Reproducibility of image interpretation in MRI of the prostate: application of the sextant framework by two different radiologists

Abstract

The aim of this study was to reproduce prostate cancer (PCA) localization by MRI based on prostatic sextants (right and left base, middle, and apex) with minimal systematic error. Combined endorectal/body-phased-array-coil MRI of the prostate at 1.5 T was retrospectively evaluated twice, with an interval of more than 1 month, by each of two independent radiologists (R1 readings R11 and R12, and R2 readings R21 and R22) in 23 patients (age 51-75 years) who had radical prostatectomy within 1 month of MRI. PCA stage was pT2 in 14 patients, and pT3 in nine. Median Gleason score was 7 (range 5-9). Histopathology showed 83 sextants with PCA and 55 without. Reproducibility of sextant positions was within one MRI slice (3 mm) in over 80% of cases. For PCA localization, ROC analysis (AUC=0.584+/-0.048-0.724+/-0.043) yielded no significant intra-reader differences. R11 and R21 differed slightly (P=0.035). Intra-observer agreement (kappa=0.52-0.58) exceeded inter-observer agreement (kappa=0.35-0.45). Intra-observer Spearman correlation (r=0.72-0.74) exceeded inter-observer correlation (r=0.43-0.51) for sextants with PCA, but not for sextants without (r=0.69-0.74). Per-sextant localization and reporting provides a highly reliable framework in MRI of the prostate. MRI of the prostate should be followed up by the same radiologists to minimize systematic error of interpretation.