To elucidate the performance of apparent diffusion coefficient (ADC) map in localizing prostate carcinoma (PC) using stepwise histopathology as a reference.
To elucidate the performance of apparent diffusion coefficient (ADC) map in localizing prostate carcinoma (PC) using stepwise histopathology as a reference.
Preoperative MR images of 37 patients with PC who had undergone radical prostatectomy were retrospectively evaluated. First, T2-weighted images (T2WI) alone were interpreted (T2WI reading), and then T2WI along with ADC map were interpreted (T2WI/ADC map reading). Sextant-based sensitivity and specificity, and the ratio of the detected volume to the whole tumor volume (% tumor volume) were compared between the two interpretations, and results were also correlated to Gleason's scores (GS). ADC values were correlated to histological grades.
Sensitivity was significantly higher in T2WI/ADC map reading than in T2WI reading (71% vs. 51%), but specificity was similar (61% vs. 60%). By adding ADC map to T2WI, % tumor volume detected increased significantly in transitional zone (TZ) lesions, but not in peripheral zone (PZ) lesions. % tumor volume detected with T2WI/ADC map reading showed a positive correlation with GS of the specimens. Less differentiated PC were associated with lower ADC values and higher detectability.
T2WI/ADC map reading was better than T2WI reading in PC detection and localization. This approach may be particularly useful for detecting TZ lesions and biologically aggressive lesions. J. Magn. Reson. Imaging 2007. © 2007 Wiley-Liss, Inc.
PROSTATE CARCINOMA (PC) is one of the most common malignancies in males, and it accounts for approximately 30,000 new annual deaths in the United States (1). To date, surgical resection of the whole organ has been the only method of eradicating this type of malignancy; however, less invasive alternative local therapies, including intensive modulated radiation therapy (IMRT), high-intensity focused ultrasound (HIFU), and brachytherapy, are being introduced due to the increasing clinical demand for the preservation of functional aspects of the prostate and related organs (2–4).
Various MR approaches have been investigated and applied to localize PC; such attempts have included dynamic studies and MR spectroscopy, which provided promising but inconsistent results (6–16). Diffusion-weighted images (DWI) and calculated apparent diffusion coefficient (ADC) mapping are additional approaches, involving the representation of the Brownian movement of water molecules. This new parameter differs from conventional T1 or T2 relaxivity, dynamic enhancement characteristics, or spectroscopic information. This DWI or the ADC map has been applied for examining various part of the body, in particular to detect or differentiate malignancies (17–20). Regarding PC, several preliminary investigations have been sporadically reported (21–25), with promising results. In this article, we applied the ADC map obtained from DWI to the diagnosis of PC, and evaluated its performance at detecting and localizing PC using stepwise histopathological data as a gold standard. We also correlated the results to the histological grades of the lesions and Gleason's scores (GS) of the specimens, in order to characterize the potential clinical usefulness of this novel technique.
Between January 2000 and March 2004, 124 patients underwent radical prostatectomy at our institute. Among these patients, 37 who had undergone preoperative MRI (including DW imaging) were retrospectively selected, and these 37 patients formed the present study population. One of the 37 subjects had received hormonal therapy prior to surgery. The age of the selected patients ranged from 56 to 75 years old (mean = 66 years). The preoperative prostate-specific antigen (PSA) level ranged from 0.7 to 54.8 ng/mL, with a mean of 11.9 (normal range < 4.00 ng/mL). All subjects had undergone transrectal or transperineal biopsy prior to MR imaging and had been pathologically diagnosed with malignant foci in the prostate. The period between biopsy and MR ranged from six to nine weeks (mean = 7.2 weeks), and the period between MR and surgery ranged from zero to eight weeks (mean = 2.5 weeks). The institutional review board at our hospital did not require that written informed consent be obtained for this study due to its retrospective nature. The current study was designed and performed according to the declaration of Helsinki (26).
A total of two 1.5T units (Magnetom Symphony and Vision; Siemens, Erlangen, Germany) were used with a pelvic multichannel phased-array coil (12 channels). After routine T1-weighted spin-echo (TR/TE/number of excitations [NEX] = 500 msec/12 msec/2) axial images had been obtained, T2-weighted fast spin-echo (TR/TE/Turbo factor/NEX = 3000 msec/102 msec/15/3, slice thickness = 5 mm, interslice gap = 30%) axial and coronal images with axial DWI were obtained using the single-shot spin-echo echo-planar imaging (EPI) technique. The matrix and field-of-view of the T1- and T2-weighted images (T1WI and T2WI) were 256 × 512, and 20 cm, respectively. The slice thickness and gap of DWI were identical to those of routine T1WI and T2WI. Sequential sampling of the k-space was used with echo-time (TE) = 110–135 msec and bandwidth = 1250 Hz/pixel, and 128 lines of data were acquired in 0.3 seconds. No parallel imaging technique was applied. Other parameters included a field-of-view = 240 mm, matrix size = 128 × 128, and the acquisition of four signals. All images were obtained while the patients maintained normal and consistent breathing, and a fat-saturated pulse was used for the DWI to exclude severe chemical-shift artifacts. A contrast-enhanced dynamic study was performed and postcontrast T1WI were obtained in all cases, the details of which are not given here, as they are out of the scope of this work.
DWI were acquired with motion-probing gradient pulses applied along three (x-, y-, and z-axes) directions with three b factors of 0, 500, and 1000 seconds/mm2. ADC maps were automatically generated on the operating console using all seven images (b = 0 and two b-values in each direction), and the ADC values were obtained by measuring the intensity of the map.
One experienced pathologist created transverse sections of the specimens: the most apical and basic sides of the specimen were cut to a thickness of 6 mm, and the remaining portion (majority of the specimen) were cut to a thickness of 4 mm. Each section of each specimen was digitally photographed together with a ruler along the edge, which serve as a size reference, and the areas of the PC that had been microscopically determined were marked on the digital photographs by the pathologist (pathological map) using commercially available presentation software (Microsoft PowerPoint 2002; Microsoft Corporation, Redmond, WA, USA). All PC foci, including infiltrating foci that did not form apparent masses, were marked on the map and divided into the sextants according to their location, as follows: right and left apices (lower third), midglands (middle third), and bases (upper third). Lesions consisting of uniform histological grades were also documented as such on the photograph. This pathological map was used as the gold standard in this study.
First, we subdivided the glands of all patients into sextants on the MR images. The presence of PC in each sextant was retrospectively evaluated and recorded by two radiologists in a consensus. Initially, T2WI alone (T2WI reading) were interpreted and then T2WI and the ADC map (T2WI/ADC map reading) were interpreted. The readers were informed that the patients had undergone surgery for PC, but no other clinical information, (e.g., PSA level or biopsy results) was provided to the readers. On either T2WI or the ADC map, areas with apparently lower signal intensity than that of the surrounding tissue were considered to represent PC, according to the previously reported descriptions (5–16, 21–25). Regarding T2WI/ADC map reading, when the findings on either sequence were equivocal, those on the ADC map were considered to have priority, if image degradation is not prominent. Sensitivity and specificity were thus calculated based on the sextant evaluation and the results of the two interpretations were compared.
We then directly compared the MR images and the pathological map on a lesion basis, and we excluded positive sextants in which the noncancerous areas had been interpreted as PC on the MR images (false-positive lesions in positive sextants); thus the true sensitivity was calculated. As for a lesion whose location at least partially overlapped on the MR images and on the pathological map, the lesion was considered false-positive when the maximum transverse diameter measured at MRI was out of the range of 50% to 150% of the maximum transverse diameter measured on the pathological map (16). Lesions seen at MRI were only considered truly positive if the suspected foci were in the same relative portion of the prostate.
We also marked the approximate areas of PC on a pathological map, which had been detected on MR images (detected PC) using MR images as reference, and the areas of the whole PC and the detected PC were traced and measured using NIH software (NIH Image, version 1.63; National Institutes of Health, Bethesda, MD, USA). Volume was calculated by multiplying the measured areas by thickness (0.4 cm). False-positive lesions, including those in positive sextants as defined above, were excluded in this evaluation. Thus, % tumor volume (percentage of the tumor volume detected to the whole tumor volume; range = 0–100%) per patient was compared between the T2WI reading and T2WI/ADC map reading. The lesions were subclassified into peripheral zone (PZ) and transitional zone (TZ) lesions according to their location on the pathological map, and these two groups were compared in terms of % tumor volume per patient. Then, the % tumor volume per patient was also correlated to the GS of the patients.
Lesions with uniform histological grades that were larger than 1 cm in their short axes were selected on the pathological map and their ADC values were measured at the corresponding sites on the ADC map by one radiologist, even when there were no detectable abnormal areas on the ADC map. Correlation of ADC value and histological grades was thus evaluated. We also evaluated the detectability of these lesions in correlation with the their histological grades.
Finally, by comparing the MR images and pathological maps, we selected areas of PC larger than 1 cm in their short axes that had not been detected on ADC map (false-negative lesions). The possible reasons for these lesions not being detected on MR images were analyzed. We also selected areas of low signal intensity larger than 1 cm in their short axes on the ADC map that did not correspond to PC on the pathological map (false-positive lesions). The corresponding sites were marked on the pathological map, and pathological details of these areas were then reevaluated by the pathologist.
In one patient, microscopic evaluation of the resected gland revealed no PC, although preoperative biopsy had suggested the presence of PC in one of the sextants. A total of 222 sextants in 37 patients were evaluated, among which 147 sextants were positive and 75 were negative for the presence of PC. A total of 79 and 105 positive sextants, and 45 and 46 negative sextants were correctly diagnosed with T2WI reading and T2WI/ADC map reading, respectively. Among the 79 and 105 positive sextants that were classified as positive, 45 and 31 were excluded for the calculation of true sensitivity because they were regarded as noncancerous areas on the lesion-based evaluation. The sensitivity, true sensitivity, and specificity of T2WI reading were 53%, 23%, and 60%, whereas the sensitivity, true sensitivity, and specificity of the T2WI/ADC map reading were 71%, 50%, and 61%, respectively. There was a significant difference between the two interpretations in terms of the sensitivity and true sensitivity (P < 0.01, McNemar chi-squared test), but no difference in the specificity (P = 0.97).
The areas of PC on the pathological map ranged from 3 mm to 22 mm in their shortest dimension (mean = 7 mm). The calculated volumes of PC per patient ranged from 0 to 5.75 cm3 (mean = 1.49 cm3). There were 261 and 151 areas of the PC in PZ and TZ; the sum of tumor volume were 38.8 and 16.3 cm3, respectively. The majority (33/37) of our patients had areas of PC in both PZ and TZ. The mean % tumor volume per patient detected by T2WI reading were 20% (range = 0–100%), 41% (range = 0–100%), and 7% (range = 0–91%), for total lesions, the PZ lesions, and TZ lesions, respectively. The mean % tumor volume per patient detected by T2WI/ADC map reading for total lesions, the PZ lesions, and TZ lesions, were 47% (range = 0–100%), 48% (range = 0–100%), and 44% (range = 0–100%), respectively. Overall (not per patient, but total sum) % tumor volume by T2WI reading were 30%, 55%, and 20%, for total lesions, the PZ lesions, and TZ lesions, respectively. Overall % tumor volume by T2WI/ADC map for total lesions, the PZ lesions, and TZ lesions, reading were 55%, 57%, and 52%, respectively. Addition of the ADC map to T2WI interpretation revealed a significant increase in % tumor volume in total lesions (P = 0.0002, Wilcoxon signed rank test) and in TZ lesions (P < 0.0001), but not in PZ lesions (P = 0.158, not significant [NS]).
For the resected specimens in 35 patients (excluding one patient whose specimen revealed no evidence of PC and another who had received hormonal therapy), GS were assigned. There was a weak but significant correlation between % tumor volume by T2WI/ADC map reading and GS (ρ = 0.40, P = 0.022, Spearman's rank correlation test) (Fig. 1). Patients with higher GS tended to have higher % tumor volume, namely had more chance for PC to be detected by T2WI/ADC map reading. No significant correlation was observed between % tumor volume detected by T2WI reading alone and GS (P = 0.41, NS). Because it has been reported that the larger the tumor is, there is the better correlation between tumor volume measured on T2WI and histopathologic volume (14), we might need to exclude the effect of the volume of PC in evaluating the correlation between % tumor volume by T2WI/ADC map reading and GS. We therefore evaluated the partial correlation coefficient between either tumor volume or GS and % tumor volume detected (Table 1). The results showed that % tumor volume detected with T2WI/ADC map reading significantly correlated to GS, but not to tumor volume, and also that % tumor volume detected with T2WI reading correlated to the tumor volume, but not to GS.
|% Tumor volume detected|
Regarding the ADC value measurement, 53 lesions with uniform histological grades that were larger than 1 cm in their short axes were selected from 36 patients. There were 21, 26, and six lesions, in well-, moderately-, and poorly-differentiated adenocarcinomas, respectively. The mean size (short axis diameter) of these lesions was 1.14 cm (range: 1.0–1.9 cm), 1.23 cm (1.0–2.2 cm), and 1.03 cm (1.0–1.2), respectively, showing no significant difference (P = 0.38, one-way factorial analysis of variance [ANOVA]). The ADC values of well-, moderately-, and poorly-differentiated PC were 1.19 ± 0.15, 1.10 ± 0.24, and 0.93 ± 0.20 × 10–3 mm2/second (mean ± standard deviation [SD]), respectively. Significant difference in ADC values was seen only between well- and poorly-differentiated PC (P = 0.019), and difference between well- and moderately-differentiated, or that between moderately- and poorly-differentiated PC was not significant (P = 0.38 and 0.13, one-way factorial ANOVA with Scheffe's post hoc test). There was a subtle but significant correlation between the histological grades and ADC values (ρ = –0.18, P = 0.014, Spearman's rank correlation) (Fig. 2). Of these 53 lesions, 13 (62%), 24 (92%), and six (100%) were detected on the ADC map, in well-, moderately-, and poorly-differentiated adenocarcinomas, respectively. The detectability of these lesions differed significantly among histological grades (Kruskal-Wallis test, P < 0.01) and a significant correlation was observed, whereby the less differentiated lesions were associated with higher detectability (P < 0.01, Cochran-Armitage test for trend).
As for false-positive lesions, 54 foci were selected from 27 patients. The pathological details of these lesions included hyperplastic nodules in 22, normal structure in 14 (periejaculatory duct tissue in five, asymmetric central zone tissue in four, base of the seminal vesicle in three, asymmetric anterior fibromuscular stroma in one, and vermontanum in one), intraacinar hemorrhage in 10, and chronic prostatitis in eight. As for false-negative lesions, 15 foci in 15 patients were selected. Possible causes for these false-negatives were well-differentiated infiltrative lesions with preserved gland formation in six lesions, susceptibility artifact from rectal or intestinal gas in four lesions, and susceptibility artifact from metallic prosthesis at the hip joint in one lesion. Causes for the remaining four lesions remained unknown. Representative cases are shown in Figs. 3, 4, and 5.
Previous investigators have reported significant difference in ADC values between PC and normal prostatic tissue, using biopsy-proven histopathology as a reference (21–25). To date, there have been few reports regarding PC detection with DWI or ADC mapping in patients who had undergone prostatectomy. We therefore attempted in the present study to clarify the clinical usefulness and significance of DWI or ADC mapping using a stepwise histopathology as a gold standard.
According to our results, sextant-based sensitivity and true sensitivity improved significantly, i.e., up to 71% and 50%, respectively, when the ADC map was interpreted along with T2WI, although the specificity remained unchanged at around 60%. In terms of the volume of PC, T2WI/ADC map reading detected approximately one-half of the tumor in the gland. Previously reported sensitivity and/or detection rate of PC on MRI (i.e., T2WI with or without spectroscopy), has been in a range between 20% to 80% (6–16). Recently, Hom et al (16) reported a detection rate of around 20% using endorectal coil MR images and MR spectroscopy for cases of PC in PZ, and using meticulous histopathologic evaluation methods and strict criteria. Our data also revealed a rather low sensitivity or detection rate, at least in part because we also used the pathological map as a strict gold standard, including all small foci of PC. Another explanation for this low sensitivity may have been the criteria we used to detect PC in our study. Namely, we considered “areas with low signal intensity relative to the surrounding tissue” as PC in both PZ and TZ; however, Li et al (27) recently applied more meticulous criteria to diagnose TZ lesions and achieved better results. Although the application of different criteria might have improved the sensitivity, our present data suggest that a T2WI/ADC map reading cannot be used as a reasonable guide yet in localization of PC in the context of local therapies such as IMRT or HIFU. Practical use of ADC maps may therefore be limited at present, for example, to a guide for rebiopsy in patients with high PSA levels but with negative biopsy results.
To improve sensitivity of the T2WI/ADC map reading, it will be necessary to detect well-differentiated PC with glandular formation, which was the most common cause of false negativity in our series. In our evaluation of the lesions larger than 1 cm in the shortest dimension, the ADC values of well-differentiated PC were significantly higher than moderately- or poorly-differentiated PC, and nearly one-half of these lesions were visually missed on the T2WI/ADC map interpretation. The pathological architecture of these lesions, namely, preserved glandular formation with a significant volume of fluid-filled luminal space, which is similar to that of normal prostatic glandular tissue, supports the relatively high ADC values of these lesions; these features rendered it difficult to detect such lesions on the ADC map. Improving the signal-to-noise ratio (SNR) by using 3T hardware or a coil with more channels may help detect these lesions. Concurrent usage of a parallel imaging technique (24, 28, 29) may also help reduce the susceptibility artifacts, the second most common cause of false negativity.
To improve specificity, it will be necessary to differentiate between hyperplastic nodules and PC; the lack of such differentiation was the most common cause of false positivity in our results. Again, improving the SNR by either the use of 3T hardware or new coils may help differentiate these two entities. The second most common reason for false positivity was normal prostatic tissue, including periejaculatory duct tissue, central zone tissue, tissue at the base of the seminal vesicle, and so on. The ADC values of these structures were 0.97 ± 0.18 × 10–3 mm2/second (not shown in the results), which were within the range of ADC values of PC. Precise anatomic evaluation may obviate this misdiagnosis; however, it would remain difficult to discriminate these structures from focal involvement of PC on an ADC map. Combination with MR spectroscopy or dynamic MR study may be of some help in resolving this problem.
The promising aspects of our findings are as follows. First, the T2WI/ADC map reading significantly increased the detection of PC located in TZ. In terms of the % tumor volume, the detection rate of T2WI/ADC map reading was approximately 50% regardless of the location of PC, whereas that of T2WI reading alone was significantly worse in the case of lesions in TZ than in those in PZ. Because PC in TZ has remained a diagnostic problem either for transrectal ultrasound (US) or conventional MR, use of DWI and ADC maps could be of aid in the detection of lesions in TZ in particular. Although Li et al (27) recently reported improved detection of PC in TZ using the combined criteria of T2WI and postcontrast T1WI, adding information of ADC map might further improve the detection of PC in TZ.
A second promising point suggested by our results was that the higher the histological grade of PC, the lower the ADC values of PC, which increases the chances of PC being detected on an ADC map. This finding may also be related to the positive correlation between % tumor volume detected by T2WI/ADC map reading and GS (Table 1). Such information may be clinically important because it may suggest that the biologically aggressive components or subsets of PC are more likely to be detected by T2WI/ADC map reading than would less aggressive components. As the histological grade or GS increases, there have been shown to be more chances of cellular architectures exhibiting little gland formation, such as medullary or solid patterns (30, 31); these features may explain the low ADC values in these lesions. Although the ADC values obtained in our study were comparable to the previously reported values (21–25), this is the first study to reveal the relationship between histological grade or GS and ADC values in cases of PC.
A third promising issue suggested by the results from but a single patient was that the areas of PC was relatively clearly depicted on ADC map within areas of diffusely decreased signal intensity on T2WI; this was a patient who had received preoperative hormonal therapy (androgen deprivation) (Fig. 4). It was already well known that posthormonotherapy prostatic tissue becomes atrophic and diffusely hypointense, which impairs pertinent MR detection of PC (32, 33). The ADC map might be of help in evaluating patients still undergoing or following hormonal therapy.
There are several limitations to the present study. First, this study was retrospective in nature, and all patients in this series had undergone biopsy prior to MR examination. Although the period between biopsy and MR examination in our study was more than one month, which is reportedly sufficient to avoid biopsy effects on MR images (34, 35), the histopathological evaluation revealed intraacinar hemorrhage, which was the third most common cause of false positivity. Another technical aspect related to the retrospective nature of the study was the subtle differences between MR images and pathological section in terms of the slice interval and slice direction; such differences might have exerted an influence on the precise correlation between the depicted abnormality and the lesions observed on the pathological map. Prospective studies are needed that are designed in such a manner that MR examination is performed prior to biopsy and in which the MR and pathological sections are identical. In addition, due to the limitation associated with the hardware, TE of EPI used for the DWI was rather long (110–135 msec) in our study, as compared to that of previous reports (96–120 msec) (21–25). This long TE may have led to the low SNR of the images, particularly in cases involving tissues with short T2 characteristics, which in turn possibly led to the incorrect calculation of ADC values. Use of a parallel imaging technique might have improved this situation, allowing for a shorter TE (24, 28, 29), although this technique was not available at the time we started this study. Third, because of the significant image distortion of the ADC map, we did not perform any volume measurement on MR images, although in this study, such measurements were carried out on the pathological map by encircling the approximate area of PC by visual inspection using MR images as a reference. Again, a parallel imaging technique would have been useful to reduce the image distortion secondary to the susceptibility effect from intestinal gas or metallic prosthesis (24, 28, 29), which would in turn have enabled us to measure the areas of interest on the ADC map.
In conclusion, the ADC map derived from DWI of MRI performed with a phased array coil without parallel imaging technique significantly improved PC detection and localization when interpreted together with T2WI, although the performance of this method might not yet be sufficient for it to serve as a guide for local therapies. This novel approach to the diagnosis of PC is expected to be particularly useful for the detection of PC lesions located in TZ, as well as detection of lesions with relatively little differentiation, and lesions with a relatively high GS.
We thank Professor Masasumi Tsuneyoshi, Chair of the Department of Anatomic Pathology, Graduate School of Medical Sciences, Kyushu University, for providing us pathologic specimens for this study.