Y-90 SPECT maximum likelihood image reconstruction with a new model for tissue-dependent bremsstrahlung production

Objectives: Y-90 SPECT quantification is complicated by material dependence of the bremsstrahlung generation. The probability of bremsstrahlung production is proportional to the square of the atomic number of the absorber. Hence, the same amount of activity in different tissue regions of the body will produce different numbers of bremsstrahlung photons. However, existing reconstruction methods disregard this tissue-dependent probability of bremsstrahlung production, which can impact quantification accuracy when imaging heterogeneous regions such as marrow near bone. We propose a new maximum likelihood method that accounts for this dependency, enabling images of the desired ‘Y-90 distribution’ to be reconstructed instead of the ‘bremsstrahlung distribution’ that is obtained with existing methods.

Methods: To assess the tissue dependence, we first performed EGS5 MC simulations for bone, marrow, ICRP tissue, and lung in infinite media. The simulations show that at typical Y-90 measurement energies bremsstrahlung production is 1.7 to 2.1 times as high in bone as in other tissues. Subsequent simulations of bone and marrow cavities of various sizes verified that bremsstrahlung production of photons above 100 keV was roughly 1.8 times production in pure marrow, almost independent of the cavity geometry. We incorporated the tissue-dependent probability into the image reconstruction method by adding a term to the system matrix (i.e., projector/backprojector) to model the bremsstrahlung spectra produced in each voxel as a bone-volume fraction (BVF) weighted mixture of the bone-only and tissue-only spectra. We used CT information to determine the BVF of each voxel in SPECT. To validate the efficacy of the system matrix, we used XCAT phantoms and investigated the region from spine to upper femur. We simulated different uptake ratios (1:1, 3:1, 1:3) for bone:marrow. Images were reconstructed with in-house developed 3D OS-EM including attenuation, collimator detector response and with and without tissue-dependent probabilities.

Results: We compared SPECT reconstructions using the new model and standard model. Qualitatively, visual comparison showed better agreement between SPECT and the true activity map when the system model for the reconstruction included the tissue dependent probability. Quantitatively, we calculated the count recovery between the SPECT image and the true activity map in volumes of interest in (1) lumbar (2) pelvis and (3) upper femur, and the new model led to reconstruction closer to the true image in marrow regions (-11.0 - 48.5%, mean 19.8%) and in bone regions (-20.0 - 50.7%, mean 14.2%) than the reconstruction using the standard model.

Conclusion: Y-90 SPECT reconstruction in heterogeneous regions is enhanced by accounting for tissue-dependent bremsstrahlung production. Research Support: NIH R01EB022075