A preoperative CT-based deep learning radiomics model in predicting the stage, size, grade and necrosis score and outcome in localized clear cell renal cell carcinoma: A multicenter study.

BACKGROUND AND PURPOSE: The Stage, Size, Grade and Necrosis (SSIGN) score is the most commonly used prognostic model in clear cell renal cell carcinoma (ccRCC) patients. It is a great challenge to preoperatively predict SSIGN score and outcome of ccRCC patients. The aim of this study was to develop and validate a CT-based deep learning radiomics model (DLRM) for predicting SSIGN score and outcome in localized ccRCC. METHODS: A multicenter 784 (training cohort/ test 1 cohort / test 2 cohort, 475/204/105) localized ccRCC patients were enrolled. Radiomics signature (RS), deep learning signature (DLS), and DLRM incorporating radiomics and deep learning features were developed for predicting SSIGN score. Model performance was evaluated with area under the receiver operating characteristic curve (AUC). Kaplan-Meier survival analysis was used to assess the association of the model-predicted SSIGN with cancer-specific survival (CSS). Harrell's concordance index (C-index) was calculated to assess the CSS predictive accuracy of these models. RESULTS: The DLRM achieved higher micro-average/macro-average AUCs (0.913/0.850, and 0.969/0.942, respectively in test 1 cohort and test 2 cohort) than the RS and DLS did for the prediction of SSIGN score. The CSS showed significant differences among the DLRM-predicted risk groups. The DLRM achieved higher C-indices (0.827 and 0.824, respectively in test 1 cohort and test 2 cohort) than the RS and DLS did in predicting CSS for localized ccRCC patients. CONCLUSION: The DLRM can accurately predict the SSIGN score and outcome in localized ccRCC.

A CT-based deep learning radiomics nomogram outperforms the existing prognostic models for outcome prediction in clear cell renal cell carcinoma: a multicenter study.

OBJECTIVES: To develop and validate a CT-based deep learning radiomics nomogram (DLRN) for outcome prediction in clear cell renal cell carcinoma (ccRCC), and its performance was compared with the Stage, Size, Grade, and Necrosis (SSIGN) score, the University of California, Los Angeles, Integrated Staging System (UISS), the Memorial Sloan-Kettering Cancer Center (MSKCC), and the International Metastatic Renal Cell Database Consortium (IMDC). METHODS: A multicenter of 799 localized (training/ test cohort, 558/241) and 45 metastatic ccRCC patients were studied. A DLRN was developed for predicting recurrence-free survival (RFS) in localized ccRCC patients, and another DLRN was developed for predicting overall survival (OS) in metastatic ccRCC patients. The performance of the two DLRNs was compared with that of the SSIGN, UISS, MSKCC, and IMDC. Model performance was assessed with Kaplan-Meier curves, time-dependent area under the curve (time-AUC), Harrell's concordance index (C-index), and decision curve analysis (DCA). RESULTS: In the test cohort, the DLRN achieved higher time-AUCs (0.921, 0.911, and 0.900 for 1, 3, and 5 years, respectively), C-index (0.883), and net benefit than SSIGN and UISS in predicting RFS for localized ccRCC patients. The DLRN provided higher time-AUCs (0.594, 0.649, and 0.754 for 1, 3, and 5 years, respectively) than MSKCC and IMDC in predicting OS for metastatic ccRCC patients. CONCLUSIONS: The DLRN can accurately predict outcomes and outperformed the existing prognostic models in ccRCC patients. CLINICAL RELEVANCE STATEMENT: This deep learning radiomics nomogram may facilitate individualized treatment, surveillance, and adjuvant trial design for patients with clear cell renal cell carcinoma. KEY POINTS: • SSIGN, UISS, MSKCC, and IMDC may be insufficient for outcome prediction in ccRCC patients. • Radiomics and deep learning allow for the characterization of tumor heterogeneity. • The CT-based deep learning radiomics nomogram outperforms the existing prognostic models in ccRCC outcome prediction.

The radiomics-based tumor heterogeneity adds incremental value to the existing prognostic models for predicting outcome in localized clear cell renal cell carcinoma: a multicenter study.

PURPOSE: Tumor heterogeneity, which is associated with poor outcomes, has not been exhibited in the University of California, Los Angeles, Integrated Staging System (UISS), and the Stage, Size, Grade and Necrosis (SSIGN) scores. Radiomics allows an in-depth characterization of heterogeneity across the tumor, but its incremental value to the existing prognostic models for clear cell renal cell carcinoma (ccRCC) outcome is unknown. The purpose of this study was to evaluate the association between the radiomics-based tumor heterogeneity and postoperative risk of recurrence in localized ccRCC, and to assess its incremental value to UISS and SSIGN. METHODS: A multicenter 866 ccRCC patients derived from 12 Chinese hospitals were studied. The endpoint was recurrence-free survival (RFS). A CT-based radiomics signature (RS) was developed and assessed in the whole cohort and in the subgroups stratified by UISS and SSIGN. Two combined nomograms, the R-UISS (combining RS and UISS) and R-SSIGN (combining RS and SSIGN), were developed. The incremental value of RS to UISS and SSIGN in RFS prediction was evaluated. R statistical software was used for statistics. RESULTS: Patients with low radiomics scores were 4.44 times more likely to experience recurrence than those with high radiomics scores (P<0.001). Stratified analysis suggested the association is significant among low- and intermediate-risk patients identified by UISS and SSIGN. The R-UISS and R-SSIGN showed better predictive capability than UISS and SSIGN did with higher C-indices (R-UISS vs. UISS, 0.74 vs. 0.64; R-SSIGN vs. SSIGN, 0.78 vs. 0.76) and higher clinical net benefit. CONCLUSIONS: The radiomics-based tumor heterogeneity can predict outcome and add incremental value to the existing prognostic models in localized ccRCC patients. Incorporating radiomics-based tumor heterogeneity in ccRCC prognostic models may provide the opportunity to better surveillance and adjuvant clinical trial design.

2D and 3D texture analysis to predict lymphovascular invasion in lung adenocarcinoma.

PURPOSE: Lymphovascular invasion (LVI) impairs surgical outcomes in lung adenocarcinoma (LAC) patients. Preoperative prediction of LVI is challenging by using traditional clinical and imaging factors. The purpose of this study was to evaluate the value of two-dimensional (2D) and three-dimensional (3D) CT texture analysis (CTTA) in predicting LVI in LAC. METHODS: A total of 149 LAC patients (50 LVI-present LACs and 99 LVI-absent LACs) were retrospectively enrolled. Clinical data and CT findings were analyzed to select independent clinical predictors. Texture features were extracted from 2D and 3D regions of interest (ROI) in 1.25 mm slice CT images. The 2D and 3D CTTA signatures were constructed with the least absolute shrinkage and selection operator algorithm and texture scores were calculated. The optimized CTTA signature was selected by comparing the predicting efficacy and clinical usefulness of 2D and 3D CTTA signatures. A CTTA nomogram was developed by integrating the optimized CTTA signature and clinical predictors, and its calibration, discrimination and clinical usefulness were evaluated. RESULTS: Maximum diametre and spiculation were independent clinical predictors. 1125 texture features were extracted from 2D and 3D ROIs and reduced to 11 features to build 2D and 3D CTTA signatures. There was significant difference (P < 0.001) in AUC (area under the curve) between 2D signature (AUC, 0.938) and 3D signature (AUC, 0.753) in the training set. There was no significant difference (P = 0.056) in AUC between 2D signature (AUC, 0.856) and 3D signature (AUC, 0.701) in the test set. Decision curve analysis showed the 2D signature outperformed the 3D signature in terms of clinical usefulness. The 2D CTTA nomogram (AUC, 0.938 and 0.861, in the training and test sets), which incorporated the 2D signature and clinical predictors, showed a similar discrimination capability (P = 1.000 and 0.430, in the training and test sets) and clinical usefulness as the 2D signature, and outperformed the clinical model (AUC, 0.678 and 0.776, in the training and test sets). CONCLUSIONS: 2D CTTA signature performs better than 3D CTTA signature. The 2D CTTA nomogram with the 2D signature and clinical predictors incorporated provides the similar performance as the 2D signature for individual LVI prediction in LAC.