Journal
ACADEMIC RADIOLOGY
Volume 30, Issue 1, Pages 64-76Publisher
ELSEVIER SCIENCE INC
DOI: 10.1016/j.acra.2022.04.014
Keywords
Bladder cancer; muscle-invasive status; multi-parametric magnetic resonance imaging; radiomics; vesical imaging-reporting and data system
Ask authors/readers for more resources
This study developed three predictive models using tri-parametric MRI sequences (T2-weighted, diffusion-weighted, and dynamic contrast-enhanced) to predict the invasive status of bladder cancer. The results showed that the tri-parametric model performed significantly better than the bi-parametric models and VI-RADS scoring system in discriminating invasive bladder cancer. In conclusion, tri-parametric MRI provides additional value to the radiomics-based identification of invasive bladder cancer.
Rationale and Objectives: Identification of muscle-invasive status (MIS) of bladder cancer (BCa) is critical for treatment decisions. The Vesical Imaging-Reporting and Data System (VI-RADS) has been widely used in preoperatively predicting MIS using tri-parametric MR imaging including T2-weighted (T2W), diffusion-weighted (DW), and dynamic contrast-enhanced (DCE) sequences. While the diagnostic values of radiomics features from bi-parametric MRI such as T2W + DW to identification of MIS have been reported, whether the tri-parametric MRI could provide additional diagnostic value to the radiomics prediction task, and how to integrate DCE features into the radiomics model, which is the objectives of this study, remain unknown. Materials and Methods: Patients with postoperatively confirmed BCa lesions (150 in non-muscle-invasive BCa and 56 in muscle-invasive BCa groups) were retrospectively included. Their T2W, DW with apparent diffusion coefficient (ADC) maps, and DCE sequences were acquired using a 3.0T MR system. Regions of interest were manually depicted and VI-RADS scores were assessed by three radiologists. Three predictive models were developed by the radiomics features extracted from sequence combinations of T2W + DW (Model one), T2W + DCE (Model two), and T2W + DW + DCE (Model three), respectively, using the least absolute shrinkage and selection operator. The performance of each model was quantitatively assessed on both the training (n = 165) and testing (n = 41) cohorts. Then a 10 times five-fold cross validation was conducted to assess the overall performance. Results: Three models achieved area under the curve of 0.888, 0.869, and 0.901 in the cross validation, respectively. The tri-parametric model performed significantly superior than the two bi-parametric models and VI-RADS. The analysis of feature coefficients derived from least absolute shrinkage and selection operator algorithm showed features from the tri-parametric MRI are effective in MIS discrimination. Conclusion: The tri-parametric MRI provides additional value to the radiomics-based identification of MIS.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available