Development and validation of a CT-based deep learning radiomics nomogram to predict muscle invasion in bladder cancer.

Objective: This study aimed to develop a nomogram combining CT-based handcrafted radiomics and deep learning (DL) features to preoperatively predict muscle invasion in bladder cancer (BCa) with multi-center validation. Methods: In this retrospective study, 323 patients underwent radical cystectomy with pathologically confirmed BCa were enrolled and randomly divided into the training cohort (n = 226) and internal validation cohort (n = 97). And fifty-two patients from another independent medical center were enrolled as an independent external validation cohort. Handcrafted radiomics and DL features were constructed from preoperative nephrographic phase CT images. Least absolute shrinkage and selection operator (LASSO) regression was used to identify the most discriminative features in train cohort. Multivariate logistic regression was used to develop the predictive model and a deep learning radiomics nomogram (DLRN) was constructed. The predictive performance of models was evaluated by area under the curves (AUC) in the three cohorts. The calibration and clinical usefulness of DLRN were estimated by calibration curve and decision curve analysis. Results: The nomogram that incorporated radiomics signature and DL signature demonstrated satisfactory predictive performance for differentiating non-muscle invasive bladder cancer (NMIBC) from muscle invasive bladder cancer (MIBC), with an AUC of 0.884 (95 % CI: 0.813-0.953) in internal validation cohort and 0.862 (95 % CI: 0.756-0.968) in external validation cohort, respectively. Decision curve analysis confirmed the clinical usefulness of the nomogram. Conclusions: A CT-based deep learning radiomics nomogram exhibited a promising performance for preoperative prediction of muscle invasion in bladder cancer, and may be helpful in the clinical decision-making process.

Validity of a multiphase CT-based radiomics model in predicting the Leibovich risk groups for localized clear cell renal cell carcinoma: an exploratory study.

OBJECTIVE: To develop and validate a multiphase CT-based radiomics model for preoperative risk stratification of patients with localized clear cell renal cell carcinoma (ccRCC). METHODS: A total of 425 patients with localized ccRCC were enrolled and divided into training, validation, and external testing cohorts. Radiomics features were extracted from three-phase CT images (unenhanced, arterial, and venous), and radiomics signatures were constructed by the least absolute shrinkage and selection operator (LASSO) regression algorithm. The radiomics score (Rad-score) for each patient was calculated. The radiomics model was established and visualized as a nomogram by incorporating significant clinical factors and Rad-score. The predictive performance of the radiomics model was evaluated by the receiver operating characteristic curve, calibration curve, and decision curve analysis (DCA). RESULTS: The AUC of the triphasic radiomics signature reached 0.862 (95% CI: 0.809-0.914), 0.853 (95% CI: 0.785-0.921), and 0.837 (95% CI: 0.714-0.959) in three cohorts, respectively, which were higher than arterial, venous, and unenhanced radiomics signatures. Multivariate logistic regression analysis showed that Rad-score (OR: 4.066, 95% CI: 3.495-8.790) and renal vein invasion (OR: 12.914, 95% CI: 1.118-149.112) were independent predictors and used to develop the radiomics model. The radiomics model showed good calibration and discrimination and yielded an AUC of 0.872 (95% CI: 0.821-0.923), 0.865 (95% CI: 0.800-0.930), and 0.848 (95% CI: 0.728-0.967) in three cohorts, respectively. DCA showed the clinical usefulness of the radiomics model in predicting the Leibovich risk groups. CONCLUSIONS: The radiomics model can be used as a non-invasive and useful tool to predict the Leibovich risk groups for localized ccRCC patients. CRITICAL RELEVANCE STATEMENT: The triphasic CT-based radiomics model achieved favorable performance in preoperatively predicting the Leibovich risk groups in patients with localized ccRCC. Therefore, it can be used as a non-invasive and effective tool for preoperative risk stratification of patients with localized ccRCC. KEY POINTS: • The triphasic CT-based radiomics signature achieves better performance than the single-phase radiomics signature. • Radiomics holds prospects in preoperatively predicting the Leibovich risk groups for ccRCC. • This study provides a non-invasive method to stratify patients with localized ccRCC.

Stratification of malignant renal neoplasms from cystic renal lesions using deep learning and radiomics features based on a stacking ensemble CT machine learning algorithm.

Using nephrographic phase CT images combined with pathology diagnosis, we aim to develop and validate a fusion feature-based stacking ensemble machine learning model to distinguish malignant renal neoplasms from cystic renal lesions (CRLs). This retrospective research includes 166 individuals with CRLs for model training and 47 individuals with CRLs in another institution for model testing. Histopathology results are adopted as diagnosis criterion. Nephrographic phase CT scans are selected to build the fusion feature-based machine learning algorithms. The pretrained 3D-ResNet50 CNN model and radiomics methods are selected to extract deep features and radiomics features, respectively. Fivefold cross-validated least absolute shrinkage and selection operator (LASSO) regression methods are adopted to identify the most discriminative candidate features in the development cohort. Intraclass correlation coefficients and interclass correlation coefficients are employed to evaluate feature's reproducibility. Pearson correlation coefficients for normal distribution features and Spearman's rank correlation coefficients for non-normal distribution features are used to eliminate redundant features. After that, stacking ensemble machine learning models are developed in the training cohort. The area under the receiver operator characteristic curve (ROC), calibration curve, and decision curve analysis (DCA) are adopted in the testing cohort to evaluate the performance of each model. The stacking ensemble machine learning algorithm reached excellent diagnostic performance in the testing dataset. The calibration plot shows good stability when using the stacking ensemble model. Net benefits presented by DCA are higher than the Bosniak 2019 version classification when employing any machine learning algorithm. The fusion feature-based machine learning algorithm accurately distinguishes malignant renal neoplasms from CRLs, which outperformed the Bosniak 2019 version classification, and proves to be more applicable for clinical decision-making.