During radiation therapy, appropriate radiation is fired from various positions around the patient. Radiation therapy planning determines the values of the radiation at various positions. This optimization problem requires a huge matrix, whose elements are the doses deposited by unit strength beamlets at all possible positions around the patient at all the voxels of the patient volume. Thus each pair of beamlet and voxel has an entry in this matrix. This matrix is generally referred to by the following names: dose deposition matrix/dose influence matrix/dose information matrix.
To get an idea about what algorithms can be used for dose calculation, this paper is good starting point: Monte Carlo- versus pencil-beam-/collapsed-cone-dose calculation in a heterogeneous multi-layer phantom
To get an idea about how this matrix is subsequently used in radiation therapy, one could look at: Real-Time Radiation Treatment Planning with Optimality Guarantees via Cluster and Bound Methods
The main bottleneck in radiation therapy planning is the computation of this matrix. While searching the literature, it is hard to find ideas from computer graphics (especially, rendering) being used to accelerate the computation of this matrix. Why is it so? Could it be just a lack of communication between the two communities? It appears unlikely, though. The radiation in radiation therapy is different from visible light. However, is it possible to borrow ideas from participating media rendering?