A maximum-a-posteriori EM reconstruction method based on total variation regularization for Compton camera imaging

International audienceDose monitoring is a key issue to improve the quality of the treatment in proton therapy. Detection of the prompt-gamma rays produced during the treatment with a Compton camera might be a solution to meet this objective. Compton camera imaging requires to reconstruct the image of the source with specific algorithms. The reference method for gamma-ray tomography is the list-mode maximum likelihood expectation maximization (LM-MLEM).An important issue is the high level of noise in the reconstructed images. This noise increases as the image of the source gets more precise over iterations. The low number of counts and the acquisition uncertainties faced in prompt-gamma imaging suggest the use of a priori information. We recently developed a maximum-a-posteriori EM algorithm based on total variation regularization which is particularly well suited for low-dose acquisitions. Specifically designed for Poisson noise, this algorithm allows to reduce the noise uniformly in the image, as it naturally adapts to the intensity-dependent variance. Moreover, the algorithm is faster than the methods based on splitting approaches.We simulated a box source with non-uniform intensity and camera geometry depicted in figure 1. For the reconstruction we used 20,000 detected events corresponding to the expected number of counts for the acquisition of a single spot in proton therapy (10^8 protons). Results show that the TV a priori strongly reduces the noise and facilitates the reconstruction of the prompt-gamma distribution