What is the optimum leaf width of a multileaf collimator?

The following question is investigated: How narrow do the leaves of a multileaf collimator have to be such that further reduction of the leaf width does not lead to physical improvements of the dose distribution. Because of the physical principles of interaction between radiation and matter, dose distributions in radiotherapy are generally relatively smooth. According to the theory of sampling, the dose distribution can therefore be represented by a set of evenly spaced samples. The distance between the samples is identified with the distance between the leaf centers of a multileaf collimator. The optimum sampling distance is derived from the 20% to 80% field edge penumbra through the concept of the dose deposition kernel, which is approximated by a Gaussian. The leaf width of the multileaf collimator is considered to be independent from the sampling distance. Two cases are studied in detail: (i) the leaf width equals the sampling distance, which is the regular case, and (ii) the leaf width is twice the sampling distance. The practical delivery of the latter treatment geometry requires a couch movement or a collimator rotation. The optimum sampling distance equals the 20%-80% penumbra divided by 1.7 and is on the order of 1.5-2 mm for a typical 6 MV beam in soft tissue. The optimum leaf width equals this sampling distance in the regular case. A relatively small deterioration results if the leaf width is doubled, while the sampling distance remains the same. The deterioration can be corrected for by deconvolving the fluence profile with an inverse filter.

With the help of the sampling theory and, more generally, the theory of linear systems, one can find a general answer to the question about the optimum leaf width of a multileaf collimator from a physical point of view. It is important to distinguish between the sampling distance and the leaf width. The sampling distance is more critical than the leaf width. The leaf width can be up to twice as large as the sampling width. Furthermore, the derived sampling distance can be used to select the optimum resolution of both the fluence and the dose grid in dose calculation and inverse planning algorithms.