Toward best practice voxel-wise 177Lu dosimetry: kernel generation, scanner characterization, and convolution-based dose calculation

Stephen Graves; Ashok Tiwari; Yusuf Menda; Mark Madsen; John Sunderland

Objectives: Patient-specific dosimetry has the potential to significantly increase the therapeutic benefit of targeted radionuclides by delivering the maximum administered activity without exceeding normal tissue toxicity limits. Voxel-based dosimetry has the potential to provide patient-specific dose volume information that is not available through conventional pre-tabulated anthropomorphic phantom-based dosimetry methods. However, challenges arise when considering the transition to voxel-wise dosimetry. Most notably among these are the partial volume effect, tissue-specific energy absorption characteristics, and density heterogeneity within the patient. The objectives of this work were to (1) validate the quantitative accuracy of 177Lu-SPECT/CT, (2) optimize the SPECT reconstruction and post-processing to minimize partial volume errors, and (3) develop an in-house algorithm for voxel-wise convolution dose calculation. Methods: Four 177Lu-containing phantoms were prepared: A uniformity phantom, a line-source, a calibrated fiducial, and the SNMMI Clinical Trials Network (CTN)-3 anthropomorphic chest phantom. At the time of scanning, the CTN-3 phantom contained 84 mCi of 177Lu with a 20:1 tumor to background ratio. SPECT/CT scans were acquired of all phantoms (Siemens Symbia T, ME collimator), and images were reconstructed by OSEM-3D (3i6s). Following reconstruction, a manual resolution recovery step was performed by blind deconvolution using the accelerated and damped Richardson-Lucy algorithm. The number of resolution recovery iterations was chosen to maximize the measured recovery coefficients within the CTN-3 phantom. Dose point kernels (DPKs) for 177Lu were generated using MCNP v6.2 in a spherical water phantom. The resultant DPKs (beta, gamma, and electron) were summed and converted to Cartesian coordinates by sampling the kernel to a high-resolution matrix, down-sampling the matrix to the SPECT resolution, followed by scaling for conservation of energy. A dose map was generated from the CTN-3 scan by convolution with this kernel, and the resultant dose map was written as an RT-Dose DICOM file for subsequent analysis. Results: For the chosen acquisition parameters, an efficiency factor of 0.127 µCi/cps was measured. Spatial resolution (FWHM) within the raw reconstructed image was 12 mm. Following resolution recovery, the spatial resolution was 8 mm, and the subsequent improvement in recovery coefficients for intermediate-sized spheres was substantial. For example, activity recovery in the 22 mm sphere increased from 41% to 80% following the resolution recovery deconvolution. Minimal Gibbs ringing artifact was observed in the resultant images, and noise was considered acceptable within a dosimetry context. The generated 177Lu Cartesian coordinate dose kernels were in good agreement with literature dose kernels. Resultant dose maps were compared against theoretical predictions for the mean dose to each sphere, and it was found that residual partial volume effect resulted in a size-dependent underestimation of mean dose. Conclusions: We have developed a dosimetry workflow that begins to address the challenges associated with voxel-wise dosimetry. This work can be immediately extended to approaches that correct for tissue-specific energy absorption and density heterogeneity, such as the collapsed-cone convolution superposition algorithms commonly employed in external-beam radiation dosimetry.

Toward best practice voxel-wise 177Lu dosimetry: kernel generation, scanner characterization, and convolution-based dose calculation

View Online

Abstract

Details

Metrics