A comparison of direct reconstruction algorithms in proton computed tomography

Area of Science:

• Medical Imaging
• Computational Physics
• Radiological Sciences

Background:

• Proton computed tomography (pCT) imaging requires algorithms that account for non-linear proton trajectories.
• Accurate image reconstruction is crucial for clinical applications of pCT.

Purpose of the Study:

• To compare the spatial resolution and relative stopping power (RSP) accuracy of five direct reconstruction algorithms for pCT.
• To evaluate algorithm performance under simulated realistic tracking uncertainties.

Main Methods:

• Simulated a single-tracking pCT setup using Geant4/GATE with proton sources and detectors.
• Evaluated five algorithms: distance-driven (DD), maximum likelihood (ML), backproject-then-filter (BTF), differentiated backprojection (DBP), and directional ramp filter (DR).
• Assessed spatial resolution using modulation transfer function and RSP accuracy using a Gammex phantom, incorporating tracker uncertainties.

Main Results:

• Directional ramp filter (DR) and BTF showed slightly better spatial resolution with ideal trackers.
• With realistic trackers, DR, BTF, DD, and DBP exhibited comparable spatial resolution.
• DR, DD, and BTF achieved mean RSP errors around 0.1%; DBP had <0.2% error.

Conclusions:

• DD and DBP algorithms provide efficient pCT image reconstruction with competitive spatial resolution and RSP accuracy.
• DD offers similar performance to more computationally intensive methods (DR, BTF) with reduced reconstruction time.
• DBP is a viable option for reduced reconstruction time, especially with truncated data, and offers good accuracy.