A Diffeomorphic MLR Framework for Surrogate-Based Motion Estimation and Situation-Adapted Dose Accumulation

Respiratory motion a is major source of uncertainty in radio-therapy. Current approaches to cope with it – like gating or tracking tech-niques – usually make use of external breathing signals, interpreted as surrogates of internal motion patterns. Due to the complex nature of in-ternal motion, a trend exists toward the application of multi-dimensional surrogates. This requires the development and evaluation of appropriate correspondence models between the surrogate data and internal motion patterns. We suggest using a multi-linear regression (MLR) and exploit the Log-Euclidean Framework to embed the MLR within a correspon-dence model yielding diffeomorphic estimates of motion fields of internal structures. The framework is evaluated using 4D CT data of lung tu-mor patients and different surrogates (spirometry, diaphragm tracking, monitoring chest wall motion). Further, the application of the framework for incorporating surrogate-based information about breathing variations into the process of dose accumulation is illustrated.