Purpose: To assess the accuracy of using several criteria to evaluate endorectal coil magnetic resonance (MR) images for penetration of the prostatic capsule.
Materials and methods: Thirty patients with prostate carcinoma underwent MR imaging and prostatectomy. Specified sites of potential capsular penetration on MR images were blindly evaluated by three readers for six diagnostic criteria. These evaluations were compared with pathologic findings of capsular penetration and were analyzed by means of receiver operating characteristic (ROC) analysis.
Results: The area under the ROC curve (Az) for the readers' overall impression of capsular penetration varied from .72 to .77. Highest mean Az's were for the criteria of capsular thickening (.74) and nodular extracapsular tumor (.72), although the latter finding had poor sensitivity (15%). Interobserver variation was low for all findings.
Conclusion: Sensitivity and specificity were generally low for the diagnostic criteria. The usefulness of endorectal coil MR imaging in staging prostate cancer may be limited by the lack of diagnostic signs that uniformly identify extracapsular penetration.