear whether this approach could be universally applied, and the optimal choice of feature selection and classifier in modeling were not fully explored. In this study, we identified that the pipeline integrating KW subregion selection with a GP classifier was a potentially optimal strategy. Compared with other feature selection methods and classifiers, this combination has several notable advantages. Specifically, unlike parametric approaches, the KW-GP pipeline is well suited for capturing nonlinear relationships, thereby improving classification accuracy. The KW test, which is based on data ranks rather than absolute values, is robust to outliers, allowing more reliable identification of discriminative features and minimizing the impact of extrema. Furthermore, the KW method allows direct comparisons among three or more groups simultaneously, leading to highly efficient feature selection in multiclass radiomic analyses(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). Importantly, GP classifiers possess inherent advantages in small-sample learning scenarios, maintaining generalizability while mitigating the risk of overfitting(\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). Collectively, these characteristics make the KW-GP pipeline particularly advantageous for the data characteristic and analytical requirements of this study.\u003c/p\u003e\u003cp\u003eHowever, this study had several limitations. First, given the single-center nature of the study, our findings require validation through larger-scale, independent, prospective multicenter studies to assess their generalizability. Second, given the novelty of this technology, the follow-up period in our cohort was relatively limited. Investigating the value of DR-CSI in the long term could be particularly insightful. Third, while the DR-CSI parameters demonstrated significant predictive value, incorporating additional feature parameters into the model should be considered in future research.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, this study highlights the potential of DR-CSI as a novel imaging technique for assessing the aggressiveness of RCC and predicting patient outcomes. Our findings demonstrate that DR-CSI outperforms conventional clinicopathological and MRI parameters in distinguishing between patients with high and low risks of disease recurrence.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eRCC Renal cell carcinoma\u003c/p\u003e\u003cp\u003eTNM Tumor-node-metastasis\u003c/p\u003e\u003cp\u003eMRI Magnetic resonance imaging\u003c/p\u003e\u003cp\u003eROI Region of interest\u003c/p\u003e\u003cp\u003eDR-CSI Diffusion‒relaxation correlation spectroscopic imaging\u003c/p\u003e\u003cp\u003ePSM Propensity score matching\u003c/p\u003e\u003cp\u003eH\u0026amp;E Hematoxylin and eosin\u003c/p\u003e\u003cp\u003eICC Intraclass correlation coefficient\u003c/p\u003e\u003cp\u003ePCC Pearson correlation coefficient\u003c/p\u003e\u003cp\u003eKW Kruskal‒Wallis\u003c/p\u003e\u003cp\u003eANOVA Analysis of variance\u003c/p\u003e\u003cp\u003eRFE Recursive feature elimination\u003c/p\u003e\u003cp\u003eGP Gaussian process\u003c/p\u003e\u003cp\u003eSVM Support vector machine\u003c/p\u003e\u003cp\u003eLDA Linear discriminant analysis\u003c/p\u003e\u003cp\u003eLR Lasso Logistic regression via Lasso\u003c/p\u003e\u003cp\u003eLOOCV Leave-one-out cross-validation\u003c/p\u003e\u003cp\u003eAUC The area under the curve\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eEthics approval and consent to participate\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was performed in line with the principles of the Declaration of Helsinki. The Institutional Review Board of the Ethics Committee of Renji Hospital School of Medicine, Shanghai Jiao Tong University (institutional ethics approval number: LY2025-058-B) approved the research protocol for this retrospective study and waived the requirement for written informed consent from the patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eConsent for publication\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAvailability of data and materials\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data of this article can be obtained with the consent of the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eCompeting interests\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFunding\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by National Natural Science Foundation of China (NSFC 82371912), Shanghai Science and Technology Committee of Shanghai Municipality (grant no. 23Y11903900), and Transverse Project from Renji Hospital, School of Medicine, Shanghai Jiao Tong University (grant no. RJKY22-004).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAuthors\u0026apos; contributions\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMZ and YL : Data Curation, Formal analysis, Validation, Writing - Original Draft\u003c/p\u003e\n\u003cp\u003eGW and GL : Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Supervision, Writing - Review \u0026amp; Editing\u003c/p\u003e\n\u003cp\u003eXW, JC and SP : Resources\u003c/p\u003e\n\u003cp\u003eAll authors reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAcknowledgements\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHirsch MS, Signoretti S, Dal Cin P. Adult Ren Cell Carcinoma Surg Pathol Clin. 2015;8(4):587\u0026ndash;621.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZnaor A, Lortet-Tieulent J, Laversanne M, Jemal A, Bray F. International Variations and Trends in Renal Cell Carcinoma Incidence and Mortality. Eur Urol. 2015;67(3):519\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTeloken PE, Thompson RH, Tickoo SK, et al. Prognostic Impact of Histological Subtype on Surgically Treated Localized Renal Cell Carcinoma. J Urol. 2009;182(5):2132\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLeibovich BC, Lohse CM, Crispen PL, et al. Histological Subtype is an Independent Predictor of Outcome for Patients With Renal Cell Carcinoma. J Urol. 2010;183(4):1309\u0026ndash;16.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRavaud A, Motzer RJ, Pandha HS, et al. Adjuvant Sunitinib in High-Risk Renal-Cell Carcinoma after Nephrectomy. N Engl J Med. 2016;375(23):2246\u0026ndash;54.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYoshida T, Ohe C, Tsuzuki T, et al. Clinical impact of segmental renal vein invasion on recurrence in patients with clinical T1 renal cell carcinoma undergoing partial nephrectomy. Int J Clin Oncol. 2019;25(3):464\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShah PH, Moreira DM, Patel VR, et al. Partial Nephrectomy is Associated with Higher Risk of Relapse Compared with Radical Nephrectomy for Clinical Stage T1 Renal Cell Carcinoma Pathologically Up Staged to T3a. J Urol. 2017;198(2):289\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCheville JC, Lohse CM, Weaver AL, et al. Sarcomatoid renal cell carcinoma: an examination of underlying histologic subtype and an analysis of associations with patient outcome. Am J Surg Pathol. 2004;28(4):435\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKlatte T, Said JW, de Martino M, et al. Presence of Tumor Necrosis is Not a Significant Predictor of Survival in Clear Cell Renal Cell Carcinoma: Higher Prognostic Accuracy of Extent Based Rather Than Presence/Absence Classification. J Urol. 2009;181(4):1558\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOh J, Lee JM, Park J et al. Hepatocellular Carcinoma: Texture Analysis of Preoperative Computed Tomography Images Can Provide Markers of Tumor Grade and Disease-Free Survival. Korean J Radiol. 2019; 20(4).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou H, Valli\u0026egrave;res M, Bai HX, et al. MRI features predict survival and molecular markers in diffuse lower-grade gliomas. Neurooncology. 2017;19(6):862\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim D, Doyle EK, Wisnowski JL, Kim JH, Haldar JP. Diffusion-relaxation correlation spectroscopic imaging: A multidimensional approach for probing microstructure. Magn Reson Med. 2017;78(6):2236\u0026ndash;49.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang Z, Wu HH, Priester A, et al. Prostate Microstructure in Prostate Cancer Using 3-T MRI with Diffusion-Relaxation Correlation Spectrum Imaging: Validation with Whole-Mount Digital Histopathology. Radiology. 2020;296(2):348\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBenjamini D, Priemer DS, Perl DP, Brody DL, Basser PJ. Mapping astrogliosis in the individual human brain using multidimensional MRI. Brain. 2023;146(3):1212\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWei X, Zhu L, Zeng Y et al. Detection of prostate cancer using diffusion-relaxation correlation spectrum imaging with support vector machine model \u0026ndash; a feasibility study. Cancer Imaging. 2022; 22(1).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu F, Hu W, Sun Y, et al. Exploration of Interstitial Fibrosis in Chronic Kidney Disease by Diffusion-Relaxation Correlation Spectrum MR Imaging: A Preliminary Study. J Magn Reson Imaging. 2022;58(2):415\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDai Y, Hu W, Wu G, et al. Grading Clear Cell Renal Cell Carcinoma Grade Using Diffusion Relaxation Correlated MR Spectroscopic Imaging. J Magn Reson Imaging. 2023;59(2):699\u0026ndash;710.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDai Y, Zhu M, Hu W, et al. To characterize small renal cell carcinoma using diffusion relaxation correlation spectroscopic imaging and apparent diffusion coefficient based histogram analysis: a preliminary study. Radiol Med. 2024;129(6):834\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eElkassem AA, Allen BC, Sharbidre KG, Rais-Bahrami S, Smith AD. Update on the Role of Imaging in Clinical Staging and Restaging of Renal Cell Carcinoma Based on the AJCC 8th Edition, From the AJR Special Series on Cancer Staging. Am J Roentgenol. 2021;217(3):541\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXiao Q, Yi X, Guan X, et al. Validation of the World Health Organization/International Society of Urological Pathology grading for Chinese patients with clear cell renal cell carcinoma. Translational Androl Urol. 2020;9(6):2665\u0026ndash;74.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoch H. WHO-ISUP-Graduierungssystem f\u0026uuml;r Nierenkarzinome. Pathologe. 2016;37(4):355\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYushkevich PA, Gao Y, Gerig G. ITK-SNAP: An interactive tool for semi-automatic segmentation of multi-modality biomedical images. 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)2016; 3342-5.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDimitriadis SI, Song Y, Zhang J et al. FeAture Explorer (FAE): A tool for developing and comparing radiomics models. PLoS ONE. 2020; 15(8).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUrbanowicz RJ, Meeker M, La Cava W, Olson RS, Moore JH. Relief-based feature selection: Introduction and review. J Biomed Inform. 2018;85:189\u0026ndash;203.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFrank I, Blute M, Cheville J, Lohse C, Weaver A, Zincke H. An Outcome Prediction Model for Patients with Clear Cell Renal Cell Carcinoma Treated with Radical Nephrectomy Based on Tumor Stage, Size, Grade and Necrosis: The Ssign Score. J Urol. 2002;168(6):2395\u0026ndash;400.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLeibovich BC, Blute ML, Cheville JC, et al. Prediction of progression after radical nephrectomy for patients with clear cell renal cell carcinoma. Cancer. 2003;97(7):1663\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZisman A, Pantuck AJ, Dorey F, et al. Improved prognostication of renal cell carcinoma using an integrated staging system. J Clin Oncol. 2001;19(6):1649\u0026ndash;57.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuo Y, Zhu M, Wei X, et al. Investigation of clear cell renal cell carcinoma grades using diffusion-relaxation correlation spectroscopic imaging with optimized spatial-spectrum analysis. Br J Radiol. 2024;97(1153):135\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaeys Y, Inza I, Larra\u0026ntilde;aga P. A review of feature selection techniques in bioinformatics. Bioinformatics. 2007;23(19):2507\u0026ndash;17.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eManiruzzaman, Rahman MJ, Benojir A, et al. Statistical characterization and classification of colon microarray gene expression data using multiple machine learning paradigms. Comput Methods Programs Biomed. 2019;176:173\u0026ndash;93.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Renal Cell Carcinoma, Diffusion‒Relaxation Correlation Spectroscopic Imaging, Prognosis, Prediction Model","lastPublishedDoi":"10.21203/rs.3.rs-7287776/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7287776/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eThe prognosis of renal cell carcinoma (RCC) varies greatly, and a highly efficient prognostic strategy is crucial for treatment selection. This study aim to evaluate the feasibility of predictive models based on diffusion relaxation correlation spectroscopic imaging (DR-CSI) in distinguishing RCC patients with different clinical outcomes.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eOne hundred and twenty-seven RCC patients who underwent DR-CSI were enrolled, including 48 patients as cohort 1 for development and 79 postoperative follow-up patients as cohort 2 for validation. DR-CSI results were analyzed using spectral equipartition method combined with various feature selection methods and classifiers, from 2*2 to 9*9. The Kruskal‒Wallis (KW), analysis of variance (ANOVA), recursive feature elimination (RFE), and Relief methods were used for subregion selection. The classifiers including Gaussian process (GP), support vector machine (SVM), linear discriminant analysis (LDA), and logistic regression via Lasso (LR Lasso). Clinicopathological and conventional MR parameters were obtained. Diagnostic performance was evaluated using AUC and compared with DeLong\u0026rsquo;s test. Kaplan‒Meier method and multivariable analysis were used for evaluating the performance of prediction model.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eDR-CSI-based equipartition models demonstrated excellent interobserver agreement (ICC: 0.86\u0026ndash;0.99). The equipartition method (6*6) combined with KW feature selection and GP classifier achieved the highest diagnostic performance in distinguishing patients with metastatic RCC from those without metastatic RCC, with an AUC of 0.87, significantly outperforming clinicopathological and conventional MR parameters (vs. age, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; vs. tumor diameter, P\u0026thinsp;=\u0026thinsp;0.002; vs. WHO/ISUP grade, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; vs. ADC, P\u0026thinsp;=\u0026thinsp;0.001; vs. T2 value, P\u0026thinsp;=\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The 6*6 model could effectively predict the recurrence of patients in cohort 2 (P\u0026thinsp;=\u0026thinsp;0.005), whereas the other models could not. Additionally, the 6*6 model might serve as an independent predictive factor for recurrence in RCC.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eDR-CSI-based 6*6 model combined with KW and GP may assess the aggressiveness of RCC and have great promise in predicting prognostic risk stratification.\u003c/p\u003e","manuscriptTitle":"Exploration of Prognostic Prediction Models for Renal Cell Carcinoma using Diffusion Relaxation Correlation Spectroscopic Imaging","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-22 11:05:55","doi":"10.21203/rs.3.rs-7287776/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bd6a9137-02ef-4300-9039-0eaeba7779e4","owner":[],"postedDate":"August 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-10-09T06:09:31+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-22 11:05:55","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7287776","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7287776","identity":"rs-7287776","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}