What is the impact of dynamic contrast-enhancement sequence in the Vesical Imaging, Reporting and Data System (VI-RADS)? A subgroup analysis

The therapeutic approach for bladder cancer (BCa) is largely driven by the clinical staging at the time of diagnosis [1, 2]. Urothelial lesions are the dominant histologic type of BCa, representing about 90% of all malignant vesical lesions. Their local staging is quite variable, and a significant proportion of the cancers is limited to the mucosa [2]. Defining the status of the muscular layer is essential to determine the treatment. Non-muscular invasive bladder cancer (NMIBC) lesions are treated conservatively, and the bladder and micturition mechanism are preserved [3, 4]. However, for muscular invasive (MIBC) lesions, the usual approach is radical cystectomy or cystoprostatectomy, depending on the extension of the primary lesion [5].

The standard way to define the extension of vesical cancers is cystoscopy followed by biopsy using transurethral procedures. Transurethral resection of bladder tumors (TURBT) can be performed as the first treatment option for most NMIBCs but can also be carried out as a diagnostic procedure for MIBCs [5]. The goal is to obtain samples from the underlying bladder wall, including the mucosa, lamina, and muscularis propria. However, the accuracy of the approach of cystoscopy plus biopsy is suboptimal, and understaging is estimated to occur in about 25% of cases [6‐8].

The imaging approach for local staging of BCa had been restricted because both Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) show limited accuracy [9]. But recently, several studies have indicated very promising results for defining the status of muscular layer in BCa using multiparametric MRI, a combination of T2, diffusion-weighted imaging (DWI), and dynamic contrast-enhancement (DCE). The accuracy has ranged from 92 to 94% in recent studies using a multiparametric approach [10, 11].

Based on these promising results, in 2018, a group of experts developed a scoring system to assess the status of the muscular layer in BCa [12]. The Vesical Imaging Radiological and Data System (VI-RADS) was inspired by similar tools for the standardization imaging approaches, such as BI-RADS and PI-RADS [13, 14]. After its release in 2018, some studies have performed external validations with very encouraging results, and the overall accuracy has ranged from 0.79 to 0.94 [15, 16]. Also, the level of agreement for the categorization by each single sequence and for the final category of VI-RADS has ranged from 0.82 to 0.95, and there has been almost perfect agreement in the majority of the studies so far [17, 18].

Notwithstanding these promising results, some studies have questioned the use of contrast-enhancement sequences, arguing that biparametric MRI would lead to similar results to the multiparametric approach [19, 20]. Therefore, we carried out this retrospective, single-institution study with two purposes. First, to assess if a “biparametric” VI-RADS would have similar results to the original for distinguishing NMIBC from MIBC, and, second, to perform a subgroup analysis according to tumor burden.

Of the 44 patients, 25 were female (56.8%) and 19 were male (43.2%). The mean age was 68.7 +/− 10.8 years, ranging from 35 to 91 years. The pathologic staging was distributed as follows: 25 cases (50.0%) were T1 (Fig. 2); 8 cases (16.0%) were T2; 8 cases (16.0%) were T3; and 9 cases (18.0%) were T4. Accordingly, 25 lesions (50%) were NMIBCa, while the other half had muscular involvement (Fig. 3). The demographic data for both groups are presented in Table 1. The standard of reference was surgery and pathologic specimens in 26 patients (59.1%) and histopathological analysis from TURB in 18 patients (40.9%). The final diagnosis was high-grade urothelial cancers for 43 lesions (86.0%), low-grade urothelial cancer for 5 lesions (10.0%), and other histologic subtypes for 2 lesions (4.0%).

The MRI findings are presented in Table 2. There was no significant difference between the two groups in the presence of a stalk for both readers (p = 0.16 for both reader). The form of the lesion was significantly different for both readers. For both readers, in the NMIBCa group, 23 lesions (92.0%) were papillary and only 2 (8.0%) were flat. In the MIBCa group, for reader 1, 16 (64.0%) were papillary and 9 were flat (36.0%) (p = 0.01), and for reader 2, 17 (68.0%) were papillary and 8 were flat (32.0%), and the difference was also significant (p = 0.03). For reader 1, the mean size in NMIBCa was 2.42 +/− 1.58 cm, and for the MIBCa group, it was 5.70 +/− 2.67 cm (p < 0.0001). For reader 2, the size was 2.37 +/− 1.50 cm vs. 5.44 +/− 2.90 cm, respectively (p < 0.0001).

The Table 3 shows the distribution of mpVI-RADS score according to pathological status, and the risk of muscle invasion for VI-RADS 2,3,4 and 5 was 7.1, 0, 55.5 and 86.4%, respectively for reader 1 and 6.3, 25.0, 63.6 and 94.1% for reader 2.

Table 4 shows the diagnostic accuracy of the final VI-RADS classification for both readers, for all lesions and the small lesions only. A trend was observed towards a better performance of the more experienced reader for final mpVI-RADS. The area under the curve (AUC) for muscle invasion with mpVI-RADS, considering all lesions, was similar for both readers, for reader 1 was 0.885 +/− 0.04 (95% CI-0.79-0.98) for reader 1 and 0.924 +/− 0.04 (0.84–0.99) for reader 2, and for bpVI-RADS was 0.879+/− 0.05 and 0.916 +/− 0.04 (0.85–0.99), respectively, both differences not statistically significant (p = 0.24 and 0.07, respectively). When subgroup analysis was performed (Table 4), considering only lesions up to 3.0 cm, a drop in the overall accuracy of mpVI-RADS was observed for both readers, the accuracy for mpVI-RADS was 0.795 +/− 0.11 (0.57–1.0) for reader1, and 0.80 +/− 0.11(0.57–1.0) for reader 2, a non-significant difference (p = 0.56). However, for bpVI-RADS the AUC was 0.747 +/− 0.12 (0.50–0.99) for reader 1 and 0.80 +/− 0.11(0.57–1.0) for reader 2, a significant difference (p = 0.04.

The inter-reader agreement for the presence of a stalk in the lesions was 0.77 (95%CI: 0.51–1.0), while for the lesions’ form was 0.94 (0.68–1.0), and the for the final mpVI-RADS score was 0.81 (0.60–1.0). For the final score using the biparametric approach the kappa value was 0.85 (0.63–1.0). When assessing agreement in each sequence, the best result was obtained for DCE with an ICC of 0.85 (0.59–1.0), indicating excellent agreement. These values and those of ICC for the small lesions group are showed in Table 5.

Our results suggest that the use of intravenous contrast media only improves the diagnostic accuracy of bladder cancer staging, using the VI-RADS system score, when less experienced readers assess small lesions.

In our study, we assumed that a VI-RADS score of only 4 indicates muscle-invasive disease. There were two reasons for this. First, the statistical analysis pointed to this threshold as the optimal point for better accuracy when determining MIBCa. Second, the decision to perform bladder resection should be based on a more specific approach.

Interestingly, despite the low number of cases in each category in our cohort, the false-negative rates were only 4.0 and 8.0% (cases classified as VI-RADS 1, 2 and 3) for the less and more experienced readers, respectively. However, the rates of false-positive cases were relatively high 28.0 and 20.0% for readers 1 and 2, respectively. Our results are aligned with previous studies [15, 17, 22], but slightly better performance was obtained for false-negative cases, although slightly worse results were obtained for false-positive cases [17, 23].

When comparing the accuracy of readers, the results of the less experienced one were consistently inferior to those of the more experienced reader throughout the different sequences and for the final score, but the difference was significant only when assessing small lesions with biparametric approach. One possible explanation for these results is that our study may be underpowered for showing significant difference in other scenarios. Or, biparametric approach may have similar accuracy for experienced readers, which would be not surprising as, in many situations, the diagnostic accuracy of an imaging method is clearly dependent on the experience of the reader [23]. Highly-specialized exams (for instance, prostate MRI) have a known learning curve for interpretation, and some organizations have suggested minimum requirements for independent reporting. However, this is an open issue for bladder cancer imaging [24].

We decided to perform a subgroup analysis assessing the diagnostic accuracy for small bladder lesions (≤ 3.0 cm) because the staging tends to be more complex in these cases. The difference in diagnostic accuracy of this limited sample was not significant, but compared to the whole group, there was a clear trend of lower accuracy for both reader 1 (84.0 vs. 79.2) and reader 2 (86.0 vs. 79.2) for the final mpVI-RADS score and for the bpVI-RADS, 79.6 vs. 75.0 (reader 1) and 85.7 vs. 79.2 (reader 2). This trend was even more conspicuous, although without statistical significance for the less experienced reader, which may indicate, depending on further studies that multiparametric approach should be preserved when radiologists with limited experience were scanning and interpreting MR studies, mainly if small lesions were detected on anatomical sequences. Once the situations where the use of contrast-enhanced sequences are clear, a more rational approach (including these two protocols) for bladder cancer staging would be feasible.

In the pre-VI-RADS years, some relevant studies have placed the contrast-enhanced sequences as essential for the best results on vesical cancer staging [10, 11]. However, recently, the role of contrast sequences has been questioned. Delle Pizzi et al. [19] in a multi-reader, a prospective study found similar accuracies for MRI protocol with and without contrast media, regardless of readers’ experience. Gmeiner et al. [20] pointed out the same direction, although in a retrospective study and using a different protocol than suggested in VI-RADS [12]. Our study also indicates the same direction. Possible explanations for this change in the perception of the value of contrast sequences include the improvement in MRI techniques, including the non-Gradient Echo DWI sequences, the motion-corrected high-resolution T2-weighted images, and the ascending learning curve of all radiologists as Bladder cancer staging is a much more requested exam now than a decade ago.

The VI-RADS scoring system was designed with a focus on the standardization of MR imaging for staging BCa. Based on a multisequence approach, the system also defines an algorithm to conduct cases using findings from different image acquisitions in a hierarchical approach [12]. By using findings from different sequences, one would expect a quite variable interpretation, but the literature has shown excellent reproducibility for the final assessment of the muscle layer by VI-RADS [17, 25, 26] ranging from 0.73 to 0.92. Our data are aligned with the literature in showing substantial agreement for final classification with an ICC of 0.81 (0.60–1.00), p < 0.0001, for all lesions and 0.82 (0.71–0.95), p < 0.0001 for lesions under 3.0 cm. On the other side, in our study, in contrast to others in literature [17, 25, 27], DCE showed almost perfect agreement, slightly higher than those observed for T2, DWI and the final VI-RADS score. This represents an interesting point, as DCE and DWI are the sequences that define the final scores when discrepant findings are present, according to the algorithm proposed by VI-RADS. And, based on it, we could argue that DCE can be preserved in specific situations (small lesions/less experienced readers) as a problem-solving sequence with a high level of reproducibility.

Our study does have some limitations that should be acknowledged. First, it is a retrospective study with all inherent risks of bias, although we carefully applied inclusion and exclusion criteria to minimize them. Second, we had a small cohort, which may limit the analysis of subgroups, such as some of the five categories of VI-RADS. The reason for the small number of patients in our sample was the high number of patients who had cystoscopy/TURB prior to MRI in the first year of our study, which reflects that urologist were not so familiar with VI-RADS at that time. Third, we performed all examinations with a 1.5-T scanner. However, the exclusive use of 3.0-T scanners has not been recommended by VI-RADS, and a recent meta-analysis showed comparable results between studies performed using 1.5 T and 3.0 T [28].

Our results indicate good overall accuracy for distinguishing NMIBCa from MIBCa using the biparametric approach, similar to the mpVI-RADS original score. However, a trend towards a lower overall accuracy with a biparametric approach, more evident for small lesions and for the less experienced reader, may indicate that DCE sequences should be preserved in these situations. Further, larger, prospective, and multi-institutional studies are required for defining the exact role of contrast sequences in Bladder cancer staging.