A comparative analysis of techniques for measuring tumor contact length in predicting extraprostatic extension

Abstract

Purpose

This study aims to evaluate the diagnostic performance of curvilinear and linear measurement methods in different magnetic resonance imaging (MRI) sequences for detecting extraprostatic extension (EPE) in prostate cancer, and to evaluate the added value of apparent diffusion coefficient (ADC) in detecting EPE.

Methods

A retrospective analysis was conducted on 84 patients who underwent multiparametric MRI (mp-MRI) prior to radical prostatectomy between January 2019 and February 2022. Tumor contact length (TCL) was assessed curvilinearly and linearly on T2-weighted imaging (T2WI), ADC maps, and dynamic contrast-enhanced (DCE) MRI by two radiologists. MRI-based EPE positivity was defined as a curvilinear or linear contact length of >15 mm. Statistical comparisons were conducted using chi-squared and independent samples t-tests, with interreader agreement evaluated using weighted κ statistics. Univariate and multivariate logistic regression identified independent predictors of EPE, and two prediction models were constructed. Diagnostic performance was assessed using receiver operator characteristic (ROC) curve analysis.

Results

A total of 32 (38%) and 52 (62%) patients with EPE and non-EPE, respectively, were included in this study. Patients with EPE demonstrated significantly larger tumor sizes, lower ADC values, and lower ADC ratios than those without EPE (p < 0.001). The curvilinear and linear TCL measurements for each sequence exhibited statistically significant correlations with EPE for both readers, with strong interreader agreement. Curvilinear TCL (c-TCL) and linear TCL (l-TCL) on DCE-MRI showed higher area under the curve (AUC) values than the other measurements for EPE prediction (reader 1: 0.815 and 0.803, reader 2: 0.746 and 0.713, respectively). However, there was no statistically significant difference between c-TCL and l-TCL. Multivariable models with mean ADC value improved predictive performance. Model 2 (ADC, ISUP, and c-TCL on DCE images) surpassed model 1 (ADC and c-TCL on DCE images) with an AUC of 0.919 and 0.874, respectively.

Conclusion

DCE-MRI demonstrated superior performance in predicting EPE compared to other sequences. Linear and curvilinear measurements had comparable diagnostic performance. Being more practical and easier, radiologists may use l-TCL measurement in daily practice. The mean ADC value provided additional diagnostic value.