Associations of Anatomical Location and Reader Experience With the Diagnostic Performance of Multiparametric MRI ‐Derived VI ‐ RADS for Muscle Invasive Bladder Cancer: A Multicenter Multireader Study

Abstract

ABSTRACT Background The associations of tumor location and reader experience with Vesical Imaging‐Reporting and Data System (VI‐RADS) performance remain unclear. Purpose To evaluate the associations of tumor location and reader experience with VI‐RADS performance in detecting muscle‐invasive bladder cancer (MIBC). Study Type Retrospective. Subjects A total of 543 patients (464 males; mean age, 67.0 ± 12.1 years) with bladder cancer. Field Strength/Sequence 1.5 T and 3.0 T, T2‐weighted image with turbo spin‐echo, diffusion‐weighted imaging using single‐shot echo planar imaging, and three‐dimensional T1‐weighted spoiled gradient echo sequence. Assessment Four radiologists (two with 15 and 10 years of experience and two with < 5 years of experience) independently assigned VI‐RADS scores. Histopathology was the reference standard. Diagnostic performance of VI‐RADS ≥ 3 for identifying MIBC was assessed across seven anatomical locations: trigone, ureteral orifice, neck, lateral wall, dome, anterior and posterior walls. Statistical Tests Weighted kappa, Generalized estimating equation (GEE) models, the independent samples t ‐test or Mann–Whitney U test, the chi‐square test, ROC analysis and DeLong test. Benjamini‐Hochberg false discovery rate (FDR)‐adjust p < 0.05 was considered statistically significant. Results Of 543 patients, 168 (30.9%) had MIBC. Inter‐reader agreement was excellent ( κ = 0.943–0.986). Adjusted GEE models showed lower accuracy (OR, 0.624; 95% CI: 0.518–0.751) for inexperienced readers. Tumor location was not associated with overall accuracy (global p = 0.295), and the location‐by‐experience interaction was nonsignificant (global p = 0.469). The ureteral orifice showed the highest false‐negative rates (24/82 [29.3%] and 32/82 [39.0%], respectively) in experienced and inexperienced readers, whereas the dome showed the highest false‐positive rates (10/58 [17.2%] and 15/58 [25.9%], respectively). Only the lateral‐wall false‐positive rate differed significantly by experience (6.1% vs. 14.8%, respectively). Data Conclusion Reader experience was independently associated with overall VI‐RADS performance, whereas tumor location was not associated with overall accuracy. Descriptive analyses identified location‐specific diagnostic pitfalls. Level of Evidence 3. Technical Efficacy Stage 2.