Magnetic resonance spectroscopic imaging of benign prostatic tissue: findings at 3.0 T compared to 1.5 T—initial experience

Abstract

In a retrospective study of 71 voxels of benign peripheral zone tissue from 3 men who underwent endorectal magnetic resonance (MR) spectroscopic imaging of the prostate at both 1.5 and 3 T, 21 voxels that appeared more malignant at 3 T to either of two readers demonstrated significantly higher levels of choline and polyamines at 3 T compared to 1.5 T using a Wilcoxon ranked-sum test; awareness of this selective amplification of these metabolic signals at high field strength may help avoid overdiagnosis of prostate cancer.

1. Introduction

Four metabolic peaks have been well described when magnetic resonance (MR) spectroscopic imaging of the prostate is performed at 1.5 T. These metabolites are choline, polyamines, creatine, and citrate, occurring at downward frequency shift relative to water of approximately 3.2, 3.1, 3.0, and 2.6 ppm, respectively [1]. Prostate cancer is characterized by raised choline (a normal cell membrane constituent, which is elevated in many tumors) and reduced polyamines and citrate (constituents of normal prostatic tissue). In practice, the choline and creatine peaks are closely adjacent and both overlap with the polyamine resonance to the point of being inseparable, but fortunately, at 1.5 T, the overlap of polyamines with the choline and creatine resonances is reduced by the selection of an appropriate echo time. As such, a reasonably simple pattern recognition approach can be taken to the interpretation of MR spectroscopic imaging at 1.5 T, based on comparing the two main peaks (one peak consisting predominantly of choline combined with creatine and the other peak consisting of citrate). If the first peak (i.e., choline and creatine) is lower than the second peak (i.e., citrate), the voxel is likely healthy tissue. If the two peaks are of similar height or if the first peak is higher than the second peak, the voxel is likely malignant [2]. However, such an approach may be inappropriate at 3 T, since in order to acquire a fully upright citrate resonance at a reasonably short echo time, the MR spectroscopic imaging sequence is modified to refocus the J-modulation of citrate [3,4]. This also refocuses the polyamine resonance, which may result in a larger polyamine peak increasing the overlap with the choline and creatine resonances and altering the metabolite ratios and pattern recognition approach that have been used at 1.5 T. For example, there has been anecdotal concern that selective amplification of polyamine signal at 3 T may result in a “first peak” (choline, polyamines, creatine) appearing high relative to citrate in benign prostatic tissue and falsely suggesting the metabolic appearance of malignancy (Kurhanewicz J, personal communication). Therefore, we undertook this study to compare the metabolic findings at MR spectroscopic imaging of benign prostatic tissue performed at 3.0 vs. 1.5 T.

2. Materials and methods

2.1. 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, using our patient database, all patients with biopsy-proven prostate cancer who underwent same-day pretreatment 1.5- and 3.0-T endorectal MR and MR spectroscopic imaging of the prostate in 2005. These patients were scanned at both field strengths during a period in which we were optimizing our endorectal MR technique at 3.0 T following installation of our first 3.0 T scanner in 2004. (This technical development work was conducted as funded research separate from this study, and written informed consent was obtained.) Seventeen patients fulfilled these criteria. Fourteen of the 17 patients were excluded due to confounding prior therapy or spectroscopic data of insufficient quality from either the 1.5- or 3.0-T scan. (The large number of excluded studies reflects the challenges we initially experienced in obtaining good quality MR spectra at 3.0 T). The remaining three patients were included in the study group. Demographic and clinical characteristics of the three patients in the final study group are shown in Table 1.

Table 1.

Demographic and clinical characteristics of the three patients in the study group

Patient no.	1	2	3
Age (years)	70	55	63
Serum prostatic specific antigen (ng/dl)	10.3	33.1	5.9
Gleason score	6	6	6
No. of positive cores	3	1	2

2.2. Magnetic resonance imaging technique

All 3.0- and 1.5-T examinations were performed on a 3.0- and 1.5-T GE Signa scanner (GE Healthcare Technologies, Waukesha, WI, USA), respectively. At 3.0 T, a body coil was used for excitation and a custom-designed rigid coil in conjunction with a pelvic phase array coil was used for signal reception. At 1.5 T, a body coil was used for excitation and either an inflatable endorectal coil (Medrad, Pittsburgh, PA, USA) (n=1) or a USA Instrument rigid coil (USA Instrument, Aurora, OH, USA) (n=2) filled with perfluorocarbon (Flutec-T14; F2 Chemicals 91 Ltd, Lancashire, UK) was used for signal reception. Both 1.5- and 3.0-T examinations included the following imaging sequences and parameters obtained after a sagittal fast spin echo localizer to check coil placement:

1. Axial T1-weighted spin echo (repetition time/echo time=950/9 ms, 5-mm slice thickness, 1-mm interslice skip, 24-cm field of view, and 256 × 192 matrix).

2. Oblique axial (i.e., transverse to the long axis of the prostate) T2-weighted fast spin echo (repetition time/effective echo time=6000/102 ms, 3-mm slice thickness, no interslice skip, 12-cm field of view at 3.0 T and 14-cm field of view at 1.5 T, 256 × 192 matrix, and no phase wrap).

3. Coronal T2-weighted fast spin echo images (repetition time/effective echo time=6000/102 ms, 3-mm slice thickness, no interslice skip, 14-cm field of view at 3.0 T and 16-cm field of view at 1.5 T, 256 × 192 matrix, and no phase wrap).

2.3. Magnetic resonance spectroscopic imaging technique

A similar approach for MR spectroscopic imaging was used at both 1.5 and 3.0 T. After review of the axial T2-weighted images, a volume of prostate tissue was selected to maximize coverage of the gland without including the adjacent rectum and periprostatic fat. The influence of chemical shift on the apparent location of the selected volume was reduced by the higher spectral bandwidth of the spectral–spatial pulses [5,6]. Outer voxel saturation pulses were used to further sharpen volume selection and conform the selected volume to the shape of the prostate to eliminate susceptibility artifacts from periprostatic fat and rectal air [7]. Standard prescan preparation was performed just prior to the magnetic resonance spectroscopic imaging (MRSI) acquisition, including setting of center frequency on water resonance, setting of transmit gain, and automatic shimming of the MRSI volume with linear shim correction. At 1.5 T, spectroscopic data were acquired with a water- and lipid-suppressed double spin echo point-resolved spectroscopy sequence (PRESS) using two spectral–spatial 180° pulses for excitation. Datasets were acquired as 16 × 8 × 8 phase-encoded spectral arrays, with a repetition time/echo time of 1000/130 ms and a 17-min acquisition time. At 3.0 T, data were acquired with the Malcolm Levitt's (MLEV) composite pulse decoupling variant of PRESS [8]. This PRESS sequence differs from the 1.5-T sequence in that the phase-modulated frequency-selective spectral–spatial refocusing pulses have been optimized for 3.0 T, and a train of non-selective pulses utilizing the MLEV phase cycling scheme is added between the dual-band refocusing pulses to refocus the J-modulation of citrate. The pulses were also designed to provide attenuated refocusing of water and lipid suppression. Datasets were acquired as 12 × 8 × 8 phase-encoded spectral arrays, with a repetition time/echo time of 1300 to 1500/85 ms and a 17-min acquisition time. The spectroscopic imaging data were zero filled from 8 to 16 in both the anteroposterior and craniocaudal directions to increase the likelihood of optimal alignment between spectroscopic voxels and the peripheral zone. Voxel size was 0.34 cm³ at 1.5 T and 0.16 cm³ at 3.0 T. Custom software interfaced with Interactive Data Language was developed to analyze and display the spectroscopy data. The reconstruction began with a 3-Hz Gaussian apodization in the time domain and was followed by Fourier transform in the time domain and three spatial domains to produce the spectral arrays. The spectra were then corrected for phase and frequency variations utilizing the residual water peak. The resultant spectra were baseline corrected before peak heights were calculated. Integrated peak area value ratios for choline, polyamine, and creatine to citrate were automatically calculated for each voxel. Magnetic resonance spectroscopic imaging data, including the spectra and associated metabolic ratios, were overlaid on the corresponding axial T2-weighted images. The total examination time was 1 h, including coil placement and patient positioning.

2.4. Selection of matched voxels

An attending radiologist (FVC) with more than 10 years of experience in the interpretation of MR and MR spectroscopic imaging of the prostate reviewed all images on a picture archiving and communication system workstation (Impax; Agfa, Mortsel, Belgium) to select voxels of benign-appearing prostatic tissue that were spatially coregistered at 1.5 and 3.0 T. This was done by first aligning the axial T2-weighted images of both studies, using anatomic landmarks such as the urethra, distinctive areas of benign prostatic hyperplasia in the central gland, and periprostatic anatomy. Then voxels of peripheral zone tissue that appeared benign at 1.5 T were identified, based on typical high T2 signal intensity and benign pattern spectra. No attempt was made to incorporate biopsy findings into the selection of benign peripheral zone tissue, because of the known inaccuracy of biopsy results for tumor localization [9]. Finally, the location of these voxels was examined to see if there was a spatially coregistered voxel at 3.0 T. Voxels were considered coregistered if the center of both voxels was in the same anatomic location. Note that the spectral appearance at 3.0 T was not part of the evaluation process for selecting voxels; only the anatomic location of voxels at 3.0 T determined if they were appropriate for inclusion in the final set of matched voxels.

2.5. Image interpretation

Based on the selection process above, 71 paired voxels of presumed benign peripheral zone tissue were identified (i.e., a total of 142 voxels). The spectra from these 142 voxels were converted into 142 individual JPEG (Joint Photographic Experts Group) image files (72 dots per inch) on a personal computer (Inspiron 5100; Dell, United States). After appropriate cropping and sharpening of the individual voxels using image-editing software (Adobe Photoshop Elements, version 1.0; Adobe Systems, Seattle, WA, USA), the 142 highlighted voxels were inserted into a slideshow (Power-Point; version 2002 SP3; Microsoft Corp., United States) in a randomized order as a single presentation of 142 slides. Randomization was ensured using a random number generator. Two attending radiologists (AQ and ACW) with 9 and 5 years of experience in the interpretation of prostate MR and MR spectroscopic imaging independently reviewed the slideshow file and characterized each of the 142 voxel images on a previously described five-point scale (1=likely benign, 2=possibly benign, 3=equivocal or indeterminate, 4=possibly malignant, and 5=likely malignant) [10]. The readers interpreted the entire slideshow in one session. Readers were aware that the spectra were derived from patients with a biopsy-proven diagnosis of prostate cancer but were unaware of which voxels were from 1.5 T and which were from 3.0 T (potentially “unmasking” data such as resolution was not present because of the manner in which the spectra were individually “carved” out and presented to the readers).

2.6. Statistical analysis

Analysis of the data was performed in a two-step process. The first step was essentially qualitative, to determine if MR spectra appeared more suspicious at 3.0 T. For this step, the 5-point scale was divided into three categories, benign (score of 1 or 2), indeterminate (score of 3), or malignant (scores of 4 or 5). A significant change in interpretation was considered present if the score assigned at 3.0 T was in a different category from the score assigned at 1.5 T, whether the change was toward “overrating” (more malignant) or “underrating” (more benign). The number of under- or overrated voxels (by one or both readers) at 3.0 T compared to 1.5 T was compared using a paired t test. In the second step, which was more quantitative, the source of “overrating” at 3.0 T was investigated by comparing the peak area value ratios for choline, polyamine, and creatine to citrate at 3.0 T compared to 1.5 T using a Wilcoxon ranked-sum test for nonparametric paired data values. Data were tabulated and analyzed using a spreadsheet (Microsoft Excel 2002 SP-1, Microsoft Corp.).

3. Results

In the initial qualitative analysis, 21 of the 71 voxels were considered to have more malignant spectra at 3 T compared to 1.5 T (14 by reader 1, 2 by reader 2, and 5 by both readers), with only 3 voxels (all by reader 1) considered more benign (P<.001). A representative example is shown in Fig. 1. The mean choline to citrate integral peak ratio in the 21 more malignant-appearing voxels was significantly higher at 3 T compared to 1.5 T (0.39 vs. 0.11; P<.001), as was the mean polyamine to citrate integral peak ratio (0.25 vs. 0.07; P<.001). Creatine-to-citrate peak integral values were unusable due to noise.

Fig. 1.

(A) Photomontage showing an axial T2-weighted endorectal MR image of the prostate obtained at 1.5 T in a 63-year-old man with biopsy-proven prostate cancer and the MR spectrum derived from the single outlined region of benign-appearing peripheral zone tissue. (B) Photomontage showing the corresponding axial T2-weighted endorectal MR image of the prostate and the associated MR spectrum obtained at 3.0 T in the same patient on the same day. Note that the spectrum appears more malignant, with a higher combined choline, polyamine, and creatine peak relative to citrate than the spectrum shown in panel A.

4. Discussion

Our results indicate that MR spectra of benign prostatic tissue obtained at 3.0 T may appear more malignant than those obtained at 1.5 T, and that this is due to relative amplification of the choline and polyamine peaks at 3.0 T relative to 1.5 T. Awareness of this selective amplification of metabolic signal is important to avoid overdiagnosis of prostate cancer as 3.0-T MR scanners and endorectal MR studies of the prostate become more common. Our results also indicate that this relative amplification of the choline and polyamine peaks is of sufficient magnitude to change reader interpretation, which was based primarily on a pattern recognition approach. We suspect the choline elevation seen in our study is likely artifactual, because the structure of the choline molecule is such that J-modulation should not be a factor. Given that current commercially available software calculates metabolic ratios based on integration over a defined frequency range for each of the metabolites, it is probable that the choline elevation at 3.0 T seen in this study reflects overlap of the tail of the large polyamine peak with the choline peak. Furthermore, in practice, automatically generated ratios are often meaningless due to noise (e.g., negative numbers or values that are implausibly high or low), so that the interpreting radiologist relies principally on pattern recognition. Our study indicates that pattern recognition developed at 1.5 T for characterization of benign peripheral zone prostatic tissue may not be appropriate at 3.0 T. Relative amplification of the choline and polyamine peaks at 3.0 T is likely due to the effect the MLEV-PRESS sequence has on refocusing the J-coupling of polyamines. At 1.5 T, citrate is strongly coupled, and an upright singlet peak is observed with PRESS localization using echo times of 120 to 130 ms, with the polyamines being upright but reduced in signal intensity due to J-modulation. At 3 T, however, the citrate resonance, due to J-modulation changes, is not completely upright until an echo time of 260 ms, which would result in substantial T2 signal loss and noisy spectra. The MLEV-PRESS sequence addresses this problem by incorporating a nonselective radiofrequency pulse train, which has the effect of refocusing the J-modulation of citrate, allowing for the acquisition of an upright citrate resonance with most of the magnetization localized to the center lines at a reasonably short echo time of 85 ms [11]. However, the radiofrequency pulse train also refocuses the J-modulation of polyamines, producing a much larger polyamine peak in healthy tissue at 3 T than is observed at 1.5 T. While, theoretically, this could result in better characterization of benign prostatic tissue, since ex vivo MR spectroscopy shows that polyamine elevation is a robust biomarker of healthy glandular prostatic tissues [12,13], this is dependent on high-quality spectroscopy that can routinely allow spectral resolution of choline, polyamine, and creatine peaks. In practice, in vivo spectral quality may lack sufficient resolution, and this can be compounded by the loss in digital resolution associated with the generation of screen saves for interpretation and archiving. Additional more systematic and rigorous studies utilizing optimized display processes for spectral interpretation will be required to develop and validate spectral criteria for MR spectroscopic interpretation of prostatic tissue at 3.0 T, similar to prior studies conducted at 1.5 T [10].

The finding that MR spectral interpretation of benign prostatic tissue at 3.0 T cannot automatically be based on the same pattern recognition at 1.5 T speaks to the broader realization that abdominal MR imaging at 3.0 T poses specific challenges and development of new strategies and techniques for examination protocols [14,15]. With respect to MR spectroscopic imaging, it would seem self-evident that higher field strength is desirable because of improved sensitivity and chemical shift resolution, but such gains may be offset by relaxation time differences, line broadening due to increased magnetic susceptibility effects, and radiofrequency coil efficiency [16]. Even when these advantages are realized, MR spectral signals are more complex than MR imaging due to the greater degree of strong coupling effects at lower field strength. Such relative changes in spectral peaks have been documented at brain spectroscopy [16] and may have clinical implications; for example, studies investigating the use of MR spectroscopic imaging to distinguish postradiation necrosis from recurrent tumor in the brain have yielded different threshold values of choline to creatine and choline to N-acetylaspartate ratios at 3.0 than 1.5 T [17,18]. To date, no publications have addressed the potential clinical impact of such factors on the interpretation of prostate MR spectroscopic imaging at 3.0 vs. 1.5 T.

We recognize that our study has limitations. Perhaps the most obvious is that we do not have histopathological proof that the tissue in our selected study voxels was benign. However, while MR and MR spectroscopic imaging at 1.5 T have limited sensitivity for prostate cancer sextant localization, the specificity of both parameters combined has been reported as 90% (44/49) [19]. Given that voxels were only included if both T2 and MR spectral findings at 1.5 T were benign, we doubt that many of the voxels actually contained malignancy. Irrespective of this contention, the voxels were anatomically matched, so that the relative amplification of choline and polyamine signal observed at 3.0 T vs. 1.5 T would remain valid. Our results were based on a relatively small number of matched voxels (n=71) drawn from only three patients. Our methodology was modified to account for these small numbers, in that we first qualitatively identified whether there was a direction of change toward benign or malignant when interpreting 3.0-T spectra from benign prostate tissue, and then performed a quantitative analysis of metabolic ratios in those voxels that were in the direction of the interpretative change. While a global analysis of all voxels at 3.0 vs. 1.5 T would have been preferable, we were concerned that such an approach may have masked significant differences due to averaging across small subgroups. In order to blind readers to field strength and other potentially confounding factors such as T2 signal and metabolic findings in adjoining voxels, images were manipulated prior to interpretation, which may have reduced the spectral quality, comprising the interpretation of individual voxels. However, this was the case for all voxels regardless of field strength. A broader study of greater patients using step section coregistered radical prostatectomy specimens as the reference standard and with high resolution spectral images for interpretation will be required to further confirm and elucidate our results and address these limitations.

In conclusion, at endorectal MR spectroscopic imaging of benign peripheral zone prostatic tissue, choline and polyamine peaks are more prominent relative to citrate at 3 T compared to 1.5 T; awareness of this selective amplification of metabolic signal may help avoid overdiagnosis of prostate cancer.