Introduction

Intrahepatic arterial ⁹⁰Y therapy of unresectable colorectal liver metastases has been extensively described in the literature (1-5). Recently, radiation safety considerations in ⁹⁰Y therapy were also addressed (6) as well as evaluation by fluorine-18 (18F)-fluorodeoxyglucose (FDG) PET of response to ⁹⁰Y therapy (7). In our institution we are using commercially available SIR-Spheres (Sirtex Medical) resin microspheres, polymer beads designed to be between 20 and 60 µm in diameter and loaded with ⁹⁰Y at a specific activity of 40–70 Bq per sphere (8, 9). Prior to ⁹⁰Y therapy, however, the ⁹⁰Y activity, which will be administered and the percentage of shunting to the lungs must be determined, most commonly, by MAA gamma camera imaging. In ⁹⁰Y therapy, it is assumed that ⁹⁰Y microspheres (⁹⁰Y) will follow the MAA distribution. But, due to the differences in catheter placement and injection techniques as well as physical differences between ⁹⁰Y and MAA, the distributions ⁹⁰Y and MAA in vivo may differ, significantly affecting the reliability of the calculated percent lung shunting and of treatment planning overall.

The objective of the current retrospective study was to identify and evaluate an objective measure of the correlation of ⁹⁰Y and MAA SPECT images in order to better understand under what clinical conditions ⁹⁰Y and MAA distributions are similar or different. We also believe that a good correlation of ⁹⁰Y and MAA distributions is generally associated with a good therapeutic response, while a poor correlation is associated with an unfavorable response. To systematically test this hypothesis, however, an objective measure of the correlation of the ⁹⁰Y and MAA distributions is needed.

Materials and Methods

Acquisition and reconstruction protocols

A dual-head Infinia SPECT/CT gamma camera (GE Medical Systems, Milwaukee, WI) was used in this study. Twenty patients have been investigated in a continuing study. First, a MAA study was performed followed by ⁹⁰Y SPECT/CT acquisition with a time interval (mean +/– standard deviation) of 23 +/– 4 days between the scans. The MAA study consisted of two acquisitions. After administration of 185 MBq of MAA, a partial-body scan from head to mid thigh was performed in continuous scan mode with table speed of 10cm/min. This scan is called the nuclear medicine breakthrough scan and is performed in all patients to determine the percentage of the dose that shunts from the liver to the lungs. Immediately after the breakthrough scan, a SPECT/CT MAA acquisition over the liver area was performed. The matrix size was 128 × 128, the pixel size was 4.4 mm, there were 60 projection images, and time per projection was 30 seconds. The CT scan was performed as follows: 10mm axial sampling, 140 kpV, 2.5 mA, 2.6 rpm and 256 × 256 matrix size. For MAA acquisitions, a low-energy-high- resolution (LEHR) collimator was used and the energy window set at 140 keV ± 10%. The percent lung shunting is calculated from the geometrically averaged anterior and posterior images from the breakthrough scan as a ratio of the lung counts and the summed counts in the lungs, liver and tumor (9). The percent lung shunting may alter the activity that can be safely infused commensurate with acceptable risk of radiation pneumonitis. If the percent lung shunting is less than 10%, there is no reduction in ⁹⁰Y activity. If the percent lung shunting is in the range 10%-15%, there is a 20% reduction in ⁹⁰Y activity and if the percent lung shunting is in the range 15%-20%, there is a 40% reduction in ⁹⁰Y activity. For patients with lung shunting higher than 20%, the ⁹⁰Y therapy is not performed (9).

However, while the breakthrough scan is used to determine the lung shunting and accordingly the activity limitations described above, the SPECT/CT MAA scan is used to calculate ⁹⁰Y activity which will be subsequently administered as therapy dose. In our center, the ⁹⁰Y activity is calculated using the body surface area (BSA) method (9). The BSA (m²) is calculated from a patient’s weight and height according to the standard formula (9), while the liver size, its fraction that is perfused and the tumor volume is calculated using SPECT/CT MAA imaging. The ⁹⁰Y activity in GBq is give as (BSA-0.2) (V_tumor/ V_tumor+ V_{normal liver}) (ref. 9, Section 7.1.1.2), where Vtumor and Vnormal liver are volume of the tumor and normal liver, respectively. In all our twenty patients the lung shunting was less than 10% and the full ⁹⁰Y activities were used. These activities ranged from 777-2442 MBq of ⁹⁰Y depending on patient’s BSA and liver and tumor size determined from the SPECT/CT MAA studies.

Quantitative bremsstrahlung imaging is very difficult due to scatter, septal penetration, the continuous nature of the bremsstrahlung energy spectrum, and inefficient bremsstrahlung production. Previously (10), it was suggested that a medium-energy general-purpose (MEGP) collimator, and a very broad energy window of 55-285 keV will provide optimal ⁹⁰Y images. Currently, however, most centers, including ours (11), use an MEGP collimator and energy window of 90 keV ± 15%. Our initial ⁹⁰Y studies on eight patients were performed with a LEHR collimator and an energy window of 70 keV ± 15% and MEGP collimation was used subsequently. A phantom study using two fillable spheres filled with 18.5 MBq of ⁹⁰Y was acquired along with five patient studies both ways, i.e., comprised of the standard protocol with MEGP collimation and energy window of 90 keV ± 15% and with LEHR collimation and energy window of 70 keV ± 15%. Our conclusion was that the ⁹⁰Y images obtained with MEGP collimation, assessed visually, looked better and had more counts than those obtained with LEHR collimation. However, the differences between these two acquisitions were not great enough to compromise ⁹⁰Y bremsstrahlung imaging. For the ⁹⁰Y SPECT studies, 60 views at 30 s per view were acquired and images were reconstructed using an ordered subset expectation maximization (OSEM) algorithm. Attenuation correction was performed using the CT scan, the reconstruction matrix size was 128 × 128 and the pixel size was 4.4 mm. The CT was performed as follows: 10mm axial sampling, 140 kpV, 2.5 mA, 2.6 rpm and 256 × 256 matrix size.



Figure 1: Matching CT images from 90Y and MAA SPECT/CT studies. The first row shows CT images corresponding to the 90Y microspheres and the second row CT images corresponding to the MAA study. The third row shows the registered and fused (overlaid) CT images.

Image coregistration and matching

First, the CT studies for the ⁹⁰Y and MAA SPECT/CT studies (Figure 1) were registered using a rigid-body transform based on normalized mutual information (12) and using PMOD software, version 2.9 (13). The transform thus derived was saved and used to align the ⁹⁰Y and the MAA SPECT images (Figure 2). In both transforms, the ⁹⁰Y data were fixed and MAA data were transformed to match the ⁹⁰Y data. In the next part of the analysis, a program written in IDL (Interactive Data Language, version 6.2, Research Systems, Inc.) was used. To reduce the background and concentrate on liver activity only, a threshold of 10% and 30% of the maximum activity was used for reoriented MAA and ⁹⁰Y images, respectively. These thresholds were found empirically by interactively using the IDL program and comparing the liver area in corresponding CT images. Both reoriented MAA and ⁹⁰Y images were normalized to the same maximum value. In order to investigate the correlation between 90Y and MAA activities, for the SPECT 90Y and reoriented MAA data, the Spearman’s (rho) rank correlation (14) between the two images was calculated. In addition, image distance (L2 norm) was also calculated. This was done on a voxel-by-voxel basis. For each voxel xijz in MAA image x the corresponding voxel yijz in 90Y image y was used, where i was the row index, j the column index and z was the slice index. For i and j, the range was 1-128 and for z the range was usually 1-20, in order to cover the whole liver. For example, the image distance was calculated as d(x, y) = sqrt (Σ_ijz (x_ijz- y_ijz)²). Thus, we were able to obtain the full 3D image Spearman’s rank correlation and distance. The optimal value for the Spearman’s rank correlation is 1.0, and the optimal value for L2 distance is 0.0. We tested our IDL program by using the same image twice, i.e., calculating Spearman’s rank correlation and distance between two identical 3D images and obtaining these optimal values. In reality, of course, these values are not possible to achieve, but there were several cases with relatively high Spearman’s rank correlation and relatively low L2 distance.

The visual assessment of ⁹⁰Y and MAA activities was performed in the following way (15, 16). For the five liver segments, lateral left, medial left, caudate, anterior right and posterior right liver segment, ⁹⁰Y and MAA uptake were graded on a scale of 0-3, 0 annotating a lack of activity and 3 annotating a very high uptake. For each segment, an absolute difference between MAA and ⁹⁰Y uptake grades were obtained. If the segment had similar or the same MAA and ⁹⁰Y uptake, the absolute difference was 0, indicating high correlation of MAA and ⁹⁰Y activities in that region. If one tracer was absent completely in a particular segment and if the other tracer had a very high uptake, the absolute difference was 3. These absolute differences were summed over all five segments and a visual score for the entire liver was created for each patient. The visual score was than compared with the Spearman’s rank correlation between two images and image L2 distance, defined above. Theoretically, the visual score can have a range from 0-15, 0 indicating uptake matching in all five segments, and 15 corresponding to an extreme situation where in all segments one tracer had maximal uptake and the other tracer did not have any uptake. For our twenty patients the visual scores ranged from 0 to10.



Figure 2: Registered SPECT images from ⁹⁰Y microspheres and MAA SPECT/CT studies, aligned using the transform obtained for registration of the corresponding CT images shown in Figure 1. The first row shows ⁹⁰Y microspheres images, the second row MAA images, and the third row registered and fused images. This is an example of good correlation between ⁹⁰Y microspheres and MAA activities.

The phantom study

In order to evaluate the accuracy of the registration process we used a Jaszczak phantom (Deluxe Jaszczak Phantom™, Data Spectrum Corporation, Hillsborough, NC). The phantom was filled with water and two activity-containing spheres, both 2 cm in diameter and each containing 37 MBq of ^99mTc (in the first study) and 37 MBq of ⁹⁰Y (in the second study) were placed in the phantom. The second study was performed a week later (Figure 3). These phantoms studies were first used to verify the accuracy of alignment of the SPECT and CT images in the SPECT/CT acquisition. The second purpose of the phantom studies was to assess aligning between ⁹⁰Y and ^99mTc SPECT distributions. The average offset was calculated by using the center-of-gravity method, which finds the center of intensity distribution in an image by averaging pixel positions weighted by their intensities. This is a standard image processing method and can easily be implemented in the IDL programming environment.

Results

In the phantom study (Figure 3) the offset between SPECT and CT images in the SPECT/CT acquisition was 0.5 mm ± 0.3mm (mean ± SD) for both ⁹⁰Y and ^99mTc studies. As in the patient studies, after aligning the CT images, the transform was saved and used to align corresponding SPECT ⁹⁰Y and SPECT ^99mTc hot spheres images. The offset between the centers of the spheres measured for 3 adjacent slices, was 0.7 mm ± 0.4mm (mean ± SD), showing very good alignment. However, the alignment of patient studies was slightly poorer, presumably due to breathing motion and differences in positioning and/or posture between studies.

Figure 4 shows the results of linear regression analysis between visual scoring and A) the Spearman’s rank correlation and B) L2 distance, respectively. The Spearman’s rank correlation values were in the range of 0.451 to 0.818. The L2 distance between ⁹⁰Y and MAA images were mostly grouped between 0.626-1.5 with one distinct outlier of 2.889. Visual inspection of the ⁹⁰Y and MAA distributions showed reasonable match between ⁹⁰Y and reoriented MAA SPECT images. As mentioned before, visual scoring ranged 0-10. There were a few cases with relatively high Spearman’s rank correla­tion of 0.75 and slightly above. Such a case is shown in Figure 2 where there is a good match between ⁹⁰Y microspheres and MAA distributions with high Spearman’s rank correlation of 0.79, visual score of 2 and L2 distance of 0.77. On the other hand, Figure 5 shows a patient with a poor match between ⁹⁰Y microspheres and MAA distributions with low Spearman’s rank correlation of 0.45, poor visual score of 10 and high L2 distance of 1.28. For linear correlation between visual scoring and the Spearman’s rank correlation (Figure 4A) the slope was a = ­16.45, the intercept b = 14.46 and coefficient of regression r = 0.65. The values of the same parameters for linear correlation between visual scoring and L2 distance (Figure 4B) were a = 3.19, b = 0.231 and r = 0.61. The result shows that the regression coefficient was lower for correlation between visual scoring and L2 distance comparing with the same results for the Spearman’s rank correlation and visual scoring.

Discussion

In ⁹⁰Y therapy, the administered activity and the estimated radiation doses to the tumor and to the liver are derived from SPECT/CT MAA study, based on the assumption that the ⁹⁰Y distribution will be the same as that for the MAA. However, the data presented show that there is a range of correlations between ⁹⁰Y and MAA activity distributions. The reason for poor correlation between these activity distributions may be caused by several factors, including differences in catheter position, injection techniques, particle sizes, flow between ⁹⁰Y spheres and MAA activity, progress of disease between MAA and ⁹⁰Y studies, and difference in energy radiation between ⁹⁰Y (bremsstrahlung) and ^99mTc. The implication of these findings is that the response to therapy may not be accurate if it is based on MAA distribution. Although biologically different tumors will have variable response to 90Y therapy, if we want to investigate the therapy response as only a function of MAA distribution, then an objective correlation parameter for MAA and ⁹⁰Y distributions is important.



Figure 3: A phantom study. First column shows results of matching CT (first row) and SPECT (second row) images in ⁹⁰Y microspheres SPECT/CT study. The fused SPECT/CT image is in the third row, first column. Second column shows matching of CT images of the phantom obtained in ⁹⁰Y study (first row) and ^99mTc study (second row). The fused CT/CT image is in the third row, second column. Third column shows matching of SPECT ⁹⁰Y (first row) and SPECT ^99mTc (second row) images. The fused SPECT/SPECT image is in the third row, third column.

The comparison of the ⁹⁰Y and MAA distributions strongly depends on the accuracy of matching the ⁹⁰Y and MAA images in all 3 dimensions. In three cases, due to different positioning of the patient between MAA and ⁹⁰Y studies, the automatic rigid transform could not provide very accurate matching between CT images. However, in these cases we focused on manually matching the liver in both CT images, that is, those corresponding to MAA and ⁹⁰Y studies, respectively. The other problem in matching CT images can be due to breathing motion and although the corresponding CT images are very close, there can be small differences in the dome of the liver. It is also possible to define liver volumes of interest (VOIs) in CT images and apply the VOIs thus derived to the corresponding ⁹⁰Y and MAA SPECT images and analyze the spatial correlation of the ⁹⁰Y and MAA SPECT images based on voxels within the liver VOIs. The thresholds of 10% for MAA studies and 30% for much noisier ⁹⁰Y bremsstrahlung studies seem to provide good MAA liver images but quite noisy ⁹⁰Y bremsstrahlung images, with some activity appearing to be located outside of the liver, as shown in Figures 2 and 5.



Figure 4: Linear correlation between visual scoring and A) the Spearman’s rank correlation and B) L2 distance between 90Y and MAA images. The blue symbols represent individual patient’s Spearman’s rank correlation results as function of the visual scores (Figure A) and individual patient’s L2 distance or norm results as function of the visual scores (Figure B), respectively. The purple lines represent linear fit. The linear coefficient of regression was 0.65 and 0.61 for the Spearman’s rank correlation and L2 distance, respectively.

In the current manuscript, in addition to providing visual assessment of the differences among ⁹⁰Y and MAA distributions as determined by SPECT, objective (i.e., quantitative) parameters expressing the relative similarity of the SPECT-derived ⁹⁰Y and MAA distributions were provided and compared. More objective parameters would lead to a more uniform approach to ⁹⁰Y microsphere therapy within institutions and among different centers. This would also facilitate comparison of the ⁹⁰Y microsphere therapy among different centers. A more uniform approach in ⁹⁰Y microsphere therapy would be beneficial for patients.



Figure 5: This is an example of poor correlation between 90Y microspheres and MAA activities. The first row shows 90Y microspheres images, the second row MAA images and the third row registered and fused images.

Conclusions

The results of the study demonstrate that ⁹⁰Y microspheres and MAA activities show a range of correlations among themselves, from highly correlated to very poor correlation. We proposed two parameters to objectively measure corre­lation between ⁹⁰Y microspheres and MAA images, the Spearman’s rank correlation and L2 distance. The results of our study indicates that the Spearman’s rank correlation value is a better parameter than image distance for correlating the ⁹⁰Y microspheres and MAA distributions and that it can be used as an objective parameter. In retrospective studies, the objective parameter of the correlation between ⁹⁰Y microspheres and MAA activities can be used to find under which clinical conditions MAA and ⁹⁰Y microspheres distributions are similar or different and therefore improve ⁹⁰Y microspheres therapy in the treatment of cancer meta­stases in the liver or primary liver cancer.

References

1. Ariel, I. M., Pack, G. T. Treatment of inoperable cancer of the liver by intra-arterial radioactive isotopes and chemotherapy. Cancer 20, 793-804 (1967).
2. Harbert, J. C., Robertson, J. S., Held, K. D. Nuclear Medicine Therapy. New York, NY: Thieme Medical Publishers. 239-256 (1987).
3. Gray, B., Van Hazel, G., Hope, M., Burton, M., Moroz, P., Anderson, J., et al. Randomized trial of SIR-spheres plus chemotherapy vs chemotherapy alone for treating patients with liver metastases from primary large bowel cancer. Ann Oncol 12, 1711-1720 (2001).
4. Salem, R., Thurston, K. G. Radioembolization with yttrium-90 microspheres: a state-of-the-art brachytherapy treatment for primary and secondary liver malignancies: part 3—comprehensive literature review and future direction. J Vasc Interv Radiol 17, 1571-593 (2006).
5. Gulec, S. A., Fong, Y. Yttrium 90 microsphere selective internal radiation treatment of hepatic colorectal metastases. Arch Surg 142, 675-682 (2007).
6. Gulec, S. A., Siegel, J. A. Posttherapy Radiation Safety Conside­rations in Radiomicrosphere Treatment with ⁹⁰Y-Microspheres. J Nucl Med 48, 2080-2086 (2007).
7. Wong, C. O., Salem, R., Qing, F., Wong, K. T., Barker, D., Gates, V., et al. Metabolic Response After Intraarterial ⁹⁰Y-Glass Microsphere Treatment for Colorectal Liver Metastases, Comparison of Quantitative and Visual Analyses by 18F-FDG PET. J Nucl Med 45, 1892-1897 (2004).
8. Sirtex Medical. SIR-Spheres (yttrium-90 microspheres) (package insert). Available at, http,//www.sirtex.com/files/US20Package20 Insert1.pdf. Accessed 1 December 2009.
9. Sirtex Medical. SIR-Spheres training program for physicians and institutions. Available at http,//www.sirtex.com/files/TRN-US-0320 for20US1.pdf. Accessed 1 December 2009.
10. Shen, S., DeNardo, G. L., Yuan, A., DeNardo, D. A., DeNardo, S. J. Planar gamma camera imaging quantitation of Yttrium-90 bremsstrahlung. J Nucl Med 35, 1381-1389 (1994).
11. Mansberg, R., Sorensen, N., Mansberg, V, Van der Wall, H. Yttrium 90 Y bremsstrahlung SPECT/CT scan demonstrating areas of tracer/tumor uptake. Eur J Nucl Med Mol Imaging 34, 1887 (2007).
12. Viola, P., Wells, W. Alignment by Maximization of Mutual Infor­mation. Inter J Comp Vision 24, 137-154 (1997).
13. PMOD Technologies Ltd, http,//www.pmod.com/technologies/index.html. Accessed 15 April 2009.
14. Myers, J. L., Well, A. D. Research Design and Statistical Analysis (second edition ed.). Lawrence Erlbaum. pp. 508 (2003).
15. Machac, J., Weintraub, J., Nowakowski, F. S., Mobley, D., Zhang, Z., Warner, R. P. Variations in liver perfusion patterns with liver tumors undergoing therapy with yttrium-90 microspheres studied with SPECT/CT (abstract). J Nucl Med 48 (Suppl), 396-397 P (2007).
16. Machac, J., Weintraub, J., Nowakowski, F. S., Mobley, D., Zhang, Z., Warner, R. P. Does the perfusion imaging of the liver with Tc-99m MAA reliably predict the pattern of therapeutic deposition of Yttrium- 90 SirSpheres? (abstract). J Nucl Med 48 (suppl), 399-400 P (2007).