Feriel Khellaf

Séminaires Jeunes docteurs du GDR du 12 avril 2021

Title: List-mode proton CT reconstruction

Abstract:

Proton therapy is used for cancer treatment to achieve better dose conformity by exploiting the energy-loss properties of protons. Proton treatment planning systems require knowledge of the stopping-power map of the patient’s anatomy to compute the absorbed dose. In clinical practice, this map is generated through a conversion from X-ray computed tomography (CT) Hounsfield units to proton stopping power relative to water (RSP). This calibration generates uncertainties as photon and proton physics are different, which leads to the use of safety margins and the reduction of dose conformity. In order to reduce uncertainties, proton CT (pCT) was proposed as a planning imaging modality since the reconstructed quantity is directly the RSP. In addition to energy loss, protons also undergo multiple Coulomb scattering (MCS) inducing non-linear paths, thus making the pCT reconstruction problem different from that of X-ray CT and degrading spatial resolution. The use of a most likely path (MLP) formalism for protons to account for the effects of MCS has improved the spatial resolution in pCT, although this formalism assumes a homogeneous medium.
The objective of this work was to improve image quality of pCT list-mode reconstruction.  First, we study the accuracy of the MLP formalism in heteregeneous media by comparing the theoretical MLP against Monte Carlo generated proton paths. Results in terms of spatial, angular, and energy distributions were analyzed to determine the maximum systematic error on the MLP and assess the impact on reconstruction.
The MLP formalism provides an additional information to the MLP estimate, which is the uncertainty envelope around the MLP. This information, included in a reconstruction algorithm, could help improve spatial resolution. In addition to MCS, the resolution of the trackers used to measure the protons' position and angle has also an impact on spatial resolution. We propose a deconvolution method using the uncertainty on the MLP estimate and the tracker resolution to improve the spatial resolution of pCT images. Results on simulated data show an improved spatial resolution in simple phantoms as well as anthropomorhpic phantoms.