ROI reconstruction from truncated cone-beam projections (Q1673801)

Region-of-interest tomography is used to reduce the radiation exposure in computed tomography by truncating the projections so that only rays intersecting a small region of interest are aquired. In this paper, the authors thus consider rays \(\{a+t\theta\,:\, \theta>0\}\) with sources \(a\) on a smooth curve \(\Gamma \subset \mathbb R^n \) outside a fixed open ball \(B\) that contains the spherical region of interest \(C\). It is assumed that each plane intersecting \(B\) also intersects the curve \(\Gamma\) non tangentially at least once and that the unknown density \(f\) that one wants to reconstruct belongs to the Sobolev space \(W^5\) and has compact support in \(B\). If the cone beam operator is not restricted to the beams intersecting the region \(C\) there is an inverse and using this inverse the authors construct an iterative procedure that in the limit, with exponential convergence, gives a function \(\hat f\) satisfying \(\|\hat f-f\|\leq \epsilon \|f\|_{W^5(B)}\) provided the radius of \(C\) is greater than some critical radius. There is no explicit restriction of the location of the region of interest \(C\) and the numerical tests performed by the authors show that the critical radius is relatively small compared to the size of the support of \(f\).