Novel computational tools for dosimetry, segmentation and treatment planning in targeted radionuclide therapy

Objectives The aim of this work is a proof of principle of new, combinable algorithms to process PET/CT or SPECT/CT data for the purpose of voxel-based dosimetry calculations and treatment planning in Targeted Radionuclide Therapy.

Methods A subdivision surface (SubD) mesh superimposed on the PET/CT provides the segmentation of the organ of interest by means of a 3D gradient vector flow followed by a deformation of a predefined SubD phantom in order to locally match the individual patient anatomy. A generic, statistical approach combining Gaussian mixture models and a Markov random field is used for tumor segmentation on the co-registered PET image. Dose calculation is realized by convolution of the accumulated activity with discrete dose kernels using a Fast Fourier Transformation. The modules are tested on patient data from peptide receptor radionuclide therapy.

Results All algorithms require very little computation time (< 1 min. on a single PC) and are able to process standard DICOM files. The statistical tumor segmentation is insensitive to the chosen PET reconstruction. The output consists of dose-volume histograms and visual depiction of the dose distribution for a segmented, partial phantom that matches the individual patient’s anatomy. The chosen approach enables a much more versatile and faster 3D visualization.

Conclusions The presented computational tools enable voxel-based dosimetry in Targeted Radionuclide Therapy combined with fast and automatic segmentation, thus avoiding inter-observer variations due to manual delineation. Future works involve quantitative evaluation of the algorithms on a large series of patient data.

Research Support This work was co-funded by the Austrian Federal Ministry for Transport, Innovation and Technology within the program ModSim Computational Mathematics which is part of the program Research, Innovation, Technology and Information Technology.