Dose volume histogram-based optimization of image reconstruction parameters for quantitative 90 Y-PET imaging.

PURPOSE: 90 Y-microsphere radioembolization or selective internal radiation therapy is increasingly being used as a treatment option for tumors that are not candidates for surgery and external beam radiation therapy. Recently, volumetric 90 Y-dosimetry techniques have been implemented to explore tumor dose-response on the basis of 3D 90 Y-activity distribution from PET imaging. Despite being a theranostic study, the optimization of quantitative 90 Y-PET image reconstruction still uses the mean activity concentration recovery coefficient (RC) as the objective function, which is more relevant to diagnostic and detection tasks than is to dosimetry. The aim of this study was to optimize 90 Y-PET image reconstruction by minimizing errors in volumetric dosimetry via the dose volume histogram (DVH). We propose a joint optimization of the number of equivalent iterations (the product of the iterations and subsets) and the postreconstruction filtration (FWHM) to improve the accuracy of voxel-level 90 Y dosimetry. METHODS: A modified NEMA IEC phantom was used to emulate clinically relevant 90 Y-PET imaging conditions through various combinations of acquisition durations, activity concentrations, sphere-to-background ratios, and sphere diameters. PET data were acquired in list mode for 300 min in a single-bed position; we then rebinned the list mode PET data to 60, 45, 30, 15, and 5 min per bed, with 10 different realizations. Errors in the DVH were calculated as root mean square errors (RMSE) of the differences in the image-based DVH and the expected DVH. The new optimization approach was tested in a phantom study, and the results were compared with the more commonly used objective function of the mean activity concentration RC. RESULTS: In a wide range of clinically relevant imaging conditions, using 36 equivalent iterations with a 5.2-mm filtration resulted in decreased systematic errors in volumetric 90 Y dosimetry, quantified as image-based DVH, in 90 Y-PET images reconstructed using the ordered subset expectation maximization (OSEM) iterative reconstruction algorithm with time of flight (TOF) and point spread function (PSF) modeling. Our proposed objective function of minimizing errors in DVH, which allows for joint optimization of 90 Y-PET iterations and filtration for volumetric quantification of the 90 Y dose, was shown to be superior to conventional RC-based optimization approaches for image-based absorbed dose quantification. CONCLUSION: Our proposed objective function of minimizing errors in DVH, which allows for joint optimization of iterations and filtration to reduce errors in the PET-based volumetric quantification 90 Y dose, is relevant to dosimetry in therapy procedures. The proposed optimization method using DVH as the objective function could be applied to any imaging modality used to assess voxel-level quantitative information.

Comparison of Step-and-Shoot and Continuous-Bed-Motion PET Modes of Acquisition for Limited-View Organ Scans.

Continuous-bed-motion (CBM) acquisition mode has been made commercially available in PET/CT scanners. CBM mode is designed for whole-body imaging, with a long scan length (multiple axial fields of view [aFOVs]) and short acquisition duration (2-3 min/aFOV). PET/CT has recently been used after 90Y-microsphere therapy to quantify 90Y activity distribution in the liver. Here we compared counting efficiencies along the bed-motion direction (z-axis) between CBM and step-and-shoot (SS) acquisition modes for limited-view organ scans, such as 90Y PET/CT liver studies, that have short scan lengths (≤2 aFOVs) and long acquisition durations (10-30 min/aFOV). Methods: The counting efficiencies, that is, analytic sensitivities, in SS mode (single-aFOV and multiple-aFOV scans) and CBM mode were theoretically derived and experimentally validated using a cylindric 68Ge phantom. The sensitivities along the z-axis were compared between the SS and CBM modes. Results: The analytic and experimental count profiles were in good agreement, validating the analytic models. For fixed scan durations, the overall coincidence counting efficiency in CBM mode was lower (∼60%) than those in SS modes, and the maximum sensitivity in CBM mode was 50% or less of that in 1-aFOV SS mode and 100% or less of that in 2-aFOV SS mode. Conclusion: The ability of CBM mode to tailor-fit the PET/CT scan length and local scan duration is not realized in studies with a short scan length (≤30 cm) and long scan duration (20 min/aFOV for the scanner). SS acquisition mode is preferable to CBM mode for limited-view organ and count-starved scans, such as 90Y PET/CT liver scans, because of the higher counting efficiency of SS mode, which leads to better image quality and quantification precision.

Selective Internal Radiation Therapy With Yttrium-90 Glass Microspheres: Biases and Uncertainties in Absorbed Dose Calculations Between Clinical Dosimetry Models.

PURPOSE: To quantify differences that exist between dosimetry models used for 90Y selective internal radiation therapy (SIRT). METHODS AND MATERIALS: Retrospectively, 37 tumors were delineated on 19 post-therapy quantitative 90Y single photon emission computed tomography/computed tomography scans. Using matched volumes of interest (VOIs), absorbed doses were reported using 3 dosimetry models: glass microsphere package insert standard model (SM), partition model (PM), and Monte Carlo (MC). Univariate linear regressions were performed to predict mean MC from SM and PM. Analysis was performed for 2 subsets: cases with a single tumor delineated (best case for PM), and cases with multiple tumors delineated (typical clinical scenario). Variability in PM from the ad hoc placement of a single spherical VOI to estimate the entire normal liver activity concentration for tumor (T) to nontumoral liver (NL) ratios (TNR) was investigated. We interpreted the slope of the resulting regression as bias and the 95% prediction interval (95%PI) as uncertainty. MCNLsingle represents MC absorbed doses to the NL for the single tumor patient subset; other combinations of calculations follow a similar naming convention. RESULTS: SM was unable to predict MCTsingle or MCTmultiple (p>.12, 95%PI >±177 Gy). However, SMsingle was able to predict (p<.012) MCNLsingle, albeit with large uncertainties; SMsingle and SMmultiple yielded biases of 0.62 and 0.71, and 95%PI of ±40 and ± 32 Gy, respectively. PMTsingle and PMTmultiple predicted (p<2E-6) MCTsingle and MCTmultiple with biases of 0.52 and 0.54, and 95%PI of ±38 and ± 111 Gy, respectively. The TNR variability in PMTsingle increased the 95%PI for predicting MCTsingle (bias = 0.46 and 95%PI = ±103 Gy). The TNR variability in PMTmultiple modified the bias when predicting MCTmultiple (bias = 0.32 and 95%PI = ±110 Gy). CONCLUSIONS: The SM is unable to predict mean MC tumor absorbed dose. The PM is statistically correlated with mean MC, but the resulting uncertainties in predicted MC are large. Large differences observed between dosimetry models for 90Y SIRT warrant caution when interpreting published SIRT absorbed doses. To reduce uncertainty, we suggest the entire NL VOI be used for TNR estimates when using PM.

Comparing voxel-based absorbed dosimetry methods in tumors, liver, lung, and at the liver-lung interface for (90)Y microsphere selective internal radiation therapy.

BACKGROUND: To assess differences between four different voxel-based dosimetry methods (VBDM) for tumor, liver, and lung absorbed doses following (90)Y microsphere selective internal radiation therapy (SIRT) based on (90)Y bremsstrahlung SPECT/CT, a secondary objective was to estimate the sensitivity of liver and lung absorbed doses due to differences in organ segmentation near the liver-lung interface. METHODS: Investigated VBDM were Monte Carlo (MC), soft-tissue kernel with density correction (SKD), soft-tissue kernel (SK), and local deposition (LD). Seventeen SIRT cases were analyzed. Mean absorbed doses ([Formula: see text]) were calculated for tumor, non-tumoral liver (NL), and right lung (RL). Simulations with various SPECT spatial resolutions (FHWMs) and multiple lung shunt fractions (LSs) estimated the accuracy of VBDM at the liver-lung interface. Sensitivity of patient RL and NL [Formula: see text] on segmentation near the interface was assessed by excluding portions near the interface. RESULTS: SKD, SK, and LD were within 5 % of MC for tumor and NL [Formula: see text]. LD and SKD overestimated RL [Formula: see text] compared to MC on average by 17 and 20 %, respectively; SK underestimated RL [Formula: see text] on average by -60 %. Simulations (20 mm FWHM, 20 % LS) showed that SKD, LD, and MC were within 10 % of the truth deep (>39 mm) in the lung; SK significantly underestimated the absorbed dose deep in the lung by approximately -70 %. All VBDM were within 10 % of truth deep (>12 mm) in the liver. Excluding 1, 2, and 3 cm of RL near the interface changed the resulting RL [Formula: see text] by -22, -38, and -48 %, respectively, for all VBDM. An average change of -7 % in the NL [Formula: see text] was realized when excluding 3 cm of NL from the interface. [Formula: see text] was realized when excluding 3 cm of NL from the interface. CONCLUSIONS: SKD, SK, and LD are equivalent to MC for tumor and NL [Formula: see text]. SK underestimates RL [Formula: see text] relative to MC whereas LD and SKD overestimate. RL [Formula: see text] is strongly influenced by the liver-lung interface.