| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Clinical Radiology, University of Munich, Marchioninistrasse 15,
Munich 81377, Germany.
2 Radiologisches Zentrum Muenchen-Pasing, Munich, Germany.
3 Present address: Pathologie Lachnerstrasse, Munich, Germany.
Received March 19, 2006;
accepted after revision June 28, 2006.
Address correspondence to A. Graser.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Combined endorectal-pelvic phased-array prostatic MRI scans obtained at 1.5 T of 106 patients with biopsy-proven prostate cancer who had undergone radical prostatectomy with whole-mount step section histopathology within 28 days of MRI were retrospectively analyzed by three independent abdominal radiologists (reviewers 1, 2, and 3). Sextants of the prostate (right and left base, middle, and apex) were evaluated for the presence of prostate cancer and extracapsular extension (ECE) using a 5-point confidence scale. Data were statistically analyzed using receiver operating characteristic (ROC) analysis. Interobserver variability was assessed by kappa statistics. For calculation of sensitivity and specificity, data from the 5-point confidence scale were dichotomized into negative (score of 1-3) or positive (score of 4 or 5) findings.
RESULTS. Forty-one patients had ECE (tumor stage T3), and 65
patients had organ-confined disease (stage T2). Of 636 prostatic sextants, 417
were positive for prostate cancer and 135 were positive for ECE at
histopathology. For prostate cancer localization, ROC analysis yielded area
under the ROC curve (AUC) values ranging from 0.776 ± 0.023 (SD) to
0.832 ± 0.027. For the detection of ECE, the AUC values ranged from
0.740 ± 0.054 to 0.812 ± 0.045. Interobserver agreement
(
) ranged from 0.49 to 0.60 for prostate cancer localization and from
0.59 to 0.67 for the detection of ECE.
CONCLUSION. Using the sextant framework, independent observers reach similar accuracy with moderate to substantial agreement for the localization of prostate cancer and ECE by means of MRI of the prostate.
Keywords: genitourinary imaging • imaging-histopathology correlation • MRI • oncologic imaging • prostate cancer
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Interobserver agreement about the presence of prostate cancer was moderate in the study by Scheidler et al. [2], which was based on the analysis of prostatic sextants. Nonetheless, for follow-up studies with MRI of the prostate before and during nonsurgical therapy, it is crucial that the systematic error of interpretation be minimized. By applying the sextant scheme for prostate cancer and extracapsular extension (ECE) localization and comparing those findings with whole-mount step section histopathology results, we sought to verify the diagnostic abilities of combined endorectal-pelvic phased-array coil MRI of the prostate. Furthermore, we aimed to quantify interobserver agreement of three independent abdominal radiologists for interpreting MRI of the prostate.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Patients underwent digital rectal examination before the endorectal coil was placed to exclude anorectal disease or injury that would prohibit placement or inflation of the endorectal coil balloon. Endorectal coil balloons were routinely filled with 80-100 cm3 of air and gently pulled back to ensure complete coverage of the prostatic apex.
Patient Population
Patients were retrospectively identified either by review of pertinent
logbooks that contained data on all examinations performed on the 1.5-T MRI
scanner (scans obtained before September 30, 1998) or by a computerized search
of our radiology information system (RIS) (Medavis RIS, version 1.9, Medavis)
using the internal code for this specific examination (examinations performed
after October 1, 1998). A database containing all patient data with the names
and addresses of the referring urologists was generated. Subsequently, all
clinical data—including biopsy results, prostate-specific antigen (PSA)
values, results of digital rectal examinations, and histopathology
reports—were obtained by one of the authors. Thus, a total of 315
patients who underwent MRI for local staging of biopsy-proven prostate cancer
were identified, of which 186 patients subsequently underwent radical
retropubic prostatectomy for resection of prostate cancer.
For exact correlation of MR images with histopathology results, only the patients whose prostatectomy specimens were reviewed in a whole-mount step fashion (n = 106) were included in this study. These 106 patients (mean age ± SD, 63.0 ± 6.8 years) formed the study population. Patients underwent radical retropubic prostatectomy within 3 months of MRI of the prostate. The mean PSA value was 11.5 ± 6.4 ng/mL (range, 3.1-73.0 ng/mL).
MR Sequences and Image Interpretation
Sagittal, coronal, and axial localizer images were obtained to check the
position of the endorectal coil and to define the positions of subsequent MRI
sequences. Axial T2-weighted fast spin-echo images covered the prostate and
seminal vesicles. Coronal T2-weighted fast spin-echo images of the pelvis were
centered on the prostate. Axial T1-weighted spin-echo images covered the
pelvis from the iliac crest to the symphysis pubis to identify enlarged lymph
nodes and bone lesions. A second axial T1-weighted turbo spin-echo sequence
that included only the prostate was also obtained using the same slice
positions as in the axial T2-weighted sequence to check for postbiopsy
hemorrhage. The technical parameters of the MRI pulse sequences are listed in
Table 1.
|
MR images were interpreted on a PACS workstation (Image Devices PACS, Image Devices). MR images were considered to be of diagnostic quality when the prostate and the pelvis were adequately depicted and the images did not show pronounced artifacts from motion or from a hip prosthesis or other metal devices placed at surgery. Three independent radiologists (reviewers 1, 2, and 3) who knew that all the patients had prostate cancer but were unaware of the other patient data interpreted the MR images of all the patients. All three observers are board-certified radiology attending physicians with fellowship training in abdominal imaging and interpretation experience in MR images of the prostate that had been acquired at different institutions and that exceeded 800 cases (reviewer 1), 400 cases (reviewer 2), and 600 cases (reviewer 3).
Based on anatomic landmarks on axial T2-weighted MR images, the prostate was divided into six sextants (left and right apex, middle gland, and base) that provided the framework for interpretation and reporting of MRI findings [16]. According to previous work [3], the base of the prostate extended from the bladder floor and seminal vesicle-ejaculatory duct junctions to the level craniad to the axial MR image section with the largest transverse diameter of the prostate. The middle gland extended from the level of the largest transverse diameter of the prostate to the caudal (inferior) level of the verumontanum. The apex extended from the next caudal level to the urogenital diaphragm (Figs. 1A, 1B, 1C, 1D, 1E, and 1F). Each of the radiologists independently determined the respective slice positions in the axial T2-weighted images of the first levels of the middle gland and the apex using magnified views on the PACS workstation before evaluating the respective prostatic sextants for the presence of prostate cancer.
|
|
|
|
|
|
Data from the Excel sheet were used for calculation of prostate cancer staging accuracy dichotomizing the data from the 5-point scale. Sensitivity, specificity, and overall staging accuracy were calculated.
Pathology Review and Corroboration of MRI Findings
After radical prostatectomy, each prostate was coated in ink and fixed in
10% buffered formaldehyde. Each prostatectomy specimen was then attached to a
gauze bandage simulating its position anterior to the rectum in an attempt to
preserve the natural shape of the organ. After transfer of each specimen to
the pathologist's office, transverse step sections were obtained at 3- to 4-mm
intervals in a plane perpendicular to the long axis (base to apex) of the
prostate parallel to the MRI plane. Sections were stained with H and E.
For exact correlation of histopathology findings with MRI findings, whole-mount step sections were digitized using a commercially available flat-bed scanner (ScanExpress 1200, Mustek). All prostate cancer foci were outlined in green, areas of ECE were marked in red, and the urethra was marked in yellow by one pathologist with long-standing experience and a special interest in pathology of the male reproductive system.
Whole-mount step sections were then displayed on a notebook computer next to the PACS monitor for evaluation. MRI slices were displayed on our institution's PACS at the largest image display setting provided by the PACS workstation. Axial and coronal T2-weighted images were displayed side by side on the dual monitor system of the PACS. Correlation of MRI and histopathology slices was based on visual assessment of the size and shape of the respective slice and anatomic landmarks (verumontanum, apex, base).
Correlation of histopathology results and MRI results was based on the sextant framework. Histopathology findings located in the apex and distal periurethral tissues were correlated with the apex at MRI, as defined earlier. Histopathology findings identified at the lateral and posterior middle were correlated with the middle gland at MRI. Histopathology findings seen at the prostatic base, proximal periurethral tissues, and bladder neck were correlated with the prostatic base at MRI. Histopathology findings seen anteriorly in the prostatic base and middle were assumed to extend into the transitional or central zone unless they were clearly marked as being confined to the peripheral zone.
A mean prostate cancer Gleason score was recorded for all tumors.
Statistical Analysis
Interobserver reproducibility of respective first-level positions of the
middle gland and the apex was determined for reviewers 1 and 2, reviewers 2
and 3, and reviewers 1 and 3. Differences were stated as deviation between the
three observers.
The accuracy of the MRI interpretations compared with the pathology results was evaluated with ROC analysis based on data from the 5-point confidence scale according to methods previously published [20].
|
|
The sensitivity and specificity and positive and negative predictive values were each calculated for prostate cancer per-sextant localization and for detection of ECE using the dichotomized data from the ROC analysis.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Prostate cancer was present in 417 (65.6%) and absent from 219 (34.4%) of the 636 sextants at histopathology (Figs. 2A and 2B). In all cases, prostate cancer was located within or extended macroscopically into the peripheral zone of the prostate at whole-mount histopathology. None of the patients had a tumor that was confined to the central gland only. The median prostate cancer Gleason score was 6 (range, 4-9).
ECE was present in 135 (21.2%) and absent from 501 (78.8%) of the 636 sextants at histopathology (Figs. 3A and 3B). Forty-one patients had ECE (stage T3), and 65 patients had organ-confined disease (stage T2). ECE was found in one sextant in three patients, in two sextants in five patients (10 total sextants), in three sextants in 29 patients (87 total sextants), in four sextants in six patients (24 total sextants), and in five and six sextants in one patient, respectively.
|
|
The sensitivity, specificity, and results of ROC analysis with 95% CIs for localization of prostate cancer on a per-sextant basis by means of MRI are listed in Table 3. For ECE detection, results are listed in Table 4. For prostate cancer localization and ECE detection, there were no statistically significant differences between ROC areas under the curve for all three observers.
|
|
Agreements between interpretations on a per-sextant basis among the three
independent radiologists were significantly lower (p < 0.001) for
prostate cancer localization (reviewer 1 vs 2: 499/636 sextants, overall
agreement of 78.5%, = 0.57; reviewer 2 vs reviewer 3: 488/636
sextants, overall agreement of 76.7%, = 0.55; reviewer 1 vs reviewer
3: 475/636 sextants, overall agreement of 74.7%, = 0.53) than for ECE
localization (reviewer 1 vs reviewer 2: 568/636 sextants, overall agreement of
89.3%, = 0.77; reviewer 2 vs reviewer 3: 543/636 sextants, overall
agreement of 85.4%, = 0.75; reviewer 1 vs reviewer 3: 552/636
sextants, overall agreement of 86.8%, = 0.78). These values are
consistent with moderate agreement for prostate cancer localization and
substantial agreement for ECE localization.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Although local tumor extent influences prognosis in patients with prostate cancer, until now most studies have investigated MRI staging accuracy rather then its ability to localize single tumor foci. Therefore, we undertook our study to correlate MRI findings with whole-mount step section histopathology. The whole-mount approach enables exact localization and size assessment of single tumor foci. At step section histopathology, studies based on whole mounts found higher incidences of stage T3 tumors than studies using a standard workup of suspicious areas only [25, 26]. Considering the prognostic importance of local tumor stage, working up prostatectomy specimens is warranted. Based on the shape of single whole-mount sections, correlation of MRI and histopathology findings was performed.
Dividing the prostate into sextants (right and left bases, middle gland, and apex) on the basis of morphologic and anatomic criteria [2, 3] provides a framework for localization of intraprostatic findings that follows closely the way in which urologists perform biopsies and report their findings. In this study, we sought to show that the prostatic base, mid gland, and apex can be reliably identified by different observers at MRI. Among the three observers, results differed by no more than one MRI slice in more than 95% of patients. Because anatomic landmarks that have to be identified for correct division of the prostate into sextants are well visualized and can be readily identified on endorectal coil MR images of the prostate, this pattern can be used in reporting MRI of the prostate in the clinical routine [16]. Although it has been argued that findings of lesions suspicious for prostate cancer on MR images should be reported on a per-nodule basis rather than on a per-sextant basis [27], there is as yet no evidence to suggest that findings of individual nodules are more easily reproduced than findings of signs indicating prostate cancer in an individual sextant of the prostate. Also, it has not been established that reporting on a per-nodule basis without the use of a framework for orientation would improve sensitivity, specificity, or clinical utility of MRI of the prostate [27].
Based on a standard protocol for MRI of the prostate, image interpretation for the presence or absence of prostate cancer essentially relies on axial and coronal T2-weighted images [11, 13]. In T2-weighted images, healthy prostate tissue in the peripheral zone shows a bright homogeneous signal that is set off against the low-signal rim of the prostatic capsule on the outside. On the inside, healthy prostate tissue contrasts with the mixed low to medium signal of the prostatic pseudocapsule, central zone, and transitional zone [10, 13]. Prostate cancer occurs in the peripheral zone in 70-80% of cases [11, 14, 28] and is frequently recognized by its low signal intensity [11]. However, foci of prostate cancer are easily missed when they measure less than 4 mm2 [29], and other disease entities may mimic prostate cancer when they have low signal intensity on T2-weighted MR images [30]. Such entities include foci of chronic or granulomatous prostatitis, nodules of benign prostatic hyperplasia in the peripheral zone [13, 30], or areas of hemorrhage or scarring secondary to prostate biopsy [11, 13, 14].
In a study that localized prostate cancer to a sextant of the peripheral
zone on the basis of combined MRI and MR spectroscopy, Scheidler et al.
[2] found that interobserver
agreement on MRI findings was moderate, with a kappa value of 0.43. In our
study, we found kappa values of 0.49-0.60 for prostate cancer localization,
indicating moderate to substantial interobserver agreement. Higher values were
achieved for localization of ECE ( = 0.59-0.67). Overall agreement on
per-sextant localization of prostate cancer was significantly lower than on
localization of ECE. The three observers reached comparable overall staging
accuracies based on ROC data.
Analysis of sextants showed that single sextants containing prostate cancer at histopathology were missed by all three observers. Review of these sextants showed only slight, if any, signal abnormalities and lack of characteristic prostate cancer imaging features such as a low-signal-intensity mass. Relatively low kappa scores show that there is substantial inconsistency among observers in the individual perception and interpretation of signal abnormalities in the prostatic peripheral zone associated with the presence of prostate cancer. Our study results indicate that some prostate cancer foci were missed by all three observers. In addition, we found interpretation errors concerning sextants where all observers detected an abnormality but interpreted the findings differently. In general, observers used values 2-4 (with 2 meaning probably no prostate cancer; 3, indeterminate; and 4, probably prostate cancer) from the 5-point confidence scale more often in these sextants, leading to interobserver disagreement. This underlines that previously published signs of prostate cancer [9, 10, 13] leave too much room for individual interpretation.
Similarly, single sextants containing ECE were missed by all observers. Review of these sextants showed that they contained only minimal ECE of 1 mm or less on histopathology. Our results underline that if only small clusters of tumor cells are seen outside the prostatic capsule at histopathology, they cannot be visualized on MRI using current techniques. Probably, the hypointense signal routinely encountered in the neurovascular bundle region renders detection of tumor spread more difficult.
The pathologist marked extracapsular tumor spread on the whole-mount sections using red. Because the digitized whole-mount specimens were displayed next to the MR images, the possibility of misinterpretation of the pathologist's written report was excluded in our study. In addition, working up all prostatectomy specimens in a whole-mount fashion reduces sampling error to a minimum because the entire prostate is sectioned and analyzed.
Because we sought to correlate findings by sextant, our pathologist carefully outlined each prostate cancer focus on the specimens. Reviewing the digitalized whole-mount specimens of all 106 patients showed that all but eight patients had multifocal prostate cancer. In a substantial number of cases, very small separate foci of prostate cancer were identified that were not connected to large tumor foci. This suggests that in previous studies that did not use whole mounts for correlation, small tumor foci may have been missed. In addition, in 21 patients tumor foci within the central and transitional zones were identified. In 18 of these 21 cases, extension of tumor into the central gland could be visualized retrospectively. Due to the mixed T2 signal intensity of hyperplastic transitional zone tissue, tumor foci cannot be reliably identified prospectively. The use of MRI in combination with MR spectroscopy and contrast-enhanced perfusion imaging seems a promising method by which to overcome this limitation [31].
Comparing MRI findings with whole-mount prostatectomy specimens allows exact correlation of findings. In our study, we sought to eliminate every possible confounding factor that may lead to misinterpretation of pathology findings. We directly compared single MRI slices with respective whole-mount specimen slices on two adjacent monitors. This direct comparison yielded interesting results: On the one hand, there were easy cases with MR signal abnormalities corresponding exactly to histopathologically proven foci of prostate cancer. On the other hand, there were difficult cases that showed very subtle and noncharacteristic imaging findings; in these latter cases, observer disagreement was observed more frequently.
Our results of 83-93% overall staging accuracy agree with those published in the literature. The use of the sextant framework allows a reproducible way of reporting MRI findings in prostate imaging. Being forced to exactly localize tumor foci may lead to improved staging accuracy. We found that revealing histopathology findings to the radiologists participating in this study was a very educational experience and may have induced a learning curve.
The use of MRI in prostate cancer staging is warranted only if the results of this examination influence patient management in a substantial number of patients. Certainly, performing MRI with an endorectal coil is the most exact imaging technique to date. It should be used in patients with clinically localized prostate cancer only to allow reproducibly good results.
Several studies have emphasized the importance of observer experience in MRI of the prostate. A multicenter study examining the interpretation performance of nine different observers in MRI of the prostate reported accuracies between 69% and 79% [12]. In our study, we did not find significant differences among the three observers, which was probably due to their similar levels of experience.
Our study has several limitations. First, observers were aware of the presence of prostate cancer in all cases; therefore, differentiation of prostate cancer foci and inflammatory changes of the peripheral zone may have been influenced. All other clinical information, including serum PSA values, was withheld. Prostate cancer staging is one of the major indications for prostatic MRI, and our setting therefore resembles the clinical routine. In some of the cases, overcalling of subtle signal abnormalities may have occurred because observers knew about the presence of prostate cancer.
Another limitation represents the lack of MR spectroscopy, which was not performed in our institution at the time this study was conducted. It has been well documented that MR spectroscopy increases sensitivity and specificity in prostate cancer localization and in the localization of ECE [27, 32, 33]. The widespread use of MRI without spectroscopy, however, still represents routine clinical practice in preoperative prostate cancer staging in many centers.
In our study, all three observers had extensive experience in interpreting MRI of the prostate. However, it has already been shown that radiologists without genitourinary imaging training perform poorly compared with genitourinary radiologists in interpreting MRI of the prostate [12]. Although observer experience differed slightly among the single observers of our study, each of the three observers seemed to have a specific pattern of recognizing and calling abnormalities and reporting imaging findings in MRI of the prostate. Although one of the observers tended to overcall prostate cancer presence in respective prostatic sextants, the other two observers were more specific, reporting fewer false-positive findings.
In conclusion, MRI of the prostate with combined endorectal-pelvic phased-array coils represents the state-of-the-art imaging technique for local tumor staging in patients with biopsy-proven prostate cancer. Application of the sextant framework to MRI of the prostate provides a highly reliable aid in the reporting and communication of MRI findings.
Our study underlines that preoperative local staging of prostate cancer relies on state-of-the-art MRI performed by expert observers.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  R. J. Stanley The Challenge of Managing Prostate Cancer Am. J. Roentgenol., January 1, 2007; 188(1): 1 - 1. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
