| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Radiology, New York University School of Medicine, New York, NY
10016.
3 Department of Nuclear Medicine, New York University Medical Center, Tisch
Hospital, 550 First Ave., HW 231, New York, NY 10016.
4 Department of Urology, New York University School of Medicine, New York, NY
10016.
Received August 18, 2003; accepted after revision February 2, 2004.
Address correspondence to E. L. Kramer
(elissa.kramer{at}med.nyu.edu).
OBJECTIVE. Our goal was to evaluate the impact of image fusion on the interpretation of indium-111 Prosta-Scint SPECT scans.
MATERIALS AND METHODS. Sixty-seven consecutive patients referred for rising prostate-specific antigen (PSA) levels after initial therapy for primary prostate cancer underwent SPECT 96 hr after infusion of 111In Prosta-Scint, with simultaneous technetium-99m blood pool imaging. Volume data sets from the SPECT scans were then fused with those from CT and MR images of the pelvis using a 3D landmark-based warping program. The SPECT scans were initially interpreted without benefit of MRI or CT fusion. The fused Prosta-Scint MRI-CT volumes were reevaluated by a nuclear radiologist and an MRI radiologist. Independent reviews before and after fusion were available in these patients. Validation of results after fusion was performed through correlation with PSA changes after radiation therapy.
RESULTS. Six patients with sites that could not be evaluated and three without their original Prosta-Scint scanning reports were excluded; thus, 58 patients were studied clinically. Seventy-four of 161 prefusion-positive sites were found to be negative after fusion. These 74 sites subsequently were identified primarily as showing bowel, vessel, or marrow uptake after fusion. In two patients, nodal disease was identified although the review before perfusion indicated none. Twenty-five patients previously thought to have nodal disease appeared to have only local disease after fusion. After local radiation therapy, PSA levels decreased in 12 of 25 patients, increased in five, and were unavailable in eight.
CONCLUSION. Although Prosta-Scint SPECT alone can help in the proper management of recurrent prostate cancer, fusion with MRI-CT of the pelvis can improve the specificity of the examination.
In patients with increasing prostate-specific antigen (PSA) levels after treatment for prostate carcinoma, the distribution of recurrence determines clinical management. In cases of localized prostate bed recurrence, the treatment of choice is radiation therapy. In patients with regional or distant metastatic disease, systemic therapies (e.g., hormonal or chemotherapy) are chosen. Clearly, identification of the nature and location of the recurrence is critical for proper management. In our institution, patients with evidence of disease beyond the prostate bed on two of three imaging studies (indium-111 capromab pendetide [a prostate-specific monoclonal antibody, commonly known as Prosta-Scint (Cytogen)] SPECT, pelvic MRI or CT, and technetium-99m medronate diphosphonate planar bone scanning) are treated for metastasis with systemic therapy. Patients whose disease appears localized to the prostate bed will receive external beam radiation therapy only.
Anatomic imaging techniques such as CT and MRI are routinely used to identify patients with metastatic disease or to stage disease in patients with known metastatic disease. These techniques are unable to detect small deposits of metastatic disease in structures such as lymph nodes of normal size. In a comparison of Prosta-Scint scanning and conventional techniques including CT, MRI, and sonography, the sensitivity for these conventional techniques was 48% [1]. More recently, in another series of patients, conventional MRI showed a 35% sensitivity for extraprostatic (lymph node) disease [2].
Radiolabeled antibodies such as Prosta-Scint represent a possible means for detecting small metastases by providing tissue-specific characterization. The sensitivity and specificity of 111In capromab (Prosta-Scint) have been shown to be 75% and 86%, respectively, in a series of patients with high risk for extraprostatic prostate carcinoma metastases [1]. Other studies have explored and established the utility of the Prosta-Scint scan for clinical staging of prostate cancer [1, 3-8]. To date, other techniques have not offered a clear advantage over Prosta-Scint scanning. FDG positron emission tomography (PET) shows sensitivity for prostate cancer similar to that of Prosta-Scint scanning in lymph nodes [9]. Carbon-11 choline PET has shown greater sensitivity (80%) but requires a cyclotron; therefore, it is not generally available [10]. Fluorine-18 choline is under development but is not yet clinically available [11]. To date, no positron-emitting or single photon-emitting radiopharmaceutical with better sensitivity or specificity is clinically available.
However, Prosta-Scint scans present difficulty in interpretation because of their lack of ability to depict discernible anatomy and the interference from residual blood pool activity, diminishing the specificity of uptake. These studies are usually interpreted in comparison with simultaneous blood pool SPECT scans (99mTc-labeled RBC or early imaging after the administration of the radio-labeled monoclonal antibody). 99mTc RBC scans provide some representation of the vascular structure.
The purpose of this study was to use image fusion to provide direct comparison of Prosta-Scint scanning with an anatomic imaging technique (MRI or CT). We hypothesized that interpretation of the Prosta-Scint SPECT scans with direct overlay on the CT scans or MR images would increase the specificity of the Prosta-Scint scans by decreasing false-positive results. Image fusion has previously been attempted using a vascular outlining method, in which large vessels of the abdomen were outlined using computer-generated wire rings on 99mTc blood pool SPECT and CT volumes. The vascular outlines on the two data sets were then aligned manually, and the resulting spatial relationships served as the basis for the registration [12, 13].
In our study, fusion of Prosta-Scint SPECT and MRI or CT volume data sets was performed using a computer program that was codeveloped with RAHD Oncology Products, which has been previously described and validated [14, 15]. The fusion method used here requires the user to identify points in both data volumes (SPECT and MRI or CT) that correspond in an anatomic manner. Thus there are more points of reference (e.g., liver edge and bone structures) than just the vasculature, providing in some cases a match that has more local detail. After the reference points are selected, the alignment of the two volumes is performed automatically. This procedure has the advantage of being simpler and more widely applicable than the previously mentioned technique.
Materials and Methods
Patients
We retrospectively studied a group of 67 consecutive patients who had
undergone radical prostatectomy, subsequently had increasing PSA levels, and
were referred for both Prosta-Scint scanning and MRI or CT at our institution
between January 1998 and August 2000. This retrospective protocol was reviewed
and approved by our institutional review board. The original Prosta-Scint
study results were used to treat patients clinically; the results of the image
fusion were not incorporated into patient treatment. These 67 patients ranged
in age from 47 to 81 years. In 43 patients, a PSA level was available at the
time of their Prosta-Scint scanning. Of these, the mean ± SD PSA level
was 0.83 ± 1.81 ng/mL (range, 0.12-11.4 ng/mL). In a subset of patients
in whom fusion results would have changed the impression of the extent of
disease (i.e., from extension to lymph nodes to confinement to the prostate
bed), response to local external beam radiation was evaluated by follow-up of
subsequent PSA values.
Nuclear Imaging
All patients underwent Prosta-Scint SPECT scanning 96 hr after the infusion
of 185-222 MBq (5-6 mCi) of 111In Prosta-Scint an
111In-labeled monoclonal antibody to the prostate-specific membrane
antigen. At the time of imaging, they also underwent simultaneous blood pool
imaging using 222 MBq (6 mCi) of 99mTc RBC scanning. The
Prosta-Scint and RBC volumes were obtained simultaneously and therefore were
spatially aligned. All SPECT volumes were acquired on a dual-head gamma camera
fitted with medium-energy collimators (GC7200, Toshiba America Medical
Systems). Twenty percent energy windows centered on 173 keV and 247 keV were
used to acquire the 111In images, and a 15% energy window centered
on 140 keV was used for the 99mTc blood pool images. Images were
acquired in a 128 x 128 matrix with a total of 90 projections and a time
per projection of 50 sec.
The volumes were reconstructed into 128 x 128 matrices, 2 pixels thick, resulting in an x-y pixel size of 4.3 mm and a slice distance of 8.6 mm.
MRI
Fifty-four patients underwent pelvic anatomic scanning using MRI.
High-resolution T1- and T2-weighted axial MRI volumes were obtained on a 1.5-T
scanner (Magnetom Symphony, Siemens Medical Solutions). The volumes were
reconstructed into 256 x 256 or 512 x 512 matrices, which varied
in x-y pixel size between 0.48 and 1.36 mm and in slice
distance (thickness plus gap) between 4.4 and 12.0 mm.
CT
Thirteen patients underwent CT. All CT scans were obtained on a
high-resolution scanner (Light-Speed, GE Healthcare) after the administration
of oral contrast material with or without IV contrast material ([iothalamate
meglumine, 43%] Conray 43, Mallinckrodt). The volumes were reconstructed into
512 x 512 matrices, which varied in x-y pixel size
between 0.66 and 0.84 mm and in slice distance between 5.0 and 8.8 mm.
Registration
The data sets were taken to a common workstation (HP9000/C180,
Hewlett-Packard), where the fusion was performed. The fusion program runs on
several UNIX platforms as well as Linux and uses modules from Visualization
Data Explorer (IBM). The original files were transferred over the local
network to the workstation in DICOM 3 (MRI or CT) or Toshiba National
Electronics Manufacturers' Association 1 format and converted to the standard
American Association of Physicists in Medicine format using Interformat
(Radiologic) [16]. Data were
not altered or lost in the conversion. Both the SPECT and the anatomic volume
data sets were read into the fusion program.
The volume registration and fusion in this program are based on the premise that there exists a mathematic transform that maps the points composing one volume data set (the floating volume) onto the other volume data set (the reference volume). The SPECT volume was invariably used as the floating volume, and the anatomic volume, as the reference (i.e., the SPECT was warped to fit the MRI or CT). The user finds corresponding anatomic-physiologic points apparent in both volumes and marks them in pairs, as landmark points, using any one of the three orthogonal planes. When a landmark is chosen, the corresponding point in the 3D volume is recorded in distance units (millimeters in this case) independent of any voxel location. Landmarks are chosen either by designating a single point (a point landmark) or with the aid of a 3D sphere.
A polynomial function of the form
 |
where x, y, and z are the standard Cartesian coordinates, is used. The 3D point pairs (landmarks) form the input to a least squares linear regression calculation, followed by a Gauss-Jordan matrix inversion to find the coefficients (eigenvalues) of the polynomial that best fit the user-defined landmarks. The coefficients determine the specific equation for each volume pair. Using this equation, we mapped (warped) the SPECT volume to align it with the reference volume. The algorithm allows translation, rotation, scale, and skew of the floating volume to match the reference volume and minimizes the influence of mismatched landmarks. This algorithm has undergone extensive validation [14, 15, 17-19]. The program can display the resulting warped (SPECT) volume fused with the reference (MRI or CT), allows manual tweaking of the registration, and provides several other viewing options, including 3D visualization [14].
Sixty-seven SPECT studies were fused with either MR images (n = 54) or CT scans (n = 13). For fusing, landmarks were selected using primarily the vascular anatomy of the blood pool volumes and the corresponding anatomy present in the MR or CT images. Because the Prosta-Scint SPECT and blood pool volumes were spatially aligned, the landmarks of the blood pool could be used for the Prosta-Scint SPECT volumes. The quality of the registration was judged by inspection of registration of normal organs by a nuclear radiologist, an MRI radiologist, and a medical student, and then consensus was reached (Fig. 1). Landmarks originally picked on the floating volume were transformed as the floating volume was transformed. Numeric values of the distance differences between these transformed landmarks and those originally picked on the reference volume were calculated. In cases of low-accuracy registration, the manual-adjustment feature of the program was used until the registration was satisfactory to the operator. For further numeric evaluation of the registration, a Student's t test was performed on the mean landmark-distance differences. Multivariate analysis was applied to the landmark-distance differences within each patient and across the set of patients
|
Analysis
Prosta-Scint scans were reviewed after fusion with overlay on the MR images
or CT scans available. The results were compared with clinical reviews of the
MRI by itself and the clinical reports of the Prosta-Scint scanning prefusion
on a site-by-site basis. Sites identified in the original SPECT report were
reevaluated. If fusion identified additional foci of suspicious activity, the
sites of foci were logged.
Results after fusion suggesting localized disease were validated by correlation with subsequent PSA levels in the subset of patients in whom these levels were available. Because a pathologic gold standard was not available to confirm our results after fusion, we used PSA response to local radiation therapy to assess the accuracy of the postfusion study [3, 20, 21]. Increasing PSA levels after therapy were taken to indicate disease beyond the radiation port. Decreasing PSA levels were taken to indicate disease confined to the prostate bed radiation port that had responded to the therapy.
Results
The results of image registration are illustrated in Figures 1, 2, 3. The image registration was judged by the users to be successfully performed in all cases using the forementioned method on a qualitative basis (i.e., identifiable anatomy like iliac vessels and the soft tissue of the pelvis showed excellent correspondence on successive slices in all three orthogonal planes). The resulting 3D warped image combines the functional information from the SPECT image with the anatomic-structural framework of the CT or MR image. This method did not require much time and effort (about a half hour from start to finish) but did demand a good knowledge of anatomy.
|
|
The minimal number of landmarks required for a simple affine warp that preserves the original relationship among the structure involved (i.e., the straightness of lines, parallelism, and the ratio between the lengths of two segments of the same line) is four. To obtain an acceptable registration, about 20 are usually required. For this study, approximately 40-50 landmarks were placed, but as few as 10 and as many as 80 were used. In some cases in which the soft tissue of the pelvis required more deformation to obtain a good match, a second-order warp was used; otherwise, a first-order warp was used. Nonaffine transformations, such as the polynomial warping applied here, have more degrees of freedom, allowing line lengths and angles to change and straight lines to become curved, and do not preserve parallelism. By allowing the selection of landmarks in multiple planes, this method eliminates the need for the problematic oblique projection of one of the image sets.
In addition, statistical analysis showed excellent registration of the image volumes. On the basis of the 67 patients whose data volumes were matched, the Student's t test showed that with a significant probability (p < 0.05), all the landmark-distance differences were within 7.9 mm (95% confidence interval, 7.91-9.99 mm). This is less than one SPECT voxel. A global multivariant procedure for the landmark choice within a particular patient and across the sets of patients, which did not indicate a significant difference at the p value equal to 0.05 level, confirmed that the landmark picking did not vary from patient to patent. A plot of the residuals showed them to be normally distributed.
The original Prosta-Scint SPECT reports from 64 of 67 patients were available. Six of these 64 patients had inevaluable sites because MR or CT images depicting those sites were not available for fusion. Nine patients were excluded from the analysis of potential clinical impact; thus, the number of patients was 58. For 87 lesions, representing 28 patients, fusion did not change overall nodal status (Table 1). Seventy-four of 161 prefusion-positive sites on Prosta-Scint scans were found to be negative after fusion. Although these 74 sites originally were thought to represent metastases before fusion, subsequently they were identified as showing bowel (Fig. 2), vessel (Fig. 2), or marrow uptake after fusion. In addition, reviews after fusion identified 13 lesions not previously identified on the original clinical reviews before fusion (Table 1), including findings in two patients in whom the reviews after fusion identified nodal disease previously unrecognized (Fig. 3). In these cases, activity of less intensity extended from the blood vessels and clearly overlay lymph nodes on MRI.
|
In 25 of 58 patients originally thought to have regional or distant nodal disease on the Prosta-Scint reviews before fusion (64/161 lesions, Table 1), postfusion reviews showed only local (i.e., prostate bed) recurrence. All 25 patients clinically thought to have local recurrence were treated with radiation. PSA results after therapy were available only in 17 of these 25 patients. In these 17 patients, 12 (70.6%) had decreasing PSA levels after radiation treatment, suggesting that recurrent tumor had been confined to the radiation port focused on the prostate bed; in five (29%) of 17, PSA levels continued to increase suggesting that disease outside the prostate bed had gone undetected even with fusion imaging. Thus, 12 (70.6%) of the 17 cases that could be evaluated corroborated the results after fusion.
The results of the prefusion versus postfusion lesion reviews are summarized in Table 1. The second and third columns on the table show the locations of the 161 sites identified on the prefusion review of the Prosta-Scint scans. The second column shows the locations of the 87 sites that were confirmed by fusion, and the third column shows the locations before fusion of the 74 sites that were confirmed to be false-positive results. The last column shows the locations of the 13 sites identified before fusion that were unidentified after fusion.
Discussion
In this study, Prosta-Scint SPECT scans were registered to MRI or CT volume data sets using a retrospective semiautomatic technique, with the hope that functional-anatomic fusion would increase the accuracy of the examination. Automatic image registration between anatomic volumes, particularly in the brain, has been performed successfully. The automatic fusion of structural and functional data sets presents more complications and, even in the brain, requires operator intervention [22, 23]. In part this is due to the difficulty of edge detection in functional imaging and in part due to the fact that the differences in the spatial distribution of voxel characteristics are greater between functional and anatomic images.
Although image registration capability has advanced and matured considerably, the pelvis presents serious challenges because of the abundance of blood vessels that may mimic activity in other abnormal structures or nodes and also because of the distensibility of pelvic organs, the lack of fixed anatomic positions of small bowel and redundant sigmoid colon, and the multiple degrees of freedom of motion possible in the pelvis. These challenges exist even when only CT and MRI volumes are matched [24]. For prostate cancer, using a previous method, one group [12] has outlined the vessel contours on CT and self-registered 99mTc RBC scans for the purpose of improving imaging-guided brachytherapy, but the technique is costly in terms of time and effort [25].
Automatic registration between MR images and Prosta-Scint scans using a volume-registration technique known as mutual information has also been attempted with mixed results [26]. The volume-fusion tool used in this study has proven to be suitable for routine clinical use, particularly in terms of speed and user friendliness. This tool conveniently allows the selection of landmarks in multiple planes. Thus, the user can view the physiologically and anatomically corresponding points from different points of view, often strengthening the positive identification of corresponding features. Furthermore, the incorporation of isolines into the tool adds additional information that can be useful on the actual display. The ability to merge both 2D slices and 3D volume sets to show the degree of overlap complements the isoline display [14].
Even though the present tool is partially interactive in that point pairs must be chosen, recent studies have shown that point pair selection is, in some cases, the preferred mode of volume registration [27] and that the chief objections to interactive versus automated registration (specifically inter- and intraobserver differences) are not as severe as they were once thought to be [28]. Methods that claim to be fully automatic generally require user interaction either before or during the registration process [22, 23, 29] or require a tremendous amount of prospective preparation, thus excluding retrospective correlation, especially when the region to be matched is not the brain [30].
Whereas we have not solved all possible problems, we have shown that a warping algorithm based on the information obtained from anatomically corresponding point pairs can overcome many of the difficulties encountered in registering images that are not initially well matched.
In this study, combining the Prosta-Scint scan with the MR image or CT scan primarily reduced the false-positive findings generated by physiologic uptake in normal structures (i.e., bowel, vessels, and marrow) by almost 46%. Without evidence of disease beyond the prostate bed, increasing PSA level is considered evidence of prostate bed recurrence. Thus in this group of patients, the potential impact of fusion of Prosta-Scint scanning with MRI or CT was to increase the number of patients with locoregional disease who would be eligible for radiation therapy to the prostate bed. However, if disease beyond the prostate bed is identified, systemic therapy is considered.
Adequate clinical follow-up to determine the accuracy of the fusion in predicting the extent of metastatic or recurrent disease was limited in our retrospective study. Of the 25 patients who were thought to have nodal disease before fusion and appeared to have only local recurrence after fusion, only 17 had PSA follow-up. The negative predictive value of 70.6% implied here agrees with previously reported value of 72% of Hinkle et al. [1].
Because this study compared postfusion images with prefusion reports, only sites previously identified on the original scans were reevaluated. Therefore, sites that were missed on the original interpretation and also metastases that might have arisen after the studies were completed would not have been evaluated. Also because this technique had an acknowledged overall sensitivity of 62% for disease [5], using it may have missed some foci of metastases. This possibility might account for the increasing PSA levels in the five remaining patients whose PSA levels rose despite local radiation.
Previously, fusion of Prosta-Scint images has been performed only for the purpose of improving therapeutic techniques such as brachytherapy [12, 13, 21, 26, 31] and not for primary localization of disease.
Positive findings on fused images can be reviewed with greater confidence because of the recognition and assignment of an anatomic correlate to the abnormal uptake seen on the nuclear scan. Such correlation of Prosta-Scint uptake and anatomic identification of a lymph node is highly suggestive of metastatic nodal disease and therefore has an impact on the treatment of patients with recurrent prostate cancer. Fusion appears to increase the specificity of the scintigraphic examination and therefore might foster greater confidence in the use of findings in clinical decision making (i.e., in selecting appropriate patients for locoregional radiation of the prostate bed). Proper identification of patients who will not benefit from radiation will spare them the morbidity, risk, and cost associated with radiation treatment.
Acknowledgments
We thank Vivian Lee for assisting in the interpretation of the anatomic images and Antoinette Murphy-Walcott for assisting with the image fusion.
References
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  B. Turkbey, P. S. Albert, K. Kurdziel, and P. L. Choyke Imaging Localized Prostate Cancer: Current Approaches and New Developments Am. J. Roentgenol., June 1, 2009; 192(6): 1471 - 1480. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. C. Roarke, B. D. Nguyen, and B. A. Pockaj Applications of SPECT/CT in Nuclear Radiology Am. J. Roentgenol., September 1, 2008; 191(3): W135 - W150. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. L. Speight and M. Roach III Advances in the Treatment of Localized Prostate Cancer: The Role of Anatomic and Functional Imaging in Men Managed With Radiotherapy J. Clin. Oncol., March 10, 2007; 25(8): 987 - 995. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Seo, K. H. Wong, M. Sun, B. L. Franc, R. A. Hawkins, and B. H. Hasegawa Correction of Photon Attenuation and Collimator Response for a Body-Contouring SPECT/CT Imaging System J. Nucl. Med., May 1, 2005; 46(5): 868 - 877. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. K. Bucci, A. Bevan, and M. Roach III Advances in Radiation Therapy: Conventional to 3D, to IMRT, to 4D, and Beyond CA Cancer J Clin, March 1, 2005; 55(2): 117 - 134. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
