Predicting variant histology in bladder cancer: the role of multiparametric MRI and vesical imaging-reporting and data system (VI-RADS)

Abstract

Objectives

(1) To evaluate the diagnostic performance of the VI-RADS score in detecting muscle invasion in variant urothelial carcinomas (VUC). (2) To identify spesific MRI features that may serve as predicting for VUC.

Methods

Two hundred four patients who underwent TUR-B and/or radical cystectomy and a bladder mpMRI scan within three months prior to the procedure were retrospectively enrolled. The tumors were divided into two histological cohorts: pure urothelial carcinoma (PUC) and VUC. Various MRI features, including largest tumor diameter, long-to-short axis ratio, morphology, heterogeneous signal intensity (SI), presence of necrosis, and normalized ADC (ADC_n) value, were analyzed. The diagnostic performance of the VI-RADS score in predicting muscle invasion was calculated using a cut-off point of ≥ 4 in both cohorts. Univariate logistic regression were also performed to identify MRI features that predict VUC. Inter-reader agreement was assessed with the weighted kappa coefficient.

Results

Our study identified several MRI features significantly associated with VUC, including heterogeneous SI on T2-weighted images (OR: 3.055; 95% CI: 1.312–7.112; p < 0.001), dynamic contrast enhancement images (OR: 2.935; 95% CI: 1.263–6.821; p < 0.001), and the presence of necrosis (OR: 3.575; 95% CI: 1.798–7.107; p < 0.001). Additionally, ADC_n values were significantly lower in the VUC cohort (p = 0.003). The VI-RADS score demonstrated high diagnostic performance across both VUC and PUC cohorts, with sensitivity ranging from 94.4 to 86.8% (reader 1) and 94.2–82.2% (reader 2), specificity ranging from 100 to 94.6% (reader 1) and 100–94% (reader 2), and accuracy ranging from 96 to 90.6% (reader 1) and 96–88.2% (reader 2). The area under the curve (AUC) ranged between 0.972 and 0.972 (reader 1) and 0.838–0.781 (reader 2). No significant differences in diagnostic performance were observed between readers or cohorts (p > 0.05), and inter-reader agreement for VI-RADS scores was excellent for both cohorts.

Conclusion

VI-RADS score can be used with high performance to detect muscle invasion in VUC, regardless of reader experience. Additionally, specific MRI features such as heterogeneous SI, the presence of necrosis, and ADC_n values can serve as potential predictors of VUC.

Graphical Abstract