Erschienen in:
13.03.2017 | Magnetic Resonance
Development and validation of a logistic regression model to distinguish transition zone cancers from benign prostatic hyperplasia on multi-parametric prostate MRI
verfasst von:
Yuji Iyama, Takeshi Nakaura, Kazuhiro Katahira, Ayumi Iyama, Yasunori Nagayama, Seitaro Oda, Daisuke Utsunomiya, Yasuyuki Yamashita
Erschienen in:
European Radiology
|
Ausgabe 9/2017
Einloggen, um Zugang zu erhalten
Abstract
Purpose
To develop a prediction model to distinguish between transition zone (TZ) cancers and benign prostatic hyperplasia (BPH) on multi-parametric prostate magnetic resonance imaging (mp-MRI).
Materials and methods
This retrospective study enrolled 60 patients with either BPH or TZ cancer, who had undergone 3 T-MRI. We generated ten parameters for T2-weighted images (T2WI), diffusion-weighted images (DWI) and dynamic MRI. Using a t-test and multivariate logistic regression (LR) analysis to evaluate the parameters’ accuracy, we developed LR models. We calculated the area under the receiver operating characteristic curve (ROC) of LR models by a leave-one-out cross-validation procedure, and the LR model’s performance was compared with radiologists’ performance with their opinion and with the Prostate Imaging Reporting and Data System (Pi-RADS v2) score.
Results
Multivariate LR analysis showed that only standardized T2WI signal and mean apparent diffusion coefficient (ADC) maintained their independent values (P < 0.001). The validation analysis showed that the AUC of the final LR model was comparable to that of board-certified radiologists, and superior to that of Pi-RADS scores.
Conclusion
A standardized T2WI and mean ADC were independent factors for distinguishing between BPH and TZ cancer. The performance of the LR model was comparable to that of experienced radiologists.
Key Points
• It is difficult to diagnose transition zone (TZ) cancer.
• We performed quantitative image analysis in multi-parametric MRI.
• Standardized-T2WI and mean-ADC were independent factors for diagnosing TZ cancer.
• We developed logistic-regression analysis to diagnose TZ cancer accurately.
• The performance of the logistic-regression analysis was higher than PIRADSv2.