Room: AAPM ePoster Library
Purpose: studies have shown that our 2D Joint Statistical Image Reconstruction code using a basis-vector cross-section model (JSIR-BVM) exhibits better accuracy and noise suppression than competing dual-energy CT (DECT) methods for mapping proton stopping power ratio (SPR) on simulated and experimental axial sinograms. Towards making JSIR-BVM SPR mapping clinically feasible for proton-beam radiotherapy planning, we present our GPU-based 3D JSIR-BVM reconstruction and quantitatively assess its performance on simulated and experimentally acquired helical sinograms for phantom and patient scan subjects.
Methods: JSIR-BVM reconstructs 3D CT images of two basis-component weights from 90 kVp and 140 kVp helical transmission sinograms acquired sequentially on a Philips Big Bore scanner. These BVM weights were mapped to I-value and effective electron density, from which SPR maps were calculated via the Bethe-Bloch equation. Our synthetic and experimental phantoms contained fabricated samples of known atomic composition and density (hence known SPR) spanning the range of soft and bony tissue compositions.
Results: mean percentage SPR estimation errors averaged over the 12 inserts were 0.06% and 0.04% for 2D and 3D reconstructed SPR images, respectively. The RMS average errors were 0.08% for 2D and 0.07% for 3D reconstruction. The maximum percentage errors in SPR estimation were 0.16% for both 2D and 3D reconstructions. Our early experience with head and neck patient SPR reconstructions demonstrates reasonable (0-2%) errors with ICRU-46 tabulations. Concordance between 2D axial and 3D helical SPR maps is 0.02% and 0.006% in terms of mean and RMS average percentage error, respectively.
Conclusion: work demonstrates the feasibility of accurately reconstructing high-resolution 3D SPR maps from synthetic and experimentally acquired, spectral-hardening-uncompensated dual-energy helical sinograms exported from a commercial multislice CT scanner. Finally, for the first time, a principled dual-energy statistical image reconstruction, 3D JSIR-BVM, was able to reconstruct plausible 3D SPR volumes on clinically imaged human subjects.