A fast parallel method for medical imaging problems including linear inequality constraints

When studying problems such as tomography with bounded noise or IMRT, we need to solve systems with many linear inequality constraints. Projection-based algorithms are often used to solve this kind of problem. We see how previous work for accelerating the convergence of linear algorithms can be recast within the most recent generic framework, and show that it gives better results in specific cases. The proposed algorithm allows general convex constraints as well and the conditions for convergence are less restrictive than tradition- nal algorithms. We provide numerical results carried out in the context of tomography and IMRT.