90Y SPECT/CT quantitative study and comparison of uptake with pretreatment 99 mTc‐MAA SPECT/CT in radiomicrosphere therapy

Abstract Introduction Yttrium‐90 (90Y) microsphere post‐treatment imaging reflects the true distribution characteristics of microspheres in the tumor and liver compartments. However, due to its decay spectra profile lacking a pronounced photopeak, the bremsstrahlung imaging for 90Y has inherent limitations. The absorbed dose calculations for 90Y microspheres radiomicrosphere therapy (RMT) sustain a limitation due to the poor quality of 90Y imaging. The aim of this study was to develop quantitative methods to improve the post‐treatment 90Y bremsstrahlung single photon emission tomography (SPECT)/computed tomography (CT) image analysis for dosimetric purposes and to perform a quantitative comparison with the 99mTc‐MAA SPECT/CT images, which is used for theranostics purposes for liver and tumor dosimetry. Methods Pre and post‐treatment SPECT/CT data of patients who underwent RMT for primary or metastatic liver cancer were acquired. A Jasczak phantom with eight spherical inserts of various sizes was used to obtain optimal iteration number for the contrast recovery algorithm for improving 90Y bremsstrahlung SPECT/CT images. Comparison of uptake on 99mTc‐MAA and 90Y microsphere SPECT/CT images was assessed using tumor to healthy liver ratios (TLRs). The voxel dosimetry technique was used to estimate absorbed doses. Absorbed doses within the tumor and healthy part of the liver were also investigated for correlation with administered activity. Results Improvement in CNR and contrast recovery coefficients on patient and phantom 90Y bremsstrahlung SPECT/CT images respectively were achieved. The 99mTc‐MAA and 90Y microspheres SPECT/CT images showed significant uptake correlation (r = 0.9, P = 0.05) with mean TLR of 9.4 ± 9.2 and 5.0 ± 2.2, respectively. The correlation between the administered activity and tumor absorbed dose was weak (r = 0.5, P > 0.05), however, healthy liver absorbed dose increased with administered activity (r = 0.8, P = 0.0). Conclusions This study demonstrated correlation in mean TLR between 99mTc‐MAA and 90Y microsphere SPECT/CT.


| INTRODUCTION
Radiomicrosphere therapy (RMT) using Yttrium-90 ( 90 Y) microspheres is an effective liver-directed therapy in the management of primary and metastatic liver cancer. [1][2][3] The biodistribution of 90 Y microspheres after treatment is assessed through 90 Y bremsstrahlung imaging, preferably SPECT/CT. [4][5][6] A single time point imaging is used due to the permanent implant of the microspheres where there is no elimination, redistribution and washout phases of the radiopharmaceutical. 7 Technetium-99m macroaggregated albumin ( 99m Tc-MAA) scanning is performed prior to RMT as a surrogate for 90 Y microspheres.
In RMT planning, 99m Tc-MAA planar and single photon emission tomography (SPECT)/computed tomography (CT) imaging is used to measure the percentage of particles that shunt to the lungs (lung shunt fraction, LSF), to assess any extrahepatic particle deposition, and to calculate the tumor to normal liver ratio (TLR). 1,8 Various authors have investigated the correlation of uptake and distribution between 99m Tc-MAA and 90 Y microspheres SPECT/CT. 8,9 The authors used tumor to normal liver ratio to evaluate the uptake on 99m Tc-MAA and 90 Y microsphere SPECT/CT images. 90 Y microspheres bremsstrahlung SPECT/CT imaging could potentially identify an extrahepatic uptake. Early detection of such an uptake, thus, could initiate preventative measures early on. The tumor and liver dose estimates obtained from 90 Y microspheres bremsstrahlung SPECT/CT imaging could be correlated with tumor response and liver toxicity clinical data. 10 The major problem in 90 Y bremsstrahlung imaging is the lack of a pronounced photopeak energy due to the continuous and broad energy spectrum of bremsstrahlung photons giving it a poor image quality. As a result, various studies based on phantom and Monte Carlo (MC) simulation have recommended appropriate energy windows in accordance with the collimator used. [11][12][13][14] It has been shown, often in conjunction with phantom studies that the incorporation of MC simulations to clinical images gives an optimal accuracy of bremsstrahlung images by compensating for photons through correction for attenuation, scatter, and collimator-detector response. [13][14][15] Alternately, studies have demonstrated the feasibility of 90 Y PET/CT imaging excelling in contrast and resolution compared to bremsstrahlung SPECT/CT. 16

2.A.2 | Phantom study
The Jasczak phantom with eight fillable spherical inserts of inner diameter 2, 8, 10, 12, 16, 25, 31, and 34 mm was used. The spheres (multiple sizes, see Table 1) and background (~6 L volume) were  Table 1 shows activities within the different sphere sizes. 90 Y activity was diluted and measured in a 60 ml vial before adding to the spheres. The activity inside each sphere was measured using a Capintec dose calibrator (read out scale factor = 10). Imaging was set identical to patient studies in terms of imaging window, collimator, and image reconstruction.

2.A.3 | Image processing
A MATLAB ® algorithm was developed to import and export images for Volumes of interest (VOIs) generation, semi-automatic tumor segmentation, activity estimation, absorbed dose estimation, and statistical and mathematical analysis. For the patient studies, VOIs were drawn manually on the CT image slices. Binary masks from the CT images were mapped onto the respective SPECT scans. For the phantom study, eight circular VOIs were manually drawn on the CT slices, where knowledge of the phantom composition allowed us to identify the spheres and make sure the volumes in VOIs were consistent with the true measured sphere volumes. Background VOIs consisted of all voxels within the phantom boundary excluding voxels that belong to the spheres' VOIs.

2.B | Contrast recovery
A is the reconstructed image degraded by the response of the detector, A' is the contrast recovered image where A 0 0 = A, ⊗ is the 3D (x, y, z) convolution operation and i is the iteration number. In our study the PSF was fixed and the only iterative maximum likelihood estimate was the image. The PSF of the collimator-detector response was modeled by the Gaussian function according to Eq. (2) where σ was found from the relationship with full-width at half-maximum PSFðx; y; zÞ ¼ 1 Improvements in the quantitative quality 90 Y bremsstrahlung SPECT/CT images were evaluated using contrast to noise ratio (CNR) and contrast recovery coefficients (Q H ) 16 for the patient and phantom studies respectively as given by Eqs. (3) and (4). Our analysis of quantitative image improvement is based on matched VOIs between the CT and SPECT images as this is required for dosimetry estimation.
M T is the mean count in tumor VOIs, M B is the mean count in healthy liver VOIs, C S is the mean count in the sphere VOIs, C B is the mean count in the background VOIs and R is the true sphere to background ratio. The block diagram of the employed RL algorithm is shown in Fig. 1.
The iteration number for the algorithm was chosen at the point of maximum likelihood where the contrast recovery coefficient for the 34 mm sphere was at its maximum value, which corresponded to a decline in the associated root mean square error (RMSE) between two consecutive iterative image estimates, which was found to be at the sixth iteration as shown in Fig. 2. The contrast improvement algorithm using Richardson-Lucy deconvolution is a new idea integrated in this approach.

2.C | SPECT calibration factor measurement
The calibration factor (CF) was defined as the ratio of the total reconstructed counts to the true activity. anatomical and functional images, and we found visually the best results using SPM. 22 We used the spatial normalization function of SPM12 using the standard settings of the toolbox. In order to choose the best registration approach, we calculated the mutual information between the co-registered 90 Y microsphere and 99m Tc-MAA SPECT images from the methods. The mutual information between the two images is defined as: IðA; BÞ ¼ HðAÞ þ HðBÞ À HðA; BÞ

2.G | Absorbed dose estimation
The voxel S-value method was used to estimate 3D radiation absorbed dose in 90 Y bremsstrahlung SPECT/CT images. 25,26 Cubical D T is the absorbed dose at the target voxel (mGy), A (MBq) is the cumulated activity from the surrounding source voxels, ⊗ is the 3D convolution and S is the voxel S-value (mGy/MBq) for each associated source distance to the target voxel. Cumulative dose-volume histograms (cDVHs) and isodose curves were generated for the tumor and healthy liver VOIs from the SPECT dose map images.

2.H | Statistical analysis
Quantitative parameters are presented as mean ± SD and ranges.
Linear regressions were generated between administered activities (independent variable) and cps for the purpose of predicting CFs. We report slope, R 2 , standard error and 95% CI of the regression models. Pearson correlation coefficient and its P value were used to test for significance of correlations between TLRs from 99m Tc-MAA and 90 Y microsphere SPECT/CT as well as between administered activity and absorbed doses. Statistical analyses were deemed significant as having a P < 0.05. All statistical analyses were performed with Minitab ® software package (version 17).

Results of the regression analyses of the relationships between cps
and administered activity gave the following results, the slopes being

3.C | Activity estimation and comparison with the administered dose
The total activity inside the liver was estimated for each patient using the two CFs. For CF 1, total liver activity estimation resulted in mean percent error of 59 ± 5%. Applying CF 2 gave the smallest error (−5 ± 13%), thus it was used for subsequent analysis. Table 2 shows results of total activity estimation within the liver VOIs using CF 2 . For most of the patients the estimated total liver activity percent errors were within ±10% giving an overall satisfactory results. As our study is retrospective based on anonymized data, we couldn't provide possible clinical reasons for the larger deviations of the estimated activities in some of the patients.
Results of activity estimates inside the spheres and the background for the phantom study gave total mean percent error of −23 ± 41%. comparable to the aforementioned result [ Fig. 4(c)]. But the mutual information is higher or equal in most patients where the CT scan was the reference image [ Fig. 5(a)]. In other patients, for example, patient 7, the 99m Tc-MAA and 90 Y microspheres SPECT images didn't co-register correctly with the CT image. From the SPECT scans of this patient, we observed that the patient had a hepatic tumor with a necrotic core with minimal uptake inside the liver which reduced the mutual information required for co-registering the SPECT and CT scans [ Fig. 5 Table 3).

3.E | Tumor segmentation and TLR comparison
The total mean TLR was 9.4 ± 9.2 and 5.0 ± 2.2 on 99m Tc-MAA and 90 Y microsphere SPECT/CT, respectively. Figure 7 shows the scatterplot of mean TLRs from the two images displaying a significant correlation (r = 0.9, P = 0.00). From the plot, one patient appears to be an outlier (Grubbs' outlier test, P = 0.00), and taking this patient out of the analysis gave a reduced correlation (r = 0.6, P < 0.05).

| DISCUSSION
The primary objective of this study was to develop a post-recon- Dosimetry of 90 Y microsphere distribution has the benefit of providing the real dose-response relationship for further treatment if required, instead of the predicted dose-response using 99m Tc-MAA.