TH-A-224-02: Atlas-Based Auto-Segmentation of CT Images for Radiotherapy Planning

Abstract

Treatment planning for high precision radiotherapy of cancer patients requires accurate delineation of both target volumes and critical structures from the planning CT image. Manual contouring is tedious and suffers from large inter- and intra-rater variability. Tools for automated segmentation are thus needed. Atlas-based approaches have become popular for automatic medical image segmentation due to its general applicability and advancements in deformable image registration techniques. In this talk, we present our research on developing fully automated, atlas-based methods for segmenting organs at risk from CT images in radiotherapy planning. One particular atlas registration approach that we developed employs a hierarchical scheme and makes use of object shape information in the atlas to improve the registration efficiency and robustness while still being able to account for large inter-subject shape differences. Together with a multi-atlas fusion strategy, accurate segmentation results can be obtained for images of different anatomical sites. To make multi-atlas segmentation computationally feasible, we also take advantage of recent advancements in GPU technology and have implemented our method on NVIDIA GPU cards. The GPU acceleration allows the auto-segmentation of a typical CT image with 10 atlases to be computed within a time-frame of several minutes. After presenting validation results, we conclude the talk with discussions about future research directions. Learning Objectives: 1. Understand the basic principle and workflow of atlas-based image segmentation 2. Understand how atlas-based image segmentation technology can help improve consistency and efficiency in structure contouring for treatment planning 3. Understand the promises and limitations of existing technology, as well as future research directions