bp-VI-RADS combined with normalized apparent diffusion coefficient enhances diagnostic efficacy for muscle-invasive bladder cancer

Background: Magnetic resonance imaging (MRI) facilitates the preoperative staging of bladder cancer through the Vesical Imaging-Reporting and Data System (VI-RADS), which employs a 5-point scale to assess the likelihood of muscle invasion. However, the VI-RADS relies on qualitative evaluation, and diagnostic uncertainty arises for patients with a score of 3, for whom muscle invasiveness cannot be determined. Biparametric magnetic resonance imaging (bp-MRI), which incorporates T2-weighted and diffusion-weighted imaging (DWI), has a diagnostic accuracy comparable to that of multiparametric protocols without the need for contrast agents. The apparent diffusion coefficient (ADC) derived from DWI reflects tumor aggressiveness, but is susceptible to technical variability, necessitating the normalization of the ADC to improve reproducibility. This study aimed to evaluate the diagnostic value of the normalized apparent diffusion coefficient (nADC) in bp-MRI for muscle-invasive bladder cancer (MIBC) cases, and to investigate whether integrating nADC with the VI-RADS enhances staging performance, particularly for uncertain lesions with a score of 3. Methods: This study retrospectively enrolled 172 bladder cancer patients who underwent MRI between January 2021 and March 2023. Three radiologists performed VI-RADS scoring on bp-MRI, two of whom manually delineated regions of interest (ROIs) on ADC maps using three-dimensional Slicer software. The mean apparent diffusion coefficient (ADCmean) value was calculated by measuring ROIs across three tumor sites. Bladder urine and obturator internus muscle (OIM) ROIs were used to normalize the data, and the nADC was calculated using the following formula: nADC = ADCmean (tumor)/ADC (reference). The Wilcoxon rank-sum test was used to assess the association between the nADC values and muscle invasion, and a receiver operating characteristic (ROC) curve analysis was conducted to determine the thresholds. Binomial logistic regression was used to construct composite models integrating the VI-RADS with nADC/ADC metrics. The DeLong test was used to compare the area under the curve (AUC) values. Results: The AUC for the VI-RADS was 0.87 (cut-off value ≥4, sensitivity 0.69, specificity 0.94). Among the individual parameters, the urine-normalized apparent diffusion coefficient (nADCurine) performed the best (AUC =0.82), while the ADCmean and the obturator internus muscle normalized apparent diffusion coefficient (nADCoim) exhibited equivalent performance (AUC =0.80). The composite models uniformly outperformed the VI-RADS alone (all P