Published online before print November 16, 2007, 10.1148/radiol.2453062042
(Radiology 2008;246:177-184.)
© RSNA, 2007
Peripheral Zone Prostate Cancer: Accuracy of Different Interpretative Approaches with MR and MR Spectroscopic Imaging1
Antonio C. Westphalen, MD,
Fergus V. Coakley, MD,
Aliya Qayyum, MBBS,
Mark Swanson, PhD,
Jeffry P. Simko, MD, PhD,
Ying Lu, PhD,
Shoujun Zhao, PhD,
Peter R. Carroll, MD,
Benjamin M. Yeh, MD, and
John Kurhanewicz, PhD
1 From the Departments of Radiology (A.C.W., F.V.C., A.Q., M.S., Y.L., S.Z., B.M.Y., J.K.), Anatomic Pathology (J.P.S.), Urology (P.R.C.), and Epidemiology and Biostatistics (Y.L., S.Z.), University of California San Francisco, 505 Parnassus Ave, Box 0628, M-372, San Francisco, CA 94143-0628. Received November 30, 2006; revision requested January 31, 2007; revision received March 1; accepted March 21; final version accepted May 7. Supported by National Institutes of Health grants IRGICA76423-0IRI, R01 CA59897, and R01 CA79980. A.C.W. supported by National Institute of Biomedical Imaging and Bioengineering T32 Training Grant 1 T32 EB001631-01A1.
Address correspondence to A.C.W. (e-mail: Antonio.Westphalen{at}radiology.ucsf.edu).
 |
ABSTRACT
|
|---|
Purpose: To retrospectively compare relative accuracy of different interpretative approaches to magnetic resonance (MR) and MR spectroscopic imaging of peripheral zone prostate cancer, by using histologic examination results as the reference standard.
Materials and Methods: This HIPAA-compliant study had institutional Committee on Human Research approval, with waiver of written consent requirement. Spectroscopic voxels of unequivocally benign (n = 66) or malignant (n = 77) peripheral zone tissue were identified by using step-section histopathologic tumor maps created for 28 men (mean age, 60 years; range, 46–71 years) who underwent endorectal MR and MR spectroscopic imaging before radical prostatectomy. Two readers (9 and 8 years of experience) independently scored the selected voxels on a scale from 1 (likely benign) to 5 (likely malignant) at randomized review of the corresponding tissue outlined on a transverse T2-weighted MR image (T2 approach), the MR spectrum from the selected voxel only (single-voxel approach), the MR spectra from all voxels at the same axial level (multivoxel approach), and both the corresponding tissue outlined on a transverse T2-weighted image and the MR spectra from all voxels at the same axial level (integrated approach). Readers were aware that spectra were derived in patients with biopsy-proved diagnoses of prostate cancer and represented either benign or malignant tissue but were unaware of which voxels had been labeled benign or malignant and of all other clinical, histopathologic, and MR imaging findings. Receiver operating characteristic (ROC) curve analysis was performed. Generalized estimating equation method was used to estimate sensitivity and specificity for specific cutoff values.
Results: Mean areas under the ROC curve (AUCs) for the T2, single-voxel, multivoxel, and integrated approaches were 0.69, 0.72, 0.72, and 0.76, respectively. AUC of the integrated approach was significantly higher than those of the other three approaches (P < .001). 
Values for assessment of interobserver variability for the T2, single-voxel, multivoxel, and integrated approaches were 0.39, 0.39, 0.34, and 0.48, respectively.
Conclusion: Addition of MR spectroscopic imaging to MR imaging significantly improves characterization of peripheral zone prostate tissue as benign or malignant; improved performance is obtained when both data sets are interpreted in an integrated fashion.
© RSNA, 2007
|
INTRODUCTION
|
|---|
Combined endorectal magnetic resonance (MR) and MR spectroscopic imaging of the prostate has emerged as a promising method for radiologic evaluation of the prostate, allowing integrated anatomic and metabolic evaluation of prostate cancer extent and aggressiveness (1,2). In particular, the addition of the metabolic information obtained at MR spectroscopic imaging to the anatomic information obtained at MR imaging has been shown to improve tumor localization, tumor staging, and tumor volume estimation (3–8). Furthermore, MR spectroscopic imaging enables assessment of tumor aggressiveness and the detection of metabolically active tumor recurrence after radiation therapy, both of which are observations that cannot be made with MR imaging alone (9–11).
With respect to the metabolites that are of interest in the prostate, MR spectroscopic imaging yields spectra that depict the relative concentrations of choline, polyamines, creatine, and citrate in a given voxel of tissue. Research has shown that prostate cancer is characterized by variable combinations of elevated choline (a normal cell membrane constituent, which is elevated in many tumors), reduced citrate, and reduced polyamines (citrate and polyamines are constituents of normal prostatic tissue) (12). These changes are frequently reported as the metabolic ratios of choline to creatine and choline plus polyamines and creatine to citrate (both ratios are elevated in cancer) (1–5,9,12,13).
The inclusion of these ratios in scientific reports is appropriately objective, but arguably has created a perception that the interpretation of MR spectroscopic imaging results is largely quantitative and that, conversely, interpretation of MR imaging results is largely qualitative. In reality, the day-to-day clinical interpretation of MR spectra is more complex. The automatically generated ratios are often meaningless because of noise (eg, they are negative numbers or are values that are implausibly high or low), although visual evaluation of the spectra may still be possible. Some centers have the benefit of dedicated spectroscopists who can optimize postacquisition processing of spectroscopic data and perform manual case-by-case adjustments that produce better spectra than commercially available software. Some of these spectroscopists also assist in data interpretation. However, most centers do not have the benefit of such dedicated basic science personnel.
The spectra are not reviewed in isolation, but in the context of the adjacent voxels. A voxel with equivocal elevation of choline may be more or less suspicious depending on the choline level in the neighboring voxels. In practice, a voxel can only be interpreted after inspection of the corresponding T2-weighted MR images to confirm that the tissue is predominantly prostatic and not contaminated by choline in adjacent muscle or seminal vesicles. Knowledge of the findings on T2-weighted images might also consciously or subconsciously influence spectroscopic assessment. Given these considerations, it is difficult to know the true incremental contribution of spectroscopy to the characterization of prostatic tissue and the best way of interpreting MR spectroscopic studies: Should spectra be inspected in isolation, in the context of neighboring voxels, with the corresponding T2-weighted images, or with some combination of these approaches? We are unaware of any studies that have addressed this issue. Therefore, we undertook this study to retrospectively compare the relative accuracy of different interpretative approaches to MR and MR spectroscopic imaging of peripheral zone prostate cancer, by using histologic examination results as the reference standard.
|
MATERIALS AND METHODS
|
|---|
Subjects
This study was approved by our institutional Committee on Human Research, with waiver of the requirement for written consent, and was compliant with the Health Insurance Portability and Accountability Act. We retrospectively identified, through cross correlation of our surgical and radiology information systems, all patients who met the following inclusion criteria:
1. Endorectal MR and MR spectroscopic imaging of the prostate performed for biopsy-proved prostate cancer between 1999 and 2005.
2. Radical prostatectomy specimen obtained within 180 days of imaging, with step-section evaluation of the specimen.
3. No interval treatment between imaging and prostatectomy.
One hundred one patients fulfilled these criteria. Two patients were excluded because the histopathologic slides were not available (n = 1) or the MR images were not retrievable (n = 1). Ninety-nine patients remained in the study group. We randomly selected 28 of these patients to form our final study population, because preliminary sample size calculations showed that this number would be sufficient to achieve a power greater than 80% with a significance level of .05 to reject the noneffect hypothesis. The sample size calculation assumed that at least one benign and one malignant voxel would be selected in each patient, that the average area under the receiver operating characteristic (ROC) curve (AUC) would be least 0.75, and that the noneffect hypothesis corresponded to an AUC of 0.5. These 28 men had a mean age of 60 years (range, 46–71 years), a mean serum prostate specific antigen level of 10.2 ng/dL (range, 1.7–64.1 ng/dL), and a median preoperative Gleason score of 7.
MR and MR Spectroscopic Imaging Technique
MR studies were performed with a 1.5-T whole-body MR imaging unit (Signa; GE Medical Systems, Milwaukee, Wis). Patients were imaged in a supine position by using the body coil for excitation and a pelvic phased-array coil (GE Medical Systems) in combination with a commercially available balloon-covered expandable endorectal coil (Medrad, Pittsburgh, Pa) for signal reception. MR imaging sequences included thin-section high nominal 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; intersection gap, 0 mm; field of view, 14 cm; matrix, 256 x 192; anteroposterior frequency encoding (to prevent obscuration of the prostate by endorectal coil motion artifact); and three excitations.
After review of the transverse T2-weighted images, an MR spectroscopic imaging volume was selected to maximize coverage of the prostate while minimizing the inclusion of periprostatic fat and rectal air. The imaging volume was selected by two trained spectroscopists, each of whom had 7 years of experience, supervised by one of the authors (J.K., who had 15 years of experience in spectroscopic imaging). Three-dimensional MR spectroscopic imaging data were acquired by using a water- and lipid-suppressed double spin-echo point-resolved spectroscopy sequence, with spectral-spatial pulses for the two 180° excitation pulses, that was optimized for the quantitative detection of choline, creatine, polyamines, and citrate. The spectral-spatial pulses enabled both sharp volume selection and frequency selection to reduce the water resonance and suppress lipid resonances (14,15). The influence of chemical shift on the apparent location of the selected volume was dramatically reduced by the high spectral bandwidth of the spectral-spatial pulses (14,15). This reduction in chemical shift effects was evidenced by a reduction in difference in the choline plus creatine–to-citrate ratio for the left and right edge voxels of the spectral array by 25% with conventional 180° pulses to less than 1% with spectral-spatial pulses (15). Outer-voxel saturation pulses were also employed to further sharpen volume selection and conform the selected volume to the shape of the prostate (to eliminate susceptibility artifact from periprostatic fat and rectal air) (16). Data sets were acquired as 16 x 8 x 8 phase-encoded spectral arrays (1024 voxels with a spatial resolution of 0.24–0.34 cm3) with 1000/130 and a 17-minute acquisition time.
Three-dimensional MR spectroscopic imaging data were processed off-line at a workstation (UltraSparc; Sun Microsystems, Mountain View, Calif) utilizing in-house software previously developed specifically for three-dimensional MR spectroscopic imaging studies. All spectral data were apodized with a 2-Hz Lorentzian function. Data were Fourier transformed in the time domain and in three spatial domains with phase, baseline, and frequency corrections. Integrated peak area values for choline, creatine, and citrate and peak area ratios for choline to creatine and choline plus creatine to citrate were automatically calculated for each voxel. MR spectroscopic imaging data, including the spectra and associated metabolic ratios, were overlaid on the corresponding transverse T2-weighted images. The total examination time was 1 hour, including coil placement and patient positioning.
Reference Standard Histopathologic Identification of Benign and Malignant Spectroscopic Voxels
The mean interval from MR imaging to surgery was 39 days (range, 2–152 days). Specimens removed at radical prostatectomy were marked with ink and fixed overnight in 10% buffered formalin. Axial step-sections were obtained at 3–4-mm intervals in a plane perpendicular to the prostatic urethra. Apical margin, bladder (base) margin, step-sections adjacent to these margins, and alternate sections between these were all sliced into quarters and submitted entirely for histopathologic evaluation. All slides were retrospectively examined by one genitourinary pathologist (J.P.S., with 10 years of experience) who, unaware of the MR and MR spectroscopic imaging results, measured and recorded the size, location, and Gleason score of all peripheral zone cancers on a standardized diagram of the prostate.
The lead investigator (A.C.W., with 4 years of experience in genitourinary MR imaging) retrospectively reviewed the histopathologic tumor maps together with the MR and MR spectroscopic images to identify peripheral zone voxels that were clearly composed of purely benign or malignant tissue (Fig 1). Voxels were labeled as consisting of benign or malignant tissue only when there was clear-cut concordance between MR imaging and histopathologic findings and no postbiopsy hemorrhage or potential for partial volume effects (overlap of tumorous voxel with surrounding healthy tissue, ejaculatory zone, or central gland tissue) was present. Allowances were made for differences in registration between imaging and pathologic slides. A total of 143 voxels were labeled as benign (n = 66, 46%) or malignant (n = 77, 54%) in the 28 men in the final study group. An average of 5.1 imaging sections per patient (range, three to eight sections) were selected. Although the number of sections per patient was not limited, only a single voxel per section was used. The 77 malignant voxels were derived from 33 prostate cancer nodules in the 28 patients.
View larger version (73K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1: Montage shows concordance between MR imaging, MR spectroscopic imaging, and histopathologic findings in 59-year-old man with prostate cancer. Focus of low signal intensity (black box) on transverse T2-weighted MR image (5000/96) (upper left) corresponds to focus of confirmed cancer (dashed outline) on whole-mount specimen stained with hematoxylin-eosin (upper right). Bottom: The 6 MR spectroscopic imaging voxels corresponding to the tumor show a malignant metabolic pattern, with elevation of choline (arrows) and reduction of citrate (arrowheads). On the basis of concordance between the T2-weighted images and the whole-mount step-section and independently of the spectral findings, the two central (annotated) voxels can be regarded as malignant. In this fashion, we selected unequivocally benign (n = 66) or malignant (n = 77) peripheral zone voxels for inclusion in our study.
|
|
Image Interpretation
The following four kinds of images (Fig 2) were generated for each of the 143 voxels identified as benign or malignant to reflect four interpretative approaches to MR and MR spectroscopic imaging data:
View larger version (182K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2a: MR and MR spectroscopic imaging of prostate in 63-year-old man with biopsy-proved cancer. Four images were created from each selected voxel, corresponding to the four interpretative approaches studied. (a) Transverse T2-weighted MR image (5000/96), with site of the voxel highlighted (highlighted square in grid) (T2 approach). (b) MR spectra from selected voxel in isolation (single-voxel approach). Numbers are metabolic ratios—the choline-to-creatine ratio ([A] Ch/Cr) (ie, 1.91) and the choline plus polyamines and creatine–to-citrate ratio ([B] Ch+Cr/Citrate) (ie, 0.953). (c) MR spectra from all voxels at axial level of selected voxel (multivoxel approach). Numbers represent metabolic ratios—the choline-to-creatine ratio (A) and the choline plus polyamines and creatine–to-citrate ratio (B). (d) Combined transverse T2-weighted MR image and corresponding spectral grid (integrated approach). The voxel of interest is highlighted.
|
|
View larger version (17K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2b: MR and MR spectroscopic imaging of prostate in 63-year-old man with biopsy-proved cancer. Four images were created from each selected voxel, corresponding to the four interpretative approaches studied. (a) Transverse T2-weighted MR image (5000/96), with site of the voxel highlighted (highlighted square in grid) (T2 approach). (b) MR spectra from selected voxel in isolation (single-voxel approach). Numbers are metabolic ratios—the choline-to-creatine ratio ([A] Ch/Cr) (ie, 1.91) and the choline plus polyamines and creatine–to-citrate ratio ([B] Ch+Cr/Citrate) (ie, 0.953). (c) MR spectra from all voxels at axial level of selected voxel (multivoxel approach). Numbers represent metabolic ratios—the choline-to-creatine ratio (A) and the choline plus polyamines and creatine–to-citrate ratio (B). (d) Combined transverse T2-weighted MR image and corresponding spectral grid (integrated approach). The voxel of interest is highlighted.
|
|
View larger version (59K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2c: MR and MR spectroscopic imaging of prostate in 63-year-old man with biopsy-proved cancer. Four images were created from each selected voxel, corresponding to the four interpretative approaches studied. (a) Transverse T2-weighted MR image (5000/96), with site of the voxel highlighted (highlighted square in grid) (T2 approach). (b) MR spectra from selected voxel in isolation (single-voxel approach). Numbers are metabolic ratios—the choline-to-creatine ratio ([A] Ch/Cr) (ie, 1.91) and the choline plus polyamines and creatine–to-citrate ratio ([B] Ch+Cr/Citrate) (ie, 0.953). (c) MR spectra from all voxels at axial level of selected voxel (multivoxel approach). Numbers represent metabolic ratios—the choline-to-creatine ratio (A) and the choline plus polyamines and creatine–to-citrate ratio (B). (d) Combined transverse T2-weighted MR image and corresponding spectral grid (integrated approach). The voxel of interest is highlighted.
|
|
View larger version (69K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2d: MR and MR spectroscopic imaging of prostate in 63-year-old man with biopsy-proved cancer. Four images were created from each selected voxel, corresponding to the four interpretative approaches studied. (a) Transverse T2-weighted MR image (5000/96), with site of the voxel highlighted (highlighted square in grid) (T2 approach). (b) MR spectra from selected voxel in isolation (single-voxel approach). Numbers are metabolic ratios—the choline-to-creatine ratio ([A] Ch/Cr) (ie, 1.91) and the choline plus polyamines and creatine–to-citrate ratio ([B] Ch+Cr/Citrate) (ie, 0.953). (c) MR spectra from all voxels at axial level of selected voxel (multivoxel approach). Numbers represent metabolic ratios—the choline-to-creatine ratio (A) and the choline plus polyamines and creatine–to-citrate ratio (B). (d) Combined transverse T2-weighted MR image and corresponding spectral grid (integrated approach). The voxel of interest is highlighted.
|
|
1. A transverse T2-weighted image with the corresponding tissue outlined (T2 approach).
2. The MR spectrum from the selected voxel only (single-voxel approach).
3. The MR spectra from all voxels at the same axial level (multivoxel approach).
4. A transverse T2-weighted image with the corresponding tissue outlined and with the MR spectra from all voxels at the same axial level depicted (integrated approach).
All the spectral images included the two metabolic ratios created during spectral postprocessing—that is, the ratios of choline to creatine and choline plus polyamines and creatine to citrate. The lead investigator created these 572 images (ie, 143 · 4) on a personal computer (Inspiron 5100; Dell, Round Rock, Tex) after downloading them from our departmental picture archiving and communication system (PACS) (Impax; Agfa, Mortsel, Belgium) as Joint Photographic Experts Group (JPEG) files (72 dots per inch) and after performing appropriate cropping and insertion of voxel outlines by using image-editing software (Adobe Photoshop, version 7.0; Adobe Systems, Seattle, Wash). The images were then inserted into a slide show (PowerPoint, version 2002 SP3; Microsoft, Redmond, Wash) in a randomized order to create a single presentation with 572 slides. Randomization was checked to ensure that none of the four images from each studied voxel were shown in consecutive order.
Two radiologists (F.V.C. [reader 1] and A.Q. [reader 2], with 9 and 8 years of experience in the interpretation of prostate MR and MR spectroscopic images, respectively) independently reviewed the slide show file and characterized each of the 572 voxel images by using a five-point scale (1 = likely benign, 2 = possibly benign, 3 = equivocal or indeterminate, 4 = possibly malignant, and 5 = likely malignant). The readers interpreted the entire slide show in one session. Readers assigned these scores by using their best judgment, on the basis of their years of experience. Automated metabolic ratios (choline-to-creatine ratio and choline plus polyamines and creatine–to-citrate ratio) were generated and were available to readers. These automated ratios were often meaningless, and, in such cases, readers used their expert qualitative evaluation rather than fixed quantitative criteria. Such qualitative evaluation included examining the polyamine "peak" (really a "dip"). Readers were aware that the spectra were derived in patients with a biopsy-proved diagnosis of prostate cancer and represented either benign or malignant tissue but were unaware of which voxels had been labeled benign or malignant and of all other clinical, histopathologic, and MR imaging findings.
Statistical Analysis
Reader scores of 1–5 were compared with the histopathologic reference standard of benign or malignant for each voxel. The interpretative accuracy for each reader and approach was calculated by using ROC curve analysis. Interreader agreement was evaluated with statistics calculated by using generalized estimating equations for scores of 1–3 versus scores of 4 and 5 (17). The degree of observer agreement was graded as follows: A value of 0–0.20 indicated slight agreement; a value of 0.21–0.40, fair agreement; a value of 0.41–0.60, moderate agreement; a value of 0.61–0.80, substantial agreement; and a value of 0.81–1.00, almost perfect agreement (18). Descriptive statistical data (sensitivity, specificity, and positive and negative predictive values) were determined by using a dichotomized rating system in which voxel scores of 1–3 indicated benignity and voxel scores of 4 and 5 indicated malignancy. Similarly, we used the generalized estimating equation method to estimate the sensitivity and specificity for specific cutoff values. AUCs of true-positive versus false-positive proportions of malignant voxels were generated on the basis of Wilcoxon approaches. We then used a mixed-effect analysis of variance model for jackknife pseudovalues (the Dorfman-Berbaum-Metz method) to evaluate the differences between approaches, as well as reader difference (19). The Tukey adjustment was used for P values of pairwise multiple comparisons. The analysis was performed by using software (SAS, version 9.1; SAS Institute, Cary, NC). A P value of less than .05 was considered to indicate a statistically significant difference.
|
RESULTS
|
|---|
A flowchart summarized our study design and results (Fig 3). The mean AUCs for the T2, single-voxel, multivoxel, and integrated approaches were 0.69, 0.72, 0.72, and 0.76, respectively (Table, Fig 4), with the AUC for the integrated approach being significantly higher than those for the other three approaches (P < .001). On average, reader 2 had a significantly higher accuracy than reader 1 (AUC of 0.75 vs 0.71, P < .001). Interreader agreement was fair for the four methods. Values for assessment of interobserver variability for the T2, single-voxel, multivoxel, and integrated approaches were 0.39, 0.39, 0.34, and 0.48, respectively.
View larger version (32K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3: Flowchart summarizes study results. R1 = reader 1, R2 = reader 2. In bottom three rows of boxes, numbers separated by slash marks indicate numbers of voxels assigned scores of 1, 2, 3, 4, and 5, respectively, by the designated reader.
|
|
View this table:
[in this window]
[in a new window]
|
Diagnostic Efficacy Measures for Four Interpretative Approaches to MR and MR Spectroscopic Imaging of Prostatic Peripheral Zone
|
|
|
DISCUSSION
|
|---|
Our study results have three major implications. First, they demonstrate that MR spectroscopic imaging is of incremental benefit to standard T2-weighted MR imaging for the characterization of individual voxels in the peripheral zone of the prostate as benign or malignant. An increase in accuracy of approximately 10% was seen when spectroscopic data were integrated with the information provided by T2-weighted images alone. This benefit was remarkably similar for both readers, who generally showed comparable accuracies for the four interpretative approaches studied and also demonstrated reasonable interobserver agreement, suggesting that our results are reproducible between observers. Although it is perhaps intuitively obvious that the combination of MR spectroscopic imaging and MR imaging would be better than either approach alone, we are unaware of any studies that have documented this benefit. Our isolation of T2, single-voxel, multivoxel, and integrated data sets allowed us to study the accuracy of each approach without potential confounding effects; for example, MR spectra are rarely presented without associated T2-weighted images to readers in the research or clinical setting, so the true value of MR spectroscopy is difficult to judge.
Second, our study answers the question as to whether MR spectra should be interpreted in isolation, in the context of adjacent spectra, or in combination with T2-weighted images—clearly, optimal results are obtained when MR spectra are read in an integrated fashion with T2-weighted imaging findings. Third, our results suggest that qualitative evaluation of the spectrum—that is, evaluation that is not based on the numeric values of the metabolic ratios—yields valid and reproducible results.
Although automated metabolic ratios were generated and were available to readers, these were often meaningless because of, for example, incorrect integration. In such cases, our approach was to read through erroneous numbers rather than to go back and perfect them all in the three-dimensional spectral array. The peaks for choline (3.2 ppm), creatine (3.0 ppm), and polyamines (3.1 ppm) overlap in regions of healthy prostate tissue because of the presence of high levels of the polyamine spermine in healthy glandular prostate tissue. However, in patients with cancer, the spectroscopic peak representing polyamines is markedly decreased or absent, while choline is increased (citrate is also reduced but is located at 2.6 ppm and can be easily seen). Although the loss of polyamines cannot be quantified because of overlap with choline and creatine at 1.5 T, it can be observed as increased spectral resolution of the choline from creatine resonances in cancer. Choline and creatine in regions of cancer are further resolved when 3-T MR imaging units are used, and this provides one of the motivations for using 3-T MR imaging units for MR spectroscopic imaging of prostate cancer. The readers in our study qualitatively used the increased resolution of choline and creatine as a metabolic indication of cancer. This approach to metabolically identifying prostate cancer in the peripheral zone has been previously reported (13).
The results of our study concur with those of prior studies in showing improved accuracy with the addition of the metabolic data from MR spectroscopic imaging to the information obtained at MR imaging (3–5,10,12,20,21). More recently, initial results from the American College of Radiology Imaging Network, or ACRIN, trial of MR spectroscopic imaging of the prostate (ACRIN 6659) have become available, and no incremental benefit of MR spectroscopic imaging was seen for sextant localization of prostate cancer when compared with MR imaging alone (22). Although the exact basis for the discrepancy between the results of this multi-institution trial and those of our own single-institution study are open to debate, a number of possibilities exist. The primary factor is probably study design. We selected voxels from definitely benign or malignant areas of peripheral zone tissue, and only these preselected voxels were analyzed. This somewhat painstaking preselection method, while arguably not conforming to that used in the clinical setting, may have allowed us to more precisely tease out and study the incremental benefit of different interpretative approaches to MR spectroscopic imaging. Also, the ACRIN trial included many centers and some readers with only limited MR spectroscopic experience and the use of a per-sextant rather than per-nodule method of analysis.
Our study had limitations. First, the preselection of voxels allowed us to isolate and analyze the accuracy of different interpretative approaches. However, our AUC values should not be considered as generalizable to the clinical setting for purposes of tissue characterization. Preselection inevitably includes areas of more clearly benign or malignant tissue and likely produces accuracy values that are higher than those that would be obtained across the prostate as a whole. However, the relative accuracies of the different approaches should still be valid, even if the absolute accuracies are not attainable in daily practice. Second, this was a retrospective study, with all the limitations that are inherent to that type of design. The population included in this study was a highly selected population and included only patients who had undergone endorectal MR and MR spectroscopic imaging and radical prostatectomy.
Third, each patient who was part of this study provided more than one benign or malignant voxel for interpretation. We used the generalized estimating equation and random-effects model approaches to control for correlation between voxels from the same patients and for interpretations of the same voxels by multiple readers with multiple reading approaches. Our statistical approach corrected for the possible bias due to correlated voxels. A stronger research design would have included patients rather than voxels as the study unit. Such a study design would have required a much larger number of patients, which was not possible for us. Furthermore, the goal of our study was not to identify cancer in individual patients by using different approaches but rather to compare the ability of different approaches in the discrimination of malignant voxels from normal voxels. Fourth, because images were manipulated prior to interpretation, readers could not use a standard PACS workstation to review them. Instead, we used PowerPoint and a personal computer to present the images that were created. All final images were JPEG images with 72 dots per inch, which inherently have less resolution than Digital Imaging and Communications in Medicine, or DICOM, images. This may have reduced the spectral quality. However, even if this effect was important, the direction of the bias would be detrimental to the true accuracy of MR spectroscopic imaging. It is important to note that using image manipulation was the only possible way of isolating each possible interpretative approach. Our PACS workstations do not have a feature that allows readers to review a single voxel, for instance. We believe this holds true elsewhere.
As endorectal MR and MR spectroscopic imaging are becoming more widely available, it is important to determine how these new diagnostic tools should be used. On the basis of the results of our study, we conclude that the addition of MR spectroscopic imaging to MR imaging significantly improves the characterization of peripheral zone prostate tissue as benign or malignant; improved performance is obtained when both data sets are interpreted in an integrated fashion.
|
ADVANCE IN KNOWLEDGE
|
|---|
- The addition of endorectal MR spectroscopic imaging to standard T2-weighted MR imaging significantly improves accuracy in the distinction of benign from malignant tissue in the peripheral zone of the prostate (P < .001).
|
IMPLICATION FOR PATIENT CARE
|
|---|
- MR spectroscopic imaging is of incremental benefit to standard T2-weighted imaging for the characterization of individual voxels in the peripheral zone of the prostate as benign or malignant.
|
FOOTNOTES
|
|---|
Abbreviations: AUC = area under the ROC curve • JPEG = Joint Photographic Experts Group • PACS = picture archiving and communication system • ROC = receiver operating characteristic
Author contributions: Guarantors of integrity of entire study, A.C.W., F.V.C., J.K.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, A.C.W., F.V.C., B.M.Y., J.K.; clinical studies, A.C.W., M.S., J.P.S., P.R.C., J.K.; experimental studies, F.V.C., A.Q.; statistical analysis, Y.L., S.Z.; and manuscript editing, A.C.W., F.V.C., A.Q., M.S., J.P.S., Y.L., P.R.C., B.M.Y., J.K.
Authors stated no financial relationship to disclose.
|
References
|
|---|
- Huzjan R, Sala E, Hricak H. Magnetic resonance imaging and magnetic resonance spectroscopic imaging of prostate cancer. Nat Clin Pract Urol 2005;2:434–442. [Medline]
- Coakley FV, Qayyum A, Kurhanewicz J. Magnetic resonance imaging and spectroscopic imaging of prostate cancer. J Urol 2003;170:S69–S75. [CrossRef][Medline]
- 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. [Abstract/Free Full Text]
- Scheidler J, Hricak H, Vigneron DB, et al. Prostate cancer: localization with three-dimensional proton MR spectroscopic imaging—clinicopathologic study. Radiology 1999;213:473–480. [Abstract/Free Full Text]
- Yu KK, Scheidler J, Hricak H, et al. Prostate cancer: prediction of extracapsular extension with endorectal MR imaging and three-dimensional proton MR spectroscopic imaging. Radiology 1999;213:481–488. [Abstract/Free Full Text]
- Clarke DH, Banks SJ, Wiederhorn AR, et al. The role of endorectal coil MRI in patient selection and treatment planning for prostate seed implants. Int J Radiat Oncol Biol Phys 2002;52:903–910. [CrossRef][Medline]
- Hricak H, Wang L, Wei DC, et al. The role of preoperative endorectal magnetic resonance imaging in the decision regarding whether to preserve or resect neurovascular bundles during radical retropubic prostatectomy. Cancer 2004;100:2655–2663. [CrossRef][Medline]
- Wang L, Mullerad M, Chen HN, et al. Prostate cancer: incremental value of endorectal MR imaging findings for prediction of extracapsular extension. Radiology 2004;232:133–139. [Abstract/Free Full Text]
- Kurhanewicz J, Vigneron DB, Nelson SJ. Three-dimensional magnetic resonance spectroscopic imaging of brain and prostate cancer. Neoplasia 2000;2:166–189. [CrossRef][Medline]
- Coakley FV, Teh HS, Qayyum A, et al. Endorectal MR imaging and MR spectroscopic imaging for locally recurrent prostate cancer after external beam radiation therapy: preliminary experience. Radiology 2004;233:441–448. [Abstract/Free Full Text]
- Zakian KL, Sircar K, Hricak H, et al. Correlation of proton MR spectroscopic imaging with Gleason score based on step-section pathologic analysis after radical prostatectomy. Radiology 2005;234:804–814. [Abstract/Free Full Text]
- Kurhanewicz J, Vigneron DB, Hricak H, Narayan P, Carroll P, Nelson SJ. Three-dimensional H-1 MR spectroscopic imaging of the in situ human prostate with high (0.24-0.7-cm3) spatial resolution. Radiology 1996;198:795–805. [Abstract/Free Full Text]
- 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. [Abstract/Free Full Text]
- Star-Lack J, Vigneron DB, Pauly J, Kurhanewicz J, Nelson SJ. Improved solvent suppression and increased spatial excitation bandwidths for three-dimensional PRESS CSI using phase-compensating spectral/spatial spin-echo pulses. J Magn Reson Imaging 1997;7:745–757. [Medline]
- 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. [CrossRef][Medline]
- 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. [CrossRef][Medline]
- Williamson JM, Lipsitz SR, Manatunga AK. Modeling kappa for measuring dependent categorical agreement data. Biostatistics 2000;1:191–202. [Abstract]
- Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–174. [CrossRef][Medline]
- Dorfman DD, Berbaum KS, Metz CE. Receiver operating characteristic rating analysis: generalization to the population of readers and patients with the jackknife method. Invest Radiol 1992;27:723–731. [CrossRef][Medline]
- Menard C, Smith IC, Somorjai RL, et al. Magnetic resonance spectroscopy of the malignant prostate gland after radiotherapy: a histopathologic study of diagnostic validity. Int J Radiat Oncol Biol Phys 2001;50:317–323. [CrossRef][Medline]
- Mueller-Lisse UG, Vigneron DB, Hricak H, et al. Localized prostate cancer: effect of hormone deprivation therapy measured by using combined three-dimensional 1H MR spectroscopy and MR imaging: clinicopathologic case-controlled study. Radiology 2001;221:380–390. [Abstract/Free Full Text]
- Weinreb J, Coakley FV, Blume J, Wheeler T, Cormack J, Kurhanewicz J. Hot topic: ACRIN 6659 MRI and MRSI of prostate cancer prior to radical prostatectomy: a prospective multi-institutional clinicopathologic study. Presented at the 92nd Scientific Assembly and Annual Meeting of the Radiological Society of North America, Chicago, November 26–December 1, 2006.
TOP
ABSTRACT
INTRODUCTION
