ESTRO 2024 - Abstract Book

S4130

Physics - Inter-fraction motion management and offline adaptive radiotherapy

ESTRO 2024

assumption that the highest doses are always given to the same voxels. Our goal is to more accurately estimate the accumulated DVH, particularly in the high dose region.

Material/Methods:

The new method is based on a statistical interpretation of the DVH. The most common type of DVH is the cumulative DVH, which tells us how much of the organ volume which receives at least a given dose. It is straightforward to convert between the cumulative DVH and the differential DVH, which tells us how much of the organ volume receives a specific dose (more precisely, a dose within a specific dose interval). The differential DVH can be interpreted as the probability density function (pdf) of the dose to a randomly and uniformly selected point within the organ. This also means that the accumulated DVH over multiple fractions is the pdf of the sum of this random variable for each fraction. The pdf of the sum of two independent random variables is the convolution of the pdfs of the individual variables. The dose to a voxel at two fractions is obviously not independent, but rather strongly correlated. Then, the pdf of the sum can be found through a convolution-like integral over the joint pdf of the two variables. The relationship between the marginal pdfs of two variables and their joint pdf is given by a copula function. In our setting, we know the marginal pdfs, but not the copula function. The idea is therefore to use a copula function that fits our assumptions about the dependence between the doses to a voxel at different fractions. Parameters for the copula function can be inferred from expert knowledge or estimated from previous data. For evaluation of the method, we used manually contoured recti from a data set consisting of 2 CT scans from each of 37 prostate cancer patients. We accumulated the rectal DVHs over the two scans for all patients using the new method. We used a Gaussian copula function, though use of more complex copula functions might improve the performance. For comparison, we also used simple averaging per dose bin. Both these methods were compared to the “gold standard” of dose accumulation using deformable registration. We compared the bias and the absolute error of the methods at high doses using the D5% parameter; the minimum dose to the hottest 5% of the volume.

Results:

Both the bias and the absolute error of the D5% were significantly improved with the new method as compared to MVDs. The bias was reduced from 0.73 Gy to 0.19 Gy (p=3.3e-5) and the absolute error was reduced from 0.76 Gy to 0.4 Gy (p=0.003). The D5% value errors are shown as box plots in figure 1.

The average of the accumulated DVHs calculated using averaging, copulas and deformable registration are shown in figure 2. The new method is closer to the gold standard at high doses, but further away at low doses.