ESTRO 2024 - Abstract Book

S64

Invited Speaker

ESTRO 2024

3424

Methodological challenges

Eliana Maria Vasquez Osorio

University of Manchester, Division of Cancer Science, Manchester, United Kingdom. The Christie NHS Foundation Trust, Radiotherapy Related Research, Manchester, United Kingdom

Abstract:

In this presentation, we will explore the methodological considerations for dose accumulation in patient treatment from the point of view of the registration.

There is an assumption that the results from image registration can be used for mapping any related signal between images, in this case, the related dose distribution. For intensity-based registrations, this assumption can easily be challenged whenever what we visualize in the images does not tell us the complete story. A clear example of this is tumor regression in the adaptive radiotherapy context, where different modes of regression have been reported: elastic (where the visible tumor shrinks and surrounding tissue follows this change) and inelastic (where the visible tumor shrinks, but the surrounding tissue stays in its place). As the registration aims at aligning information based on the intensities in the images, the resulting registration will erroneously stretch the region around the inelastically shrinking tumor, thereby resulting in incorrect dose mapping. Alternatively, contour-based registration could avoid reliance on intensities. However, as these registrations rely on manually/automatically delineated structures, inter- and/or intra-observer variations introduce registration uncertainties. Hybrid methods can provide a good alternative, but further work to estimate dose mapping uncertainty is required for these methods, particularly those available in commercial software. A second important assumption introduced during registration implementation is that resulting vector fields are smooth and continuous. This restriction is introduced to handle ill-defined problems, such as registration, where the problem can have infinite solutions; for example, there are infinite vector fields (with slight differences) that can be used to align the input images. However, reality is often not 'smooth and continuous', such as when there is sliding between tissues or tissue mismatch between the input images. In these cases, we are asking the registration to 'go against' its implemented rules, and as a result, the registration results are often compromised. Advanced registrations have addressed these challenges, but they are often not available in most clinical software. A general limitation for dose mapping is the lack of ground truth, limiting our ability to determine the correctness of the mapped dose distributions. Therefore, we rely on contours/landmarks to establish the geometrical performance of the registration. However, geometrical performance does not directly translate to dose mapping uncertainty as it depends on the location and steepness of dose gradients. Therefore, dose mapping uncertainties are patient- and treatment-specific, showing the need to have individualized tools to estimate these uncertainties.

3425

MLC modelling: Get it right and go Leaf your Life

Victor Hernandez 1 , Jordi Saez 2