Development and Validation of an MRI‐Based Nomogram for Preoperative Detection of Muscle Invasion in VI‐RADS 3

Abstract

BackgroundThe relationship between tumor and muscle layer in the vesical imaging‐reporting and data system (VI–RADS) 3 is ambiguous, and there is a lack of preoperative and non‐invasive procedures to detect muscle invasion in VI‐RADS 3.PurposeTo develop a nomogram based on MRI features for detecting muscle invasion in VI–RADS 3.Study TypeRetrospective.Population235 cases (Age: 67.5 ± 11.5 years) with 11.9% females were randomly divided into a training cohort (n = 164) and a validation cohort (n = 71).Field Strength/Sequence3T, T2‐weighted imaging (turbo spin‐echo), diffusion‐weighted imaging (breathing‐free spin echo), and dynamic contrast‐enhanced imaging (gradient echo).Assessment3 features were selected from the training cohort, including tumor contact length greater than maximum tumor diameter (TCL > Dmax), flat tumor morphology, and lower standard deviation of apparent diffusion coefficient (ADCSD). Three readers assessed VI‐RADS scores and the tumor morphology.Statistical TestsInterobserver agreement was assessed by Kappa analysis. Features for final analysis were selected by logistic regression. The performance of the nomogram was evaluated by the receiver operating characteristic curve, decision curve analysis, and calibration curve.ResultsTCL > Dmax, flat morphology, and lower ADCSD were the independent risk factors for muscle invasive in VI‐RADS 3. The AUCs, accuracy, sensitivity, and specificity of the nomogram 1 composed of three features for detecting muscle invasion were 0.852 (95% CI: 0.793–0.912), 0.756, 0.917, and 0.663 in the training cohort, and 0.885 (95% CI: 0.801–0.969), 0.817, 0.900, and 0.784 in the validation cohort. The nomogram 2 without ADCSD has nearly the same performance as the nomogram 1.Data ConclusionNomogram can be an efficient tool for preoperative detection of muscle invasion in VI–RADS 3.Level of Evidence3Technical EfficacyStage 2