AbstractID: 6913 Title: Correction of Surface Dose Artifacts in Convolution Models of Geometric
Uncertainties
Patient repositioning and organ motion are sources of geometric uncertainty that
can lead to dose delivery uncertainty in radiation therapy. The dose actually received by
the patient can be estimated by the convolution of the conventional dose distribution with
the probability density function describing the geometric uncertainties. However,
convolution assumes shift invariance of the dose distribution. This is not valid, and one
consequence of this assumption is inaccurate dose calculation near the surface of the
patient. A correction method is presented that improves the accuracy of convolution in
this region. This method ‘pads’ the conventional dose distribution by assigning dose
values to all points outside of the patient’s external contour. A conventional dose
distribution is calculated, the dose matrix is padded, and convolution is performed. A
clinical example of a sinus tumour patient demonstrates the use of this method. The
results of a conventional dose calculation, a standard convolution, and a corrected
convolution are compared to a direct simulation of geometric uncertainty. The
conventional dose calculation does not accurately predict the dose near the beam edges
when geometric uncertainties are present (10% error). Convolution accounts for
geometric uncertainties, but the assumption of shift invariance results in large errors
(18%) near the surface. The corrected convolution accounts for geometric uncertainties,
and predicts the dose near the surface quite accurately (2% error). This correction
method allows improved accuracy in convolution calculations of the dose to target
volumes and critical organs near the surface of the patient.