Prostate Cancer - RSNA Publications Online

ORIGINAL RESEARCH

䡲 GENITOURINARY IMAGING

Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights.

Prostate Cancer: Endorectal MR Imaging and MR Spectroscopic Imaging—Distinction of True-Positive Results from Chance-detected Lesions1 Jeffrey J. Hom, BA Fergus V. Coakley, MD Jeffry P. Simko, MD Aliya Qayyum, MB, BS Ying Lu, PhD Lars Schmitt Peter R. Carroll, MD John Kurhanewicz, PhD

1 From the Departments of Radiology (J.J.H., F.V.C., A.Q., Y.L., J.K.), Anatomic Pathology (J.P.S., L.S.), and Urology (P.R.C.), University of California, San Francisco, 505 Parnassus Ave, Box 0628, M-372, San Francisco, CA 941430628. Received September 29, 2004; revision requested December 2; final revision received May 20, 2005; accepted June 13. Supported by NIH grants IRGICA764230IRI, R01 CA59897, and R01 CA79980. J.J.H. supported by UCSF Dean’s Award for Summer Research. L.S. supported by NCI SPORE grant p50.

Purpose:

To retrospectively investigate size criteria for the identification of chance-detected lesions at endorectal magnetic resonance (MR) imaging and MR spectroscopic imaging of prostate cancer.

Materials and Methods:

Approval of the committee on human research and written informed consent were obtained. This study was HIPAA compliant. Endorectal MR imaging and MR spectroscopic imaging were performed with a 1.5-T MR imager in 48 men with a mean age of 59 years (age range, 47–75 years) prior to radical prostatectomy. Two independent readers recorded the size and location of all suspected peripheral zone tumor nodules on MR images alone and on images obtained with combined MR imaging and MR spectroscopic imaging. Nodules detected at MR imaging were classified as matched lesions if tumor was present in the same location at step-section histopathologic review. For all matched lesions, ␬ values were calculated to examine agreement between measured and actual tumor size. Lesions that were overmeasured at MR imaging with a ␬ value of less than 0.2 were considered chance-detected lesions.

Results:

At MR imaging, two of 27 and four of 35 matched lesions for readers 1 and 2, respectively, were chance-detected lesions. The corresponding numbers of lesions at combined MR imaging and MR spectroscopic imaging were one of 21 and one of 31, respectively. In all but two cases, the measured diameter of chance-detected lesions was more than twice that of the diameter at histopathologic analysis. By using this diameter threshold to distinguish true-positive results, the mean diameter of detected tumors at histopathologic analysis was 15 mm compared with 4 mm for both undetected and chance-detected tumors (P ⬍ .05).

Conclusion:

To ensure uniformity in the comparison of scientific studies, peripheral zone tumors detected at MR imaging and MR spectroscopic imaging of the prostate that are in the same location as tumors detected at histopathologic review should be considered chance-detected lesions if the MR transverse diameter is more than twice the histopathologic transverse diameter. 娀 RSNA, 2006

姝 RSNA, 2006

192

Radiology: Volume 238: Number 1—January 2006

GENITOURINARY IMAGING: Endorectal MR Imaging and MR Spectroscopic Imaging

I

n cancer imaging, location match combined with a positive biopsy finding is generally considered sufficient evidence to classify a lesion as a truepositive finding. For example, if a computed tomographic (CT) scan shows a spiculated mass in the lung and a biopsy sample of the lesion is positive for malignancy, it is implicitly and reasonably assumed that the abnormality observed at CT is a true-positive result and that the extent of this abnormality conforms to the anatomic boundary of the tumor. The definition of a true-positive result at endorectal magnetic resonance (MR) imaging and MR spectroscopic imaging of prostate cancer, however, is more problematic. Foci of low T2 signal intensity or abnormal metabolism is considered indicative of prostate cancer, but this is not a specific finding; both MR imaging and MR spectroscopic imaging are associated with false-positive results (1). Findings at MR imaging and MR spectroscopic imaging may not be clearly demarcated, and they often merge gradually into areas of normal peripheral zone T2 signal intensity or metabolism. Sextant biopsy is an imperfect technique used for tumor sampling and localization (2). The limited specificity of MR findings, the lack of clear lesion boundaries, and the poor correlation between sextant biopsy and radical prostatectomy specimens for tumor localization make it difficult to know when a focus of cancer has been depicted with MR imaging or MR spectroscopic imaging. That is, a simple match by lesion location is insufficient for diagnosis because a large lesion may coincidentally encompass a small focus of cancer and be inappropriately considered a truepositive result. For example, two independent readers interpreting preoperative endorectal combined MR imaging and MR spectroscopic imaging findings in 37 patients undergoing radical prostatectomy detected 35 and 39 of 51 prostate cancer nodules that were confirmed with step-section histopathologic analysis by using location match as the definition of a true-positive result (1). However, when the same database was used in another study in which a trueRadiology: Volume 238: Number 1—January 2006

positive result required a match of lesion location and size, the numbers of nodules detected by two independent readers were 11 and 21 (3). In the latter study, lesion size was considered matched if the maximum transverse diameter measured at MR imaging or MR spectroscopic imaging was within 50%– 150% of the maximum transverse diameter measured at histopathologic analysis. This size range was not supported by systematic analysis, but it had been arbitrarily established on the basis of the reported 18%–33% shrinkage of the prostate that occurs during fixation (4,5) and the need to provide reasonable allowance for differences in registration and morphology between imaging and histopathologic findings. The lack of established size criteria for a true-positive result is a substantial limitation of research into endorectal MR imaging and MR spectroscopic imaging of prostate cancer, particularly as these modalities become widely available (although they are still uncommon in clinical practice) and are the subject of an ongoing open American College of Radiology Imaging Network multi-institutional trial. Thus, we undertook this study to retrospectively investigate size criteria for the identification of chancedetected prostate cancer lesions at endorectal MR imaging and MR spectroscopic imaging.

Materials and Methods Subjects We retrospectively identified all patients who underwent radical prostatectomy during the 2-year period between January 2001 and December 2002 and had undergone preoperative endorectal MR imaging and MR spectroscopic imaging of the prostate at our institution (n ⫽ 56). Patients were referred for endorectal MR imaging and MR spectroscopic imaging after the diagnosis of prostate cancer had been established with transrectal ultrasonography– guided biopsy. Patients were recruited into an ongoing National Institutes of Health study in which the use of MR imaging in the evaluation of prostate cancer was

Hom et al

being investigated. This study was approved by our institutional committee on human research, and written informed consent was obtained from all patients. In addition, our retrospective study was specifically approved by this same committee, with waiver of the requirement for written informed consent. Our study was compliant with the Health Insurance Portability and Accountability Act. Patients were excluded from the study if processing of the prostatectomy specimen precluded creation of tumor maps (n ⫽ 4), if they had received preoperative hormonal or radiation therapy (n ⫽ 2), or if technical artifacts precluded interpretation of MR images (n ⫽ 2). The final study population consisted of 48 men with a mean age of 59 years (age range, 47–75 years). The median tumor Gleason score was 7 (Gleason score range, 5–9). The mean prostate-specific antigen level at diagnosis was 7.8 ng/mL (range, 1.7–24.0 ng/ mL). The mean period between biopsy and endorectal MR imaging and MR spectroscopic imaging was 62 days (range, 3–149 days). The mean interval between endorectal MR imaging and MR spectroscopic imaging and surgical resection was 75 days (range, 7–248 days).

MR Imaging Technique MR imaging was performed with a 1.5-T whole-body MR imager (Signa; GE Medical Systems, Milwaukee, Wis). Pa-

Published online 10.1148/radiol.2381041675 Radiology 2006; 238:192–199 Author contributions: Guarantor of integrity of entire study, J.J.H.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; literature research, J.J.H.; clinical studies, F.V.C., J.P.S., A.Q., L.S., P.R.C., J.K.; statistical analysis, Y.L.; and manuscript editing, all authors Address correspondence to F.V.C. (e-mail: [email protected]). Authors stated no financial relationship to disclose.

193


tients were examined in the supine position. A body coil was used for excitation, and a pelvic phased-array coil (GE Medical Systems) was used in combination with a commercially available balloon-covered expandable endorectal coil (Medrad, Pittsburgh, Pa) for signal reception. Sequences performed included thin-section high-spatial-resolution transverse and coronal T2-weighted fast spin-echo imaging of the prostate and seminal vesicles with the following parameters: repetition time msec/effective echo time msec, 5000/96; echo train length, 16; section thickness, 3 mm; no intersection gap; field of view, 14 cm; matrix, 256 ⫻ 192; anteroposterior frequency direction; and three signals acquired. All MR images were routinely postprocessed to compensate for the reception profile of the endorectal and pelvic phased-array coils.

MR Spectroscopic Imaging Technique and Evaluation After review of the transverse T2weighted MR images, an MR spectroscopic imaging volume was selected by the technologist performing the examination to maximize coverage of the prostate while minimizing inclusion of periprostatic fat and rectal air. Technologists who perform MR spectroscopic examination of the prostate at our institution have had extensive training and experience, with more than 4000 such examinations having been performed at our institution in the past decade. Three-dimensional MR spectroscopic imaging data were acquired with use of a water- and lipid-suppressed doublespin-echo point-resolved spectroscopic sequence, with spectral spatial pulses used for the two 180° excitation pulses. The spectral spatial pulses allowed for both sharp volume selection and frequency selection to reduce water resonance and suppress lipid resonance (6,7). The influence of chemical shift on the apparent location of the selected volume was also reduced by the higher spectral bandwidth of the spectral spatial pulses (6,7). Outer voxel saturation pulses were also used to further sharpen volume selection and conform the selected volume to the shape of the 194

prostate to eliminate susceptibility artifact from periprostatic fat and rectal air (8). Data sets were acquired as 16 ⫻ 8 ⫻ 8 phase-encoded spectral arrays (1024 voxels with a spatial resolution of 0.24 – 0.34 cm3) with 1000-msec repetition time, 130-msec echo time, and 17minute acquisition time. MR spectroscopic imaging data were zero filled from eight to 16 sections in both the anteroposterior direction and the craniocaudad direction to increase the likelihood of optimal alignment between the spectroscopic voxels and the peripheral zone. The total examination time, including coil placement and patient positioning, was 1 hour. Spectroscopic imaging data were overlaid on the corresponding transverse T2-weighted images and evaluated by a spectroscopist (J.K.) with 15 years of experience in spectroscopic imaging of the prostate to determine which voxels were suitable for analysis. Individual voxels were considered suitable if they consisted of at least 75% peripheral zone tissue, did not include tissue surrounding the urethra or ejaculatory ducts, had a signal-to-noise ratio greater than 5:1, and were not spectroscopically contaminated by insufficient water or fat suppression. The mean number of peripheral zone voxels per patient was 98 (range, 43–160 peripheral zone voxels per patient); of these 98 voxels, a mean number of 9 voxels (range, 6 –15 voxels) were unsuitable for analysis because of insufficient peripheral zone tissue, and a mean number of 4 voxels (range, 0 –12 voxels) were unsuitable because of inadequate signal intensity or spectral contamination. A score of 1–5 was assigned to each useable voxel, such that a score of 1 was considered probably benign; a score of 2, possibly benign; a score of 3, equivocal; a score of 4, possibly malignant; and a score of 5, probably malignant. The details of this scoring system have been described previously. In brief, a primary score of 1–5 was assigned on the basis of the ratio of choline plus creatine to citrate. An initial adjustment was made to the primary score on the basis of the choline-to-creatine ratio and the

Hom et al

polyamine peak. A final adjustment was made to incorporate poor voxel signalto-noise ratio. This scoring system has demonstrated high interobserver agreement, with ␬ values of 0.79 – 0.80 (9). The final MR spectroscopic images consisted of transverse T2-weighted images with an overlaid grid showing both the corresponding spectra and the spectral score on the scale.

Interpretation of MR Images and MR Spectroscopic Findings Two attending radiologists (F.V.C. and A.Q., with 8 and 5 years of experience, respectively, in the interpretation of endorectal MR images and MR spectroscopic images of the prostate) independently reviewed all MR images without knowledge of MR spectroscopic findings. Studies were reviewed on a picture archiving and communication system workstation (Impax; Agfa, Mortsel, Belgium). Readers were aware that patients had prostate cancer, but they were unaware of all other clinical and histopathologic findings. Readers marked the location and maximum transverse diameter of each suspected peripheral zone tumor nodule on a standardized diagram of the prostate. Central gland nodules were not recorded because these tumors were observed infrequently (1) and there was a lack of established MR imaging or MR spectroscopic imaging criteria. More than one tumor focus could be recorded for each patient, since prostate cancer is often multifocal. Readers also recorded the transverse, anteroposterior, and craniocaudal diameters of both the entire prostate and the central gland. After an interval of at least 2 weeks, readers reviewed the MR images again in combination with the MR spectroscopic imaging findings, including both the spectra and the assigned voxel scores, to minimize recall or learning bias. Readers again recorded the size and location of all suspected peripheral zone tumor nodules on the standardized diagram of the prostate. Readers used their professional judgment and experience rather than fixed objective criteria to identify tumors, but—in general—a tumor was defined at MR imaging and Radiology: Volume 238: Number 1—January 2006


MR spectroscopic imaging as a masslike nodule of low T2 signal intensity and as a cluster of voxels demonstrating abnormal metabolism, respectively. Of note, no changes were made to our MR imaging or MR spectroscopic imaging protocols during the study period.

Histopathologic Interpretation Specimens removed at radical prostatectomy were marked with ink and fixed overnight in 10% buffered formalin. Transverse step sections were obtained at 3– 4-mm intervals in a plane perpendicular to the prostatic urethra, and these samples yielded between six and 10 tissue slices per specimen. Apical margin, bladder (base) margin, step sections adjacent to these margins, and alternate sections between these margins were all sliced into quarters and submitted in their entirety for histopathologic analysis, thus yielding 18 –24 blocks per case. All slides were examined in consensus by two reviewers (J.P.S. and J.J.H.) who were unaware of MR imaging and MR spectroscopic findings; these reviewers measured and recorded the size, location, and Gleason score of all tumor foci on a standardized diagram of the prostate. The presence of extracapsular extension or seminal vesicle invasion was also noted. Statistical Analysis The unit of analysis was the peripheral zone tumor nodule. In the analysis, each suspected peripheral zone tumor nodule recorded by the readers on the schematic prostate diagram was compared with the true location of peripheral zone tumor nodules on the histopathologic tumor map by one of the reviewers (J.J.H.) who was trained in the general recognition of prostate cancer at MR imaging and histopathologic analysis but was not specifically aware of such data for patients in this study. A tumor detected with MR imaging was considered to be a matched lesion when the tumor was at least partially present in the same region on the histopathologic tumor map. Reasonable allowances were made for differences in registration and deformation between histopathologic analysis and MR imaging, but good Radiology: Volume 238: Number 1—January 2006

transverse anatomic correlation between MR images and histopathologic findings was required for nodules to be classified as “matched.” That is, a tumor seen at MR imaging was only considered “matched” to a tumor seen at histopathologic analysis if the suspected focus was in the same relative portion of the prostate in both craniocaudal and transverse directions, as confirmed by localization relative to fixed anatomic landmarks such as the urethra and distinctive nodules of benign prostatic hyperplasia. With use of the ␬ statistic, tumor size was then used to distinguish truepositive matched lesions from chancedetected lesions. Specifically, MR tumor volume (VMR) and histopathologic tumor volume (VPath) were estimated from the transverse diameter by assuming that the tumors were spherical and that the volume of overlap (VO) between nodules identified at MR imaging and nodules identified at histopathologic analysis was the smaller of the two volumes (ie, the smaller of imaging and histopathologic volumes). Peripheral zone volume (VPZ) was calculated as the difference between the total gland volume and central gland volume by using the ellipsoid formula from the mean dimensions measured by the two readers. The ␬ value was then calculated to examine the agreement between imaging and histopathologic size for each matched lesion by using the following formula (10): ␬ ⫽ [VO/VPZ ⫺ (VMR/VPZ 䡠 VPath/VPZ)]/[1 ⫺ (VMR/VPZ 䡠 VPath/VPZ)]. In this context, the ␬ value indicates the likelihood that lesions seen at MR imaging and histopathologic analysis overlap by chance alone, with adjustment for the effect of peripheral zone volume. An imaging nodule is more likely to overlap by chance a larger truepositive tumor than a smaller true-positive tumor; furthermore, this overlap is more likely to occur in a small peripheral zone than in a large peripheral zone. Possible ␬ values range from a minimum value of ⫺1 (ie, complete disagreement) to a maximum value of 1 (ie, complete agreement); a ␬ value of 0 indicates there is no agreement. Lesions overmeasured at MR imaging with a ␬

Hom et al

value of less than 0.2 were considered chance-detected lesions. This threshold was chosen to define chance-detected lesions because a ␬ value of less than 0.2 indicated only slight or worse agreement, whereas a ␬ value of more than 0.2 indicated fair or better agreement (10,11). Having established a statistical method to define chance-detected lesions, we then examined the relationship between the ␬ value and the ratio of tumor diameter at MR imaging to tumor diameter at histopathologic analysis to determine if a simpler comparison of tumor diameter at MR imaging to true tumor diameter could be used as a surrogate measure of the ␬ value. We used a generalized estimation equation to compare tumor sizes (ie, independent variables) for true-positive, false-positive, and false-negative lesions (ie, fixed dependent variables) to account for potential data clustering or dependency created by the presence of more than one nodule per patient and by the repeated observations of two readers (ie, random-effect variables).

Results Histopathologic Findings A total of 113 peripheral zone tumor nodules, with a mean size of 8 mm (median size, 6 mm; size range, 1–32 mm), were identified in the 48 men in the study. Twenty-three patients had a single nodule, 10 had two nodules, three had three nodules, and 12 had more than three separate nodules. Thirtyeight patients had organ-confined tumors, eight had extracapsular extension alone, and two had both extracapsular extension and seminal vesicle invasion. MR Imaging and MR Spectroscopic Imaging Findings At MR imaging alone, readers 1 and 2 detected 27 and 35 matched lesions, respectively. At combined MR imaging and MR spectroscopic imaging, readers 1 and 2 detected 21 and 31 matched lesions, respectively (Fig 1). A ␬ value of less than 0.2 was used to define overmeasurement of lesions on MR images as chance-detected lesions. Two of 27 195


matched lesions and four of 35 matched lesions for readers 1 and 2, respectively, were classified as chance-detected lesions at MR imaging. Similarly, one of 21 matched lesions and one of 31 matched lesions for readers 1 and 2, respectively, were classified as chancedetected lesions at combined MR imaging and MR spectroscopic imaging. At MR imaging alone, examination of the relationship between the ␬ value and the ratio of the tumor diameter at MR imaging to the tumor diameter at histopathologic analysis showed that in all but two cases the measured diameter of chance-detected lesions was two times greater than the histopathologic diameter, and none of the matched lesions with a ␬ value of more than 0.2 had a measured diameter that was more than two times greater than the actual diameter. Similarly, in all but one case, the measured diameter of chance-detected lesions at combined MR imaging and MR spectroscopic imaging was more than twice the measured diameter at histopathologic analysis; none of the matched lesions with a ␬ value greater than 0.2 had a measured diameter that was more than twice the actual diameter (Fig 2). This diameter threshold was used to distinguish true-positive results, and the mean histopathologic diameter (depending on reader and modality) of detected tumors was 15–17 mm compared with 5– 6 mm for undetected tumors (P ⬍ .001) and 3– 4 mm for chance-detected tumors (Table).

account for chance detection in the case of a large lesion seen at MR imaging that coincidentally encompasses a small tumor detected at histopathologic analysis. In addressing this issue, our study indicates that peripheral zone lesions detected at MR imaging and MR spectroscopic imaging that are more than twice the diameter of a corresponding histopathologic lesion should be consid-

Hom et al

ered chance-detected lesions (ie, falsepositive findings). With this size threshold, true-positive lesions were significantly larger than chance-detected lesions (mean diameter, 15 mm vs 4 mm, respectively; P ⬍ .05) and chancedetected and undetected lesions had the same mean diameter (ie, 4 mm). These latter findings support our hypothesis that chance-detected lesions are really a

Figure 1

Discussion The definition of a true-positive result at endorectal MR imaging and MR spectroscopic imaging of patients with prostate cancer is not straightforward, since a location match alone is insufficient for diagnosis. To illustrate the inadequacy of this definition, one need only consider the extreme situation in which the entire peripheral zone is interpreted as malignant in patients undergoing a staging MR examination; this would result in an accuracy of 100%. Use of the sextant as the unit of analysis enables this problem to be partially overcome; however, this does not 196

Figure 1: (a) Transverse T2-weighted (5000/96) endorectal MR image obtained in a 62-year-old man with a prostate-specific antigen level of 4.0 ng/mL and a positive transrectal biopsy specimen showing cancer with a Gleason score of 6 in the right gland. Reader 2 interpreted an ill-defined area of low signal intensity in the left peripheral zone as a 13-mm tumor. The part of this area within the white box corresponds to the histopathologic specimen shown in b. (b) Photomicrograph of the histopathologic specimen corresponding to the white box in a (original magnification, ⫻20). Only a small 2-mm focus (area within the oval) of cancer with a Gleason score of 3 is present, and the remaining peripheral zone is free of tumors. The 13-mm lesion seen by reader 2 at MR imaging likely represents chance detection of the 2-mm focus seen at microscopy and should be considered a false-positive result. (c) MR spectroscopic image obtained at the level of the transverse image in a shows unremarkable metabolic patterns (high citrate peak and nonelevated choline peak resonances) in the voxels corresponding to the area of low T2 signal intensity. Neither reader considered this area to contain tumor at interpretation of combined MR imaging and MR spectroscopic imaging findings, which corresponds to the utility of MR spectroscopic imaging in reducing the likelihood of false-positive results. Radiology: Volume 238: Number 1—January 2006


subset of undetected lesions, and they should be considered false-positive findings. This conclusion has implications primarily for clinical research studies that use radical prostatectomy specimens as the reference standard, and it is not clinically applicable to patients who do not undergo surgery. The increasing interest in MR imaging and MR spectroscopic imaging of the prostate indicates the importance of a uniform definition of a true-positive result, and our findings contribute to a working definition of this concept. One could argue that misclassification of chance-detected lesions as true-positive results would have had a relatively minor effect on the detection rate in this analysis, given that chance-detected lesions accounted for no more than four (11%) of 35 matched lesions for either reader. However, the subjectivity of in-

terpretation of MR images of the prostate is well known. Both readers in our study were experienced and conservative in their assessment of tumors. Readers with a different approach who designate large areas of the prostate as malignant could conceivably identify many more chance-detected lesions; lack of an appropriate size boundary to classify these chance-detected lesions as false-positive findings could subvert the comparability of studies with different readers. While our findings are primarily applicable to research studies, they also have implications for daily clinical practice. Readers who routinely describe large abnormalities in the prostate should recognize that there is a strong likelihood that some of these abnormalities are chance-detected lesions; they should not, for example, be falsely reas-

Figure 2

Figure 2: Graphs show relationship between ␬ values (indicating the level of agreement between imaging size and histopathologic size) and the ratio of imaging to histopathologic diameter for readers 1 and 2, both for MR imaging alone and for combined MR imaging and MR spectroscopic imaging. Chance-detected lesions were defined as overmeasured lesions with a ␬ value of less than 0.2. At MR imaging alone, in all but two cases (horizontal arrows), the measured diameter of chance-detected lesions was more than twice that of the histopathologic diameter, and none of the matched lesions with a ␬ value greater than 0.2 had a measured diameter that was more than twice the actual diameter. In all but one case (vertical arrow), the measured diameter of chance-detected lesions at combined MR imaging and MR spectroscopic imaging was more than twice that at histopathologic analysis, and none of the matched lesions with a ␬ value of more than 0.2 had a measured diameter that was more than twice the actual diameter. Radiology: Volume 238: Number 1—January 2006

Hom et al

sured by positive histopathologic findings in the same region at sextant biopsy. Furthermore, our study may have wider implications, given the increasing rate of detection of small lesions because of improved spatial resolution at CT scanning and MR imaging and the current interest in screening for small colonic polyps and lung nodules as precursors of colon cancer and lung cancer, respectively. Traditional correlation of radiologic and histopathologic findings becomes problematic in such cases when multiple small lesions may be seen at imaging and may be difficult to correlate with histopathologic specimens. Our study provides a template for the investigation of chance-detected lesions in these settings. In a previous study, we required that the diameter of a matched lesion at endorectal MR imaging and MR spectroscopic imaging of the prostate be between 50% and 150% of the lesion diameter at histopathologic analysis before it could be considered a truepositive finding (1). This size range was arbitrary, but it was based on reasonable allowances for differences arising from fixation, deformation, shrinkage, and registration. Few studies, in correlating imaging and histopathologic findings, have addressed the issue of size as a criterion in addition to location as part of the definition of a true-positive lesion. One analogous setting is the correlation of CT colonoscopy and optical colonoscopy, in which the potential multiplicity and variable size of polyps create a situation where a location match is an incomplete criterion for a true-positive finding. Interestingly, a recent study of CT colonoscopy used a size range of 50%– 150% as the window for lesion match by colonic segment for a lesion to be considered a true-positive finding (12). This size range appears to have been arbitrary and unsupported by systematic analysis. The findings of our current study suggest a preliminary threshold based on data for defining chance-detected lesions at MR imaging and MR spectroscopic imaging of prostate cancer, providing an upper cutoff of 200% for 197


chance-detected lesions. We deliberately chose not to provide a lower boundary for two reasons. First, our prior study of tumor volume, in which all lesions matched by location were considered to be true-positive lesions, showed that gross overmeasurement occurred in many small lesions and that these lesions likely represented chancedetected lesions. Marked overestimation of tumor volume when a true-positive lesion is defined by size alone was also observed in a similar study performed in Italy (13). We did not observe the opposite phenomenon—namely, gross undermeasurement of larger tumors—to any appreciable degree. Second, undermeasurement is fundamentally different from overmeasurement. Drawing a large lesion that coincidentally encompasses a tiny focus of tumor can be likened to the stopped clock that is right twice a day, whereas drawing a small lesion that is part of a large tumor may be likened to seeing the tip of an iceberg; failure to appreciate the entire iceberg does not negate its presence. The histopathologic determinants of prostate cancer visibility at endorectal MR imaging and MR spectroscopic imaging are poorly understood; however, given the heterogeneity of prostate cancer, it is entirely possible that lowergrade or less fibrotic portions of the tumor may not be seen as well at imaging and may account for undermeasurement of tumors. Conversely, tumor overmeasurement may reflect tumor shrinkage during fixation, differences in deformation between the resected specimen and the prostate compressed by the endorectal coil during MR imaging, or peritumoral inflammation or postbiopsy changes. Our study had several limitations. It was performed retrospectively at a single academic institution with a small number of readers and use of currently available technology. The validity of extrapolating our results to other centers and the effect of future technologic advances are unknown. Incomplete MR spectroscopic coverage of the peripheral zone and the exclusion of central gland cancer foci and the effect of postbiopsy hemorrhage from our analysis 198

Hom et al

Comparison of True-Positive, Chance-detected, and Undetected Lesions for Each Reader and Modality Lesion Type, Modality, and Reader True-positive lesions MR imaging Reader 1 Reader 2 MR imaging and MR spectroscopy Reader 1 Reader 2 Chance-detected lesions MR imaging Reader 1 Reader 2 MR imaging and MR spectroscopy Reader 1 Reader 2 Undetected lesions MR imaging Reader 1 Reader 2 MR imaging and MR spectroscopy Reader 1 Reader 2

No. of Lesions*

Mean Diameter (mm)

95% Confidence Interval

25 (24) 32 (31)

16 15

12.9, 19.1 12.5, 17.5

21 (20) 30 (26)

17 15

13.9, 20.0 12.7, 17.9

2 (2) 4 (4)

3 4

1.6, 4.4 0.2, 8.8

0 (0) 1 (1)

NA 4

NA 4.0, 4.0

89 (34) 82 (34)

5 5

4.0, 6.8 3.6, 6.2

93 (38) 84 (35)

6 5

4.2, 7.1 3.7, 6.3

Note.—NA ⫽ not applicable. * Data in parentheses are number of patients.

are additional limitations. MR images were reviewed separately from combined MR spectroscopic and MR images, with a minimum interval of 2 weeks, but learning or recall bias still could have occurred. We have not reported the number of false-positive findings (other than chance-detected lesions) or the effect of lesion size on detection rates because both false-positive findings and detection rates can be calculated only after criteria for truepositive results have been determined. These criteria were not apparent at the start of this study; however, now that they have been established, we can investigate these other parameters. Likewise, other intriguing issues, including analysis of interobserver agreement, the incremental effect of MR spectroscopic imaging, the role of image quality, and the use of morphologic criteria to identify cancer (14), are all contingent on a uniform definition of a truepositive result, which was not available to us at the start of the study.

Despite these issues, our tumor detection rates (ie, 21–35 of 113 peripheral zone lesions), which depended on the reader and modality, may seem low. We suspect the low detection rate reflects the meticulous nature of the histopathologic evaluation used in our study and the small volume of disease seen in patients undergoing radical prostatectomy at our institution (as a consequence of strict surgical selection criteria). The effect of different histopathologists on the number of tumor foci found at histopathologic examination is illustrated by considering that the mean number of detected tumor nodules per patient in this study was 2.4 (113 nodules in 48 patients) compared with 1.4 (51 nodules in 37 patients) in our prior study of a similar population where specimens were examined by a different pathologist (1). The mean interval from endorectal combined MR imaging and MR spectroscopic imaging to surgery was 75 days (range, 7–248 days). Tumor progresRadiology: Volume 238: Number 1—January 2006


sion during this interval could have been a confounding variable; however, this was unlikely to have been an important confounding variable given the indolent nature of prostate cancer in patients with low-risk tumors (15), such as the majority of the men in this study. Determining whether a nodule seen at MR imaging is matched by location with a nodule seen at histopathologic analysis also introduces an element of subjective judgment and potential bias. This is a general problem with studies that correlate histopathologic specimens with MR findings in patients with prostate cancer, although it is difficult to conceive of a practical method to improve on this reference standard. In our study, we took care to adopt a conservative approach, deeming nodules matched only if there was good transverse anatomic correlation between findings at MR imaging and histopathologic analysis. ␬ Statistical analysis, as applied in this study, is not without limitations, among which is the oversimplified assumption that all tumor nodules are spherical. This study used two-dimensional transverse diameter instead of volumetric analysis in three-dimensional space as a proxy for lesion size; we believed that this would closely conform to what could be expected in clinical practice, since even in research studies it is unusual to perform volumetric tumor measurements. Our use of ␬ statistical analysis as a tool to define chance agreement is innovative and makes use of a categorical scale to interpret continuous data. While this approach can be criticized, ␬ analysis has the advantage of having a widely used interpretative scale available for measurement of agreement (10,11). MR spectroscopic imaging is a rapidly developing technology, and it is possible that our findings might change with the use of 3-T MR imagers or improvements in spectroscopic resolution. Even so,

Radiology: Volume 238: Number 1—January 2006

Hom et al

readers interpreted MR spectroscopic imaging findings in conjunction with T2weighted MR imaging findings. T2weighted MR imaging is a relatively stable technology, and we found similar results when comparing the MR imaging and MR spectroscopic imaging technique with MR imaging alone. This suggests our conclusions may be relatively unaffected by ongoing developments in MR spectroscopic imaging. In conclusion, a match of both size and location is required for a finding to be classified as true-positive at endorectal MR imaging and MR spectroscopic imaging of prostate cancer since overestimation of tumor size at imaging may represent a chance-detected lesion. To ensure uniformity in the comparison of scientific and clinical studies, peripheral zone tumors detected at MR imaging and MR spectroscopic imaging of the prostate that are in the same location as tumors detected at histopathologic analysis should be considered chance-detected lesions if the MR transverse diameter is more than twice the histopathologic transverse diameter.

10. Landis JR, Koch GG. The measurement of observed agreement for categorical data. Biometrics 1977;33:159 –174.

References

11. Kundel HL, Polansky M. Measurement of observer agreement. Radiology 2003;228: 303–308.

1. Coakley FV, Kurhanewicz J, Lu Y, et al. Prostate cancer tumor volume: measurement with endorectal MR and MR spectroscopic imaging. Radiology 2002;223:91–97. 2. Wefer AE, Hricak H, Vigneron DB, et al. Sextant localization of prostate cancer: comparison of sextant biopsy, magnetic resonance imaging and magnetic resonance spectroscopic imaging with step section histology. J Urol 2000;164:400 – 404. 3. Dhingsa R, Qayyum A, Coakley FV, et al. Prostate cancer localization with endorectal MR imaging and MR spectroscopic imaging: effect of clinical data on reader accuracy. Radiology 2004;230:215–220. 4. Partin AW, Epstein JI, Cho KR, Gittlesohn AM, Walsh PC. Morphometric measurement of tumor volume and percent of gland involvement as predictors of pathological

stage in clinical stage B prostate cancer. J Urol 1989;141:341–345. 5. McNeal JE, Bostwick DG, Kindrachuck RA, Redwine EA, Freiha FS, Stamey TA. Patterns of progression in prostate cancer. Lancet 1986;1:60 – 63. 6. Star-Lack J, Pauly J, Vigneron DB, Kurhanewicz J, Nelson SJ. Improved solvent suppression and increased spatial excitation bandwidths for 3-D PRESS CSI using phasecompensating spectral/spatial spin-echo pulses. J Magn Reson Imaging 1997;7:745– 757. 7. Schricker AA, Pauly JM, Kurhanewicz J, Swanson MG, Vigneron DB. Dualband spectral-spatial RF pulses for prostate MR spectroscopic imaging. Magn Reson Med 2001; 46:1079 –1087. 8. Tran TK, Vigneron DB, Sailasuta N, et al. Very selective suppression pulses for clinical MRSI studies of brain and prostate cancer. Magn Reson Med 2000;43:23–33. 9. Jung JA, Coakley FV, Vigneron DB, et al. Prostate depiction at endorectal MR spectroscopic imaging: investigation of a standardized evaluation system. Radiology 2004;233: 701–708.

12. Pickhardt PJ, Choi JR, Hwang I, et al. Computed tomographic virtual colonoscopy to screen for colorectal neoplasia in asymptomatic adults. N Engl J Med 2003;349:2191– 2200. 13. Ponchietti R, Di Loro F, Fanfani A, Amorosi A. Estimation of prostate cancer volume by endorectal coil magnetic resonance imaging vs pathologic volume. Eur Urol 1999;35:32– 35. 14. Cruz M, Tsuda K, Narumi Y, et al. Characterization of low-intensity lesions in the peripheral zone of prostate on pre-biopsy endorectal coil MR imaging. Eur Radiol 2002; 12:357–365. 15. Choo R, DeBoer G, Klotz L, et al. PSA doubling time of prostate carcinoma managed with watchful observation alone. Int J Radiat Oncol Biol Phys 2001;50:615– 620.

199