TY - JOUR
T1 - MR imaging in adenocarcinoma of the prostate
T2 - Interobserver variation and efficacy for determining stage C disease
AU - Schiebler, M. L.
AU - Yankaskas, B. C.
AU - Tempany, C.
AU - Spritzer, C. E.
AU - Rifkin, M. D.
AU - Pollack, H. M.
AU - Holtz, P.
AU - Zerhouni, E. A.
PY - 1992
Y1 - 1992
N2 - Patients with adenocarcinoma of the prostate confined to the gland (stage B) are candidates for a potentially curative surgical procedure (radical retropubic prostatectomy). However, patients with adenocarcinoma that penetrates the capsule or invades the seminal vesicles (stage C) are no longer considered good candidates for surgical cure of their disease. The purpose of this study was to compare the ability of four radiologists to detect stage C disease on MR images and to evaluate interobserver variability. One hundred consecutive MR studies of the prostate were reviewed independently by four radiologists to determine whether the cancer was stage C (capsule penetration or seminal vesicle invasion by tumor). A radical prostatectomy was performed in each case, and careful histologic assessment was made of the prostatic capsule and seminal vesicles for any evidence of stage C disease. The sensitivity, specificity, and accuracy (true-positive + true-negative/100 patients) in detecting stage C disease were calculated for each of the four readers. Four receiver-operating-characteristic curves were generated and compared by means of the univariate z score. Percentage agreement was calculated for five specific areas of the prostate on MR images, and observations made by the best reader were compared with the other three to help determine interreader variability. The results showed that the sensitivity and specificity of MR imaging in detecting stage C disease ranged from .24 to .61 (mean, .48) and .49 to .79 (mean, .66), respectively. The accuracy of MR imaging ranged from .47 to .61 (mean, .55). The univariate z score test showed that one of the readers significantly differed from the other three. The average percentage agreement between that reader and the other three was 70% for the five separate anatomic regions. This study shows that considerable interobserver variation exists in the interpretation of MR images for staging cancer of the prostate. The average accuracy among four radiologists in determining the presence of stage C adenocarcinoma of the prostate from MR images was only slightly above a chance guess at .55.
AB - Patients with adenocarcinoma of the prostate confined to the gland (stage B) are candidates for a potentially curative surgical procedure (radical retropubic prostatectomy). However, patients with adenocarcinoma that penetrates the capsule or invades the seminal vesicles (stage C) are no longer considered good candidates for surgical cure of their disease. The purpose of this study was to compare the ability of four radiologists to detect stage C disease on MR images and to evaluate interobserver variability. One hundred consecutive MR studies of the prostate were reviewed independently by four radiologists to determine whether the cancer was stage C (capsule penetration or seminal vesicle invasion by tumor). A radical prostatectomy was performed in each case, and careful histologic assessment was made of the prostatic capsule and seminal vesicles for any evidence of stage C disease. The sensitivity, specificity, and accuracy (true-positive + true-negative/100 patients) in detecting stage C disease were calculated for each of the four readers. Four receiver-operating-characteristic curves were generated and compared by means of the univariate z score. Percentage agreement was calculated for five specific areas of the prostate on MR images, and observations made by the best reader were compared with the other three to help determine interreader variability. The results showed that the sensitivity and specificity of MR imaging in detecting stage C disease ranged from .24 to .61 (mean, .48) and .49 to .79 (mean, .66), respectively. The accuracy of MR imaging ranged from .47 to .61 (mean, .55). The univariate z score test showed that one of the readers significantly differed from the other three. The average percentage agreement between that reader and the other three was 70% for the five separate anatomic regions. This study shows that considerable interobserver variation exists in the interpretation of MR images for staging cancer of the prostate. The average accuracy among four radiologists in determining the presence of stage C adenocarcinoma of the prostate from MR images was only slightly above a chance guess at .55.
UR - http://www.scopus.com/inward/record.url?scp=0026606423&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0026606423&partnerID=8YFLogxK
U2 - 10.2214/ajr.158.3.1738994
DO - 10.2214/ajr.158.3.1738994
M3 - Article
C2 - 1738994
AN - SCOPUS:0026606423
SN - 0361-803X
VL - 158
SP - 559
EP - 562
JO - American Journal of Roentgenology
JF - American Journal of Roentgenology
IS - 3
ER -