Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print September 21, 2007, 10.1148/radiol.2452062201
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2452062201v1
245/2/499    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Shukla-Dave, A.
Right arrow Articles by Zakian, K. L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Shukla-Dave, A.
Right arrow Articles by Zakian, K. L.
(Radiology 2007;245:499-506.)
© RSNA, 2007

Genitourinary Imaging

Detection of Prostate Cancer with MR Spectroscopic Imaging: An Expanded Paradigm Incorporating Polyamines1

Amita Shukla-Dave, PhD, Hedvig Hricak, MD, PhD, Chaya Moskowitz, PhD, Nicole Ishill, MS, Oguz Akin, MD, Kentaro Kuroiwa, MD, Jessica Spector, BA, Mahesh Kumar, PhD, Victor E. Reuter, MD, Jason A. Koutcher, MD, PhD, and Kristen L. Zakian, PhD

1 From the Departments of Medical Physics (A.S., J.S., M.K., J.A.K., K.L.Z.), Radiology (A.S., H.H., O.A., J.A.K., K.L.Z.), Epidemiology and Biostatistics (C.M., N.I.), Urology (K.K.), Pathology (V.E.R.), and Medicine (J.A.K.), Memorial Sloan-Kettering Cancer Center, 1275 York Ave, New York, NY 10021; and Department of Urology, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan (K.K.). Received December 29, 2006; revision requested February 20, 2007; revision received March 28; final version accepted April 20. Supported by National Institutes of Health grant R01 CA76423. Address correspondence to A.S. (e-mail: davea{at}mskcc.org).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 
Purpose: To characterize benign and malignant prostate peripheral zone (PZ) tissue retrospectively by using a commercial magnetic resonance (MR) spectroscopic imaging package and incorporating the choline plus creatine–to-citrate ratio ([Cho + Cr]/Cit) and polyamine (PA) information into a statistically based voxel classification procedure.

Materials and Methods: The institutional review board approved this HIPAA-compliant study and waived the requirement for informed consent. Fifty men (median age, 60 years; range, 44–69 years) with untreated biopsy-proved prostate cancer underwent combined endorectal MR imaging and MR spectroscopic imaging. Commercial software was used to acquire and process MR spectroscopic imaging data. The (Cho + Cr)/Cit and the PA level were tabulated for each voxel. The PA level was scored on a scale of 0 (PA undetectable) to 2 (PA peak as high as or higher than Cho peak). Whole-mount step-section histopathologic analysis constituted the reference standard. Classification and regression tree analysis in a training set generated a decision-making tree (rule) for classifying voxels as malignant or benign, which was validated in a test set. Receiver operating characteristic and generalized estimating equation regression analyses were used to assess accuracy and sensitivity, respectively.

Results: The median (Cho + Cr)/Cit was 0.55 (mean ± standard deviation, 0.59 ± 0.03) in benign and 0.77 (mean, 1.08 ± 0.20) in malignant PZ voxels (P = .027). A significantly higher percentage of benign (compared with malignant) voxels had higher PA than choline peaks (P < .001). In the 24-patient training set (584 voxels), the rule yielded 54% sensitivity and 91% specificity for cancer detection; in the 26-patient test set (667 voxels), it yielded 42% sensitivity and 85% specificity. The percentage of cancer in the voxel at histopathologic analysis correlated positively (P < .001) with the sensitivity of the classification and regression tree rule, which was 75% in voxels with more than 90% malignancy.

Conclusion: The statistically based classification rule developed indicated that PAs have an important role in the detection of PZ prostate cancer. With commercial software, this method can be applied in clinical settings.

© RSNA, 2007


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 
The use of endorectal magnetic resonance (MR) imaging for the detection and staging of prostate cancer has grown steadily in the past decade, and MR spectroscopic imaging has been shown to contribute significant incremental value to MR imaging in the detection and localization of prostate cancer (P < .001) (1). With the release of commercial spectroscopic imaging software, more centers will be able to use this technology for metabolic characterization of the prostate gland, and it will be important to develop standardized criteria for the interpretation of proton MR spectra.

While the choline (Cho) plus creatine (Cr)–to-citrate (Cit) ratio ([Cho + Cr]/Cit) has been widely studied and cutoff values have been suggested for detection of cancer on a voxel-by-voxel basis (2–4), the use of newer acquisition and processing software has also enabled the routine assessment of polyamines (PAs) (predominantly spermine). Several studies have been done to investigate the role of PAs in cellular growth and differentiation in prostate cancer (5–10). In vitro MR spectroscopic studies have revealed that in normal and benign hyperplastic tissue, a high content of spermine is present, whereas spermine levels in malignant tumor are reduced (11–14). The incorporation of PA information into human prostate spectral interpretation has been reported (2). Thus, the purpose of our study was to characterize benign and malignant prostate peripheral zone (PZ) tissue retrospectively by using a commercial MR spectroscopic imaging package and incorporating the (Cho + Cr)/Cit and PA information into a statistically based voxel classification procedure.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 
Patients
The institutional review board of Memorial Sloan-Kettering Cancer Center approved and issued a waiver of informed consent for our study, which was compliant with the Health Insurance Portability and Accountability Act. We identified 50 consecutive patients who, from August 2003 and January 2004, underwent combined preoperative endorectal MR imaging and proton MR spectroscopic imaging performed with commercially available acquisition and processing software (PROSE; GE Medical Systems, Milwaukee, Wis) and for whom whole-mount step-section histopathologic maps were available for comparison with imaging findings. All patients had biopsy-proved prostate cancer (median biopsy Gleason score, 6; range, 6–9) that was untreated. The median patient age was 60 years (range, 44–69 years), and the median baseline serum prostate-specific antigen level was 5.4 ng/mL (range, 0.6–24.1 ng/mL). All patients underwent surgery after MR imaging and MR spectroscopic imaging, with a mean interval between imaging and radical prostatectomy of 31 days (range, 2–112 days) (Table 1).


View this table:
[in this window]
[in a new window]

  		Table 1. Clinical Data for the 50 Patients Included in the Study

 
MR Imaging and MR Spectroscopic Imaging Data Acquisition and Processing
MR imaging and MR spectroscopic imaging examinations were performed on a 1.5-T whole-body MR imager (GE Medical Systems) with a pelvic phased-array coil and an endorectal coil for signal reception. Patients were imaged with standard MR imaging sequences (15–18), which included the acquisition of transverse T1-weighted spin-echo MR images from the aortic bifurcation to the symphysis pubis (700/12 [repetition time msec/echo time msec]; section thickness, 5 mm; intersection gap, 1 mm; matrix, 256 x 192; two signals acquired; field of view, 24 cm) and thin-section high-spatial-resolution transverse, coronal, and sagittal T2-weighted fast spin-echo images of the prostate and seminal vesicles (4000/102; section thickness, 3 mm; no intersection gap; matrix, 256 x 192; four signals acquired). Automated correction was applied to the T1- and T2-weighted images for the reception profile of the endorectal and pelvic phased-array coils.

Transverse T2-weighted images were used for MR spectroscopic imaging volume selection, which was performed by means of the point-resolved spectroscopy (PRESS) voxel excitation technique (3,19), with an in-plane spatial resolution of 6.9 mm. The PRESS box was positioned to maximize coverage of the prostate while minimizing inclusion of periprostatic fat. Spectral-spatial pulses were used for water and lipid suppression within the PRESS-selected volume (4,20), and very selective outer voxel suppression pulses were used to reduce the contamination from surrounding tissues. The MR imaging–MR spectroscopic imaging examination took 55–60 minutes.

The MR spectroscopic imaging data were processed with the Functool package on the Advantage workstation (GE Medical Systems), which aligns the spectral data with the MR images and archives arrays of spectral data with the corresponding images in Digital Imaging and Communications in Medicine format. The MR spectroscopic imaging data were zero filled to the nearest power of two in the superior-inferior dimension and zero filled once in the spectral dimension. The time-spectral dimension was apodized with a 4-Hz shifted Gaussian function. Fourier transformations were performed in the temporal and spatial domains with automated baseline correction and frequency alignment.

The PROSE software integrates the signal over designated frequency ranges to provide estimates of metabolite peak areas and calculates metabolite ratios. Manual adjustments of the integration range centers were made if voxel-by-voxel frequency shifts were observed owing to variations in the main magnetic field strength. For all voxels, the (Cho + Cr)/Cit was calculated. The PROSE software permits the detection of a PA resonance (at 3.1 ppm), which resides between the Cr (3.0 ppm) and Cho (3.2 ppm) peaks. While the height of the PA peak may be judged qualitatively and reported relative to the Cho peak, the peak cannot always be completely differentiated from the Cho and Cr peaks and a distinct integration region cannot be designated for PA. Therefore, both the Cho and Cr integration ranges may contain some PA contribution, and the (Cho + Cr)/Cit reported in our study may be interpreted as (Cho + PA + Cr)/Cit, as has been described elsewhere (21).

Histopathologic Analysis and Comparison with MR Spectroscopic Imaging
Prostatectomy specimen whole-mount preparation (22) consisted of surface inking with tattoo dye followed by fixation in 10% formalin. The distal 5-mm portion of the apex was amputated and coned. The remainder of the gland was serially sectioned from the apex to the base at 3–4-mm intervals and submitted in its entirety for paraffin-embedded whole mounts. A pathologic Gleason grade was assigned to the whole cancer lesion (by K.K., who had more than 5 years of experience in uropathology). Cancer foci were outlined in ink on whole-mount, apical, and seminal vesicle sections and photographed. The photographs constituted the tumor maps.

A radiologist (O.A.) with more than 5 years of experience in prostate imaging matched the histopathologic step sections with the most closely corresponding T2-weighted transverse MR images. Because the spectroscopy data were acquired in the same position and with the same gradients as the imaging data, registration of the spectroscopic data with the T2-weighted images was automatic, and the spectroscopic data could be compared with the most closely corresponding histopathologic step section. The radiologist outlined the tumor on the registered MR and MR spectroscopic images and designated the voxels as malignant or benign by using the histopathologic maps. Using the tumor outline drawn by the radiologist (O.A.), the spectroscopist (A.S.) estimated and tabulated the percentage of cancer in each voxel.

Analysis of MR Spectroscopic Imaging Data
MR spectroscopic imaging data from the 50 patients were analyzed by a team of two spectroscopists with 10 years (K.L.Z.) and 5 years (A.S.) of experience in reading prostate spectroscopy data. The spectroscopists worked together by consensus. Of the total of 1839 PZ voxels evaluated, 273 (15%) were unusable owing to artifact arising from lipid contamination in the excitation volume and 315 (17%) were nondiagnostic (ie, had signal-to-noise ratios for Cho and Cit peaks of <5.0). In the remaining 1251 usable voxels, the phase and chemical shift (in parts per million) of the metabolites were checked, the (Cho + Cr)/Cit was recorded, and the PA peak was inspected. The PA peak was scored on a scale of 0–2 as follows: 0 indicated undetectable PA; 1, PA peak lower than Cho peak; and 2, PA peak as high as or higher than Cho peak (Fig 1). For all voxels, the (Cho + Cr)/Cit, PA score, benign or malignant classification based on whole-mount step-section histopathologic analysis results, and percentage of cancer in the voxel were tabulated for statistical analysis.


Figure 1A
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1a: Representative spectra for PA scores of (a) 0 (PA undetectable), (b) 1 (PA peak lower than Cho peak), and (c) 2 (PA peak as high as or higher than Cho peak).

 

Figure 1B
View larger version (10K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1b: Representative spectra for PA scores of (a) 0 (PA undetectable), (b) 1 (PA peak lower than Cho peak), and (c) 2 (PA peak as high as or higher than Cho peak).

 

Figure 1C
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1c: Representative spectra for PA scores of (a) 0 (PA undetectable), (b) 1 (PA peak lower than Cho peak), and (c) 2 (PA peak as high as or higher than Cho peak).

 
Statistical Analysis
To determine whether the PA score was associated with a benign or malignant tissue designation, Pearson {chi}2 statistics were calculated, with adjustment for correlation due to multiple observations per patient performed by using a second-order correction (23). The accuracy of the (Cho + Cr)/Cit alone and that of the PA score alone for the detection of cancer were individually assessed by means of receiver operating characteristic (ROC) analysis, with the voxel serving as the unit of analysis. The area under the ROC curve (AUC) and corresponding confidence intervals (CIs) were estimated to account for correlation due to multiple observations per patient (24). The CIs for proportions, such as estimates of sensitivity and specificity, were calculated (23).

To generate a classification rule that combined the (Cho + Cr)/Cit and the PA score, the classification and regression tree (CART) analysis method was used (25). Patient data were randomly divided into a training set of 24 patients (584 voxels) and a test set of 26 patients (667 voxels). The training set was used to create a classification rule (tree), and the test set was used for validating the rule. When nondiagnostic and artifact voxels were excluded, the training set contained 349 benign and 235 malignant voxels, and the test set contained 452 benign and 215 malignant voxels.

The models were fit with the method of recursive partitioning, in which the data were repeatedly split into binary subgroups until either all subgroups were completely homogeneous or there were too few observations left to split further. If necessary, the tree was simplified by removing branches that did not contribute substantial information to the classification rule. To determine the branching criteria, the algorithm aimed to partition the data by choosing the optimal split at each node, with "optimal" defined in terms of the deviance (likelihood ratio statistic). Thus, in each node all possible splits were examined, and the split that yielded the maximum reduction in the deviance for the tree was chosen (26). To assess whether the (Cho + Cr)/Cit was associated with malignancy and whether the percentage of cancer in a voxel affected the sensitivity of the CART rule for cancer detection, generalized estimating equation regression analysis was performed.

In our study, the PA peak was not separated from the Cho peak, so there was an interrelation. This interrelation, called collinearity in the statistical literature, can make statistical inference difficult in some analyses; however, previous work (27,28) has shown that collinearity does not substantially change analysis results when the goal is to make predictions. For example, in our study, we attempted to predict whether or not voxels were cancerous. In this setting, as long as one uses the same set of variables on data that have the same degree of collinearity when making future predictions, the collinearity should not affect the results (27,28). Because we expected the same collinearity in our testing and training sets and further expected a similar degree of collinearity in future practical applications, the collinearity should not be a problem.

P ≤ .05 was considered to indicate a significant difference. Analyses were performed with the following software: Intercooled Stata (version 8.0 for Windows, 2003; Stata, College Station, Tex), S-PLUS for Windows (version 6.2.1, 2003; Insightful, Seattle, Wash), and SAS (version 9.0 for Windows, 2002; SAS Institute, Cary, NC).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 
Of the 1251 usable PZ voxels analyzed, 801 (64%) were designated benign at histopathologic analysis, and 450 (36%) were designated malignant.

PA and Cho Peaks
In the majority of benign diagnostic voxels (90%), the PA peak was higher than the Cho peak (Table 2). Ten percent of benign voxels had PA peaks lower than Cho peaks, and less than 1% had no detectable PA. The proportion of malignant voxels with PA peaks higher than Cho peaks was substantial (56%); however, the proportions with PA peaks lower than Cho peaks (33%) or undetectable PA (11%) were significantly greater than those in benign voxels. The difference in the distribution of the PA metabolite between benign and malignant voxels was significant (P < .001). As expected, the (Cho + Cr)/Cit was greater in voxels containing cancer (median, 0.77; range, 0.14–8.0) than in benign voxels (median, 0.55; range, 0.13–2.6) (P = .027). As the height of the PA peak increased, the (Cho + Cr)/Cit decreased (Fig 2, Table 2). Since PA is implicit in the numerator of this ratio, the changes in Cho and Cit appear to outweigh the effect of PA in the ratio.


View this table:
[in this window]
[in a new window]

  		Table 2. MR Spectroscopic Imaging PA and (Cho+Cr)/Cit Data from 1251 Diagnostic Voxels Segregated according to Benign or Malignant Histopathologic Designation

 

Figure 2A
View larger version (45K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2a: MR imaging and MR spectroscopic imaging data and histopathologic whole-mount step-section tumor map from a patient with prostate cancer (biopsy Gleason score, 6). (a) MR spectroscopic imaging grid superimposed on transverse T2-weighted MR image. MR imaging parameters were as follows: 4000/102; echo train length, 12; field of view, 14 cm; acquisition matrix, 256 x 192; section thickness, 3 mm; no intersection gap; and four signals acquired. MR spectroscopic imaging parameters were as follows: volume excitation with water and lipid suppression by means of spectral-spatial pulses; 1000/130; chemical shift imaging matrix, 16 x 8 x 8; field of view, 110 x 55 x 55 mm; spatial resolution, 6.9 mm; one signal acquired; and imaging time, 17 minutes. (b) Corresponding histopathologic section shows a tumor with a surgical Gleason score of 7 (outlined by thick solid line) and the PZ (outlined by dashed line). (c) Tabulated metabolite ratios, PA scores, benign or malignant designations, and percentages of cancer for selected PZ voxels in a single MR spectroscopic imaging section.

 

Figure 2B
View larger version (82K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2b: MR imaging and MR spectroscopic imaging data and histopathologic whole-mount step-section tumor map from a patient with prostate cancer (biopsy Gleason score, 6). (a) MR spectroscopic imaging grid superimposed on transverse T2-weighted MR image. MR imaging parameters were as follows: 4000/102; echo train length, 12; field of view, 14 cm; acquisition matrix, 256 x 192; section thickness, 3 mm; no intersection gap; and four signals acquired. MR spectroscopic imaging parameters were as follows: volume excitation with water and lipid suppression by means of spectral-spatial pulses; 1000/130; chemical shift imaging matrix, 16 x 8 x 8; field of view, 110 x 55 x 55 mm; spatial resolution, 6.9 mm; one signal acquired; and imaging time, 17 minutes. (b) Corresponding histopathologic section shows a tumor with a surgical Gleason score of 7 (outlined by thick solid line) and the PZ (outlined by dashed line). (c) Tabulated metabolite ratios, PA scores, benign or malignant designations, and percentages of cancer for selected PZ voxels in a single MR spectroscopic imaging section.

 

Figure 2C
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2c: MR imaging and MR spectroscopic imaging data and histopathologic whole-mount step-section tumor map from a patient with prostate cancer (biopsy Gleason score, 6). (a) MR spectroscopic imaging grid superimposed on transverse T2-weighted MR image. MR imaging parameters were as follows: 4000/102; echo train length, 12; field of view, 14 cm; acquisition matrix, 256 x 192; section thickness, 3 mm; no intersection gap; and four signals acquired. MR spectroscopic imaging parameters were as follows: volume excitation with water and lipid suppression by means of spectral-spatial pulses; 1000/130; chemical shift imaging matrix, 16 x 8 x 8; field of view, 110 x 55 x 55 mm; spatial resolution, 6.9 mm; one signal acquired; and imaging time, 17 minutes. (b) Corresponding histopathologic section shows a tumor with a surgical Gleason score of 7 (outlined by thick solid line) and the PZ (outlined by dashed line). (c) Tabulated metabolite ratios, PA scores, benign or malignant designations, and percentages of cancer for selected PZ voxels in a single MR spectroscopic imaging section.

 
In detecting cancer in the training set, the AUC for the (Cho + Cr)/Cit alone was 0.72 (95% CI: 0.59, 0.85), while that for the PA score alone was 0.71 (95% CI: 0.61, 0.81) (Fig 3a). In the test set, the AUC for the (Cho + Cr)/Cit alone was 0.71, while that for the PA score alone was 0.64 (Fig 3b). Differences in AUCs were not significant in either the training or the test set.


Figure 3A
View larger version (13K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3a: (a) Training set and (b) test set ROC curves for the detection of cancer on a voxel-by-voxel basis with use of (Cho + Cr)/Cit (CC/C) and PA score.

 

Figure 3B
View larger version (13K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3b: (a) Training set and (b) test set ROC curves for the detection of cancer on a voxel-by-voxel basis with use of (Cho + Cr)/Cit (CC/C) and PA score.

 
In the decision tree for classifying voxels as benign or malignant (Fig 4), the tree voxels are first split according to their PA scores. If the PA peak is lower than the Cho peak or is undetectable, the voxel is determined to be malignant. If the PA peak is higher than the Cho peak, the voxel is considered benign if the (Cho + Cr)/Cit is less than 1.1 and malignant if the (Cho + Cr)/Cit is greater than or equal to 1.1. The first branching of the tree is based on the PA score, indicating that this factor was determined with the CART analysis to have greater importance than the (Cho + Cr)/Cit. For the training set, the tree yielded a sensitivity of 54% (127 of 235 voxels) (95% CI: 33%, 73%) and a specificity of 91% (316 of 349 voxels) (95% CI: 82%, 95%). In the test set, the rule yielded a sensitivity of 42% (91 of 215 voxels) (95% CI: 27%, 59%) and a specificity of 85% (383 of 452 voxels) (95% CI: 70%, 93%).


Figure 4
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4: CART-based decision-making tree for voxel-by-voxel analysis of MR spectroscopic imaging data.

 
The percentage of cancer in the voxel correlated positively with the sensitivity of the CART rule (P < .001) (Table 3). In voxels with more than 90% cancer content, the sensitivity for cancer detection was 75%. The most notable exception was the 61%–70% decile, where the low number of voxels (14) may have resulted in the aberrant low sensitivity.


View this table:
[in this window]
[in a new window]

  		Table 3. Effect of Percentage of Cancer in Histopathologically Malignant Voxels on Cancer Detection Sensitivity Derived by Using the CART Rule

 
CART Analysis
CART analysis is used to predict a binary outcome (benign or malignant) that cannot be directly characterized by means of ROC analysis, which requires a continuous or ordinal value. To compare the CART rule with the use of the (Cho + Cr)/Cit in light of this difficulty, we applied the method suggested by Beam and Wieand (29). The specificity of the CART rule may be used as a starting point. The point on the ROC curve corresponding to the same specificity for the (Cho + Cr)/Cit is fixed, and the corresponding sensitivity is compared with the CART rule sensitivity. For example, CART rule analysis of the test set yielded a sensitivity of 42% and a specificity of 85%. On the (Cho + Cr)/Cit ROC curve, the point corresponding to a specificity of 85% gives a sensitivity of 27%, which is lower than the corresponding CART rule sensitivity. Therefore, for the same specificity, the CART rule had a sensitivity 15% higher than that of the (Cho + Cr)/Cit alone.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 
In our study, the decision-making rule had high specificity and low sensitivity for the detection of prostate cancer—a trend that has been found in other MR spectroscopic imaging studies (2,30). The high specificity indicates that the rule is very good at identifying benign voxels; that is, if the (Cho + Cr)/Cit is low and the PA peak is higher than the Cho peak, the voxel is very likely benign. The low sensitivity may be due to several factors: First, to reflect the typical clinical scenario, voxels with small percentages of cancer at histopathologic analysis were included. Second, volume averaging of malignant and benign tissue dilutes the metabolic abnormality. The sensitivity of the decision tree increased as the percentage of cancer in the voxel increased. Third, the majority of the patients in the study had low clinical stages and low Gleason scores. It has been reported that the sensitivity of MR spectroscopic imaging in cancer detection increases with tumor grade (30). Furthermore, MR imaging findings, which might have increased the sensitivity, were not included in the analysis (1).

The accuracy of the (Cho + Cr)/Cit for cancer detection and that of the PA score alone were approximately the same. For a fixed specificity level, the CART rule had higher sensitivity than either the (Cho + Cr)/Cit or the PA score alone, justifying the use of the rule. CART analysis revealed that the level of PA had a greater effect on the detection of cancer than did the (Cho + Cr)/Cit. Our results strongly suggest that PA information should be considered in MR spectroscopic imaging voxel characterization in the PZ of the prostate gland.

An early in vitro study revealed higher PA levels in the spectra of normal PZ extracts than in benign prostatic hyperplasia and cancer (12). It has been observed that increased intracellular spermine has been associated with well-differentiated nonproliferative cells and that because spermine is also a secretory product in the prostate, the spermine level would decrease if the ductal volume were reduced (eg, if tissue morphology changed because of tumor) (14). MR spectroscopy of tissue samples revealed that levels of putrescine and spermine were reduced in prostate cancer compared with the levels in benign tissue (14). High-resolution magic angle spinning hydrogen 1 MR spectroscopic examinations of intact human prostate tissue samples have revealed that the amount of spermine in tissue correlates positively with the volume percentage of normal prostatic epithelial cells (11).

PAs in the prostate gland were detected in vivo by means of single-voxel oversampled J-resolved spectroscopy at 1.5 T (31). Later, Jung et al (2) incorporated the (Cho + Cr)/Cit and the PA level into a scoring system for cancer detection, resulting in specificities of 89.3% and 84.6% and sensitivities of 69% and 64% for two readers. Jung et al used only high-quality data and included voxels with clear-cut concordance between imaging and histopathologic findings (ie, a high percentage of cancer in the voxel) in the analysis, possibly resulting in higher sensitivity. Our findings, obtained by using a larger population of consecutive patients, commercially available software, and a statistically based decision-making rule, reinforce these results. The use of consecutive patients results in a wider range of data quality that may more accurately reflect day-to-day clinical examinations.

Use of the (Cho + Cr)/Cit in prostate cancer detection is well known (2–4,30,32–35) and is based on the elevation of Cho-containing compounds and the reduction of Cit in prostate cancer relative to normal prostatic tissue. The mean (Cho + Cr)/Cit for benign tissue in this study (0.59 ± 0.03) is greater than the value previously published for spectral-spatial MR spectroscopic imaging acquisition (0.31 ± 0.17) (4). A possible reason for this discrepancy is differences in the software packages used for data acquisition and processing.

A limitation of our study is that PA levels were assessed qualitatively. At 1.5 T, differentiation of Cho from PA is inherently difficult because of the proximity of the peaks and the limited spectral resolution available within a reasonable imaging time; thus, PA could not be integrated and quantified. As discussed in the Materials and Methods, we did not expect the lack of separation of Cho and PA to influence the prediction of cancer. Higher magnetic field strengths will yield better spectral resolution with MR spectroscopic imaging (36–38). A recent examination performed at 3 T (36) revealed that Cho could be better distinguished from PA owing to increased spectral dispersion, although complete differentiation of PA from Cr was still difficult to achieve. Another limitation of our study is that MR imaging findings were not included. We believe the model developed in our study may serve as the foundation for future prospective studies that incorporate MR imaging.

In conclusion, PAs have an important role in determining which voxels contain cancer and which are benign. The statistically based classification rule developed in our study is simple to apply for the detection of prostate cancer on a voxel-by-voxel basis and is applicable to consecutive data sets obtained with a commercially available software package. If institutions apply this rule, multicenter comparative studies may be facilitated.


   ADVANCES IN KNOWLEDGE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 


   IMPLICATION FOR PATIENT CARE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
References
 


   ACKNOWLEDGMENTS
 
The authors thank Ada Muellner, BA, for editing the manuscript.


   FOOTNOTES
 

Abbreviations: AUC = area under the ROC curve • CART = classification and regression tree • Cho = choline • CI = confidence interval • Cit = citrate • Cr = creatine • PA = polyamine • PZ = peripheral zone • ROC = receiver operating characteristic

Guarantor of integrity of entire study, A.S.; 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.S.; clinical studies, A.S., H.H., O.A., K.K., M.K., V.E.R., K.L.Z.; experimental studies, A.S., H.H., K.L.Z.; statistical analysis, C.M., N.I.; and manuscript editing, A.S., H.H., J.A.K., K.L.Z.

Authors stated no financial relationship to disclose.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE...
 References
 

  1. 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]
  2. 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]
  3. 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]
  4. Males RG, Vigneron DB, Star-Lack J, et al. Clinical application of BASING and spectral-spatial water and lipid suppression pulses for prostate cancer staging and localization by in vivo 3D 1H magnetic resonance spectroscopic imaging. Magn Reson Med 2000;43:17–22. [CrossRef][Medline]
  5. Cohen RJ, Fujiwara K, Holland JW, McNeal JE. Polyamines in prostatic epithelial cells and adenocarcinoma: the effects of androgen blockade. Prostate 2001;49:278–284. [CrossRef][Medline]
  6. Gerner EW, Meyskens FL Jr. Polyamines and cancer: old molecules, new understanding. Nat Rev Cancer 2004;4:781–792. [CrossRef][Medline]
  7. Medina MA, Correa-Fiz F, Rodriguez-Caso C, Sanchez-Jimenez F. A comprehensive view of polyamine and histamine metabolism to the light of new technologies. J Cell Mol Med 2005;9:854–864. [Medline]
  8. Schipper RG, Romijn JC, Cuijpers VM, Verhofstad AA. Polyamines and prostatic cancer. Biochem Soc Trans 2003;31:375–380. [CrossRef][Medline]
  9. Takyi EE, Fuller DJ, Donaldson LJ, Thomas GH. Deoxyribonucleic acid and polyamine synthesis in rat ventral prostrate: effects of age of the intact rat and androgen stimulation of the castrated rat with testosterone, 5 alpha-dihydrotestosterone and 5 alpha-androstane-3 beta, 17 beta-diol. Biochem J 1977;162:87–97. [Medline]
  10. Wallace HM, Fraser AV, Hughes A. A perspective of polyamine metabolism. Biochem J 2003;376(pt 1):1–14.
  11. Cheng LL, Wu C, Smith MR, Gonzalez RG. Non-destructive quantitation of spermine in human prostate tissue samples using HRMAS 1H NMR spectroscopy at 9.4 T. FEBS Lett 2001;494:112–116. [CrossRef][Medline]
  12. Kurhanewicz J, Dahiya R, Macdonald JM, Chang LH, James TL, Narayan P. Citrate alterations in primary and metastatic human prostatic adenocarcinomas: 1H magnetic resonance spectroscopy and biochemical study. Magn Reson Med 1993;29:149–157. [Medline]
  13. Swanson MG, Vigneron DB, Tabatabai ZL, et al. Proton HR-MAS spectroscopy and quantitative pathologic analysis of MRI/3D-MRSI-targeted postsurgical prostate tissues. Magn Reson Med 2003;50:944–954. [CrossRef][Medline]
  14. van der Graaf M, Schipper RG, Oosterhof GO, Schalken JA, Verhofstad AA, Heerschap A. Proton MR spectroscopy of prostatic tissue focused on the detection of spermine, a possible biomarker of malignant behavior in prostate cancer. MAGMA 2000;10:153–159.
  15. 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]
  16. Mullerad M, Hricak H, Kuroiwa K, et al. Comparison of endorectal magnetic resonance imaging, guided prostate biopsy and digital rectal examination in the preoperative anatomical localization of prostate cancer. J Urol 2005;174:2158–2163. [CrossRef][Medline]
  17. 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]
  18. 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]
  19. Bottomley P. Selective volume method for performing localized NMR spectroscopy. U.S. patent 4,480,228. 1984.
  20. 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]
  21. Pels P, Ozturk-Isik E, Swanson MG, et al. Quantification of prostate MRSI data by model-based time domain fitting and frequency domain analysis. NMR Biomed 2006;19:188–197. [CrossRef][Medline]
  22. Aihara M, Wheeler TM, Ohori M, Scardino PT. Heterogeneity of prostate cancer in radical prostatectomy specimens. Urology 1994;43:60–66. [CrossRef][Medline]
  23. Rao JN, Scott AJ. The analysis of categorical data from complex sample surveys: chi-squared tests for goodness of fit and independence in two-way tables. J Am Stat Assoc 1981;76:221–230. [CrossRef]
  24. Obuchowski NA. Nonparametric analysis of clustered ROC curve data. Biometrics 1997;53:567–578. [CrossRef][Medline]
  25. Breiman L, Friedman J, Olshen R, Stone C. Classification and regression trees. Belmont, Calif: Wadsworth, 1984.
  26. Chambers JM, Hastie TJ. Statistical models. London, England: Chapman & Hall, 1996.
  27. Harrell FE. Regression modeling strategies. New York, NY: Springer, 2001.
  28. Myers RH. Classical and modern regression with applications. Boston, Mass: PWS-Kent, 1990.
  29. Beam CA, Wieand HS. A statistical method for the comparison of a discrete diagnostic test with several continuous diagnostic tests. Biometrics 1991;47:907–919. [CrossRef][Medline]
  30. 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]
  31. Swanson MG, Vigneron DB, Tran TK, Sailasuta N, Hurd RE, Kurhanewicz J. Single-voxel oversampled J-resolved spectroscopy of in vivo human prostate tissue. Magn Reson Med 2001;45:973–980. [CrossRef][Medline]
  32. Casciani E, Polettini E, Bertini L, et al. Prostate cancer: evaluation with endorectal MR imaging and three-dimensional proton MR spectroscopic imaging. Radiol Med (Torino) 2004;108:530–541. [Medline]
  33. Futterer JJ, Heijmink SW, Scheenen TW, et al. Prostate cancer localization with dynamic contrast-enhanced MR imaging and proton MR spectroscopic imaging. Radiology 2006;241:449–458. [Abstract/Free Full Text]
  34. Prando A, Kurhanewicz J, Borges AP, Oliveira EM Jr, Figueiredo E. Prostatic biopsy directed with endorectal MR spectroscopic imaging findings in patients with elevated prostate specific antigen levels and prior negative biopsy findings: early experience. Radiology 2005;236:903–910. [Abstract/Free Full Text]
  35. Squillaci E, Manenti G, Mancino S, et al. MR spectroscopy of prostate cancer: initial clinical experience. J Exp Clin Cancer Res 2005;24:523–530. [Medline]
  36. Chen AP, Cunningham CH, Kurhanewicz J, et al. High-resolution 3D MR spectroscopic imaging of the prostate at 3 T with the MLEV-PRESS sequence. Magn Reson Imaging 2006;24:825–832. [CrossRef][Medline]
  37. Cunningham CH, Vigneron DB, Marjanska M, et al. Sequence design for magnetic resonance spectroscopic imaging of prostate cancer at 3 T. Magn Reson Med 2005;53:1033–1039. [CrossRef][Medline]
  38. Scheenen TW, Gambarota G, Weiland E, et al. Optimal timing for in vivo 1H-MR spectroscopic imaging of the human prostate at 3T. Magn Reson Med 2005;53:1268–1274. [CrossRef][Medline]




This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2452062201v1
245/2/499    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Shukla-Dave, A.
Right arrow Articles by Zakian, K. L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Shukla-Dave, A.
Right arrow Articles by Zakian, K. L.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE