Dose Reconstruction from PET Images in Carbon Ion Therapy: A Deconvolution Approach Using an Evolutionary Algorithm

Abstract

Dose monitoring and range verification are important tools in carbon ion therapy. For their implementation, positron emission tomography (PET) can be used to image the $\beta^{+}$-activation of tissue during treatment. Predictions of these $\beta^{+}$-activity distributions are usually obtained from Monte Carlo simulations, which demands high computational time and thus limits the applicability of this technique in clinical scenario. Nevertheless, it is desirable to explore faster approaches able to give such a prediction, since only its comparison with the measured distributions allows a definite assessment of potential range deviations from the planned treatment. For the first time, we present an approach to perform deconvolution from PET data in carbon ion therapy and reconstruct the dose. A filtering method is used to predict positron emitter profiles from dose profiles in short time. In order to reverse the convolution and estimate a dose distribution from a positron emitter distribution, we apply an evolutionary algorithm. Filters are obtained from either a library or are created in advance for a specific problem, assuming that a prediction of the positron emitter distribution is available. To perform the latter method and find the best filter for a specific problem, we use another evolutionary algorithm, hence optimizing the filter on-the-fly for the given treatment scheme. The application of our method is shown for dose and positron emitter distributions in homogeneous phantoms using simulated and newly measured online PET data. Carbon ion ranges can be predicted within 2 mm and the shape of the dose distribution is reconstructed with an overall promising agreement.