A Spatio-Temporal Based Scheme for Efficient Registration-Based Segmentation of Thoracic 4D MRI
Y. Yang, E. Van Reeth, C. L. Poh, C. H. Tan, and I.W. K. Tham
This paper presents a novel automated 4D registration-based segmentation scheme that is based on spatiotemporal information for the segmentation of thoracic 4D MR lung images. Dynamic 3D (4D) MR imaging is gaining importance in the study of pulmonary motion for respiratory diseases and pulmonary tumor motion for radiotherapy. To perform quantitative analysis using 4D MR images, segmentation of anatomical structures such as the lung and pulmonary tumor is required. Manual segmentation of entire thoracic 4D MRI data that typically contains many 3D volumes acquired over several breathing cycles is extremely tedious, time consuming, and suffers high user variability. This requires the development of new automated segmentation schemes for 4D MRI data segmentation. Registration-based segmentation technique that uses automatic registration methods for segmentation has been shown to be an accurate method to segment structures for 4D data series. However, directly applying registration-based segmentation to segment 4D MRI series lacks efficiency. Our proposed scheme presented in this paper saved up to 95% of computation amount while achieving comparable accurate segmentations compared to directly applying registration-based segmentation to 4D dataset. The scheme facilitates rapid 3D/4D visualization of the lung and tumor motion and potentially the tracking of tumor during radiation delivery.