Prostate magnetic resonance imaging: The truth lies in the eye of the beholder

Purpose: To determine the diagnostic accuracy and interobserver variability of radiologic interpretation of magnetic resonance imaging (MRI) performed for surgical planning before prostatectomy. Patients and methods: The records of 233 men undergoing prostatectomy with presurgical multiparametric 3T surface body coil MRI were reviewed. All initial films were read by a fellowship-trained body radiologist provided with relevant clinical information. A senior radiologist then reread all pelvic MRIs blinded to the initial interpretation with findings from both readings compared to final pathology. Kappa (κ) scores as well as sensitivity, specificity, positive predictive values (PPV), negative predictive value (NPV), and accuracy were determined. Results: When considering extraprostatic extension (EPE), there was low concordance comparing the initial vs. repeat MRI interpretation (κ = 0.22). Additionally, when the senior radiologist reread his own initial interpretation (n = 93, blinded to initial result), concordance for EPE was greater (κ = 0.36) albeit similarly low. With regard to EPE, a comparison of initial MRI interpretation vs. reread by senior radiologist noted universal improvements in diagnostic characteristics including sensitivity (30.3% vs. 56.1%), specificity (80.2% vs. 88.6%), PPV (37.7% vs. 66.1%), NPV (74.4% vs. 83.6%), and accuracy (66.1% vs. 79.4%). In contrast, seminal vesicle invasion interpretation was more uniform whereby initial MRI interpretation vs. reread yielded similar sensitivity (18.2% vs. 27.3%), specificity (97.2% vs. 93.8%), PPV (40.0% vs. 31.6%), NPV (91.9% vs. 92.5%), and accuracy (89.7% vs. 87.6%). Conclusions: Even at a tertiary referral center, interobserver variability among radiologists regarding local extent of disease on prostate MRI is high. These observations underscore the importance of uniformity when defining criteria for EPE and seminal vesicle invasion to allow for optimal presurgical planning.