Program Information
Respiration-Induced Landmark Motion Tracking in Ultrasound Guided Radiotherapy
P Huang1,3, D Li1* , K Cao2 , Q Song2 , P Kazanzides3 , I Iordachita3 , M Bell3 , J Wong3 , K Ding3 , (1) Shandong Normal University, Jinan, Shandong Province, (2) CuraCloud Corporation, Seattle, Washington, (3) Johns Hopkins University, Baltimore, MD.
Presentations
SU-F-708-4 (Sunday, July 30, 2017) 2:05 PM - 3:00 PM Room: 708
Purpose: To accurately track anatomical landmarks (e.g. vessel) in real-time in ultrasound (US) guided radiotherapy.
Methods: Ten sets of 2D US sequences provided in the MICCAI challenge on liver ultrasound tracking (CLUST 2015) are used in this study. All datasets were preprocessed to extract feature point, following by a principle component analysis (PCA) to improve computational efficiency. The slow feature analysis (SFA) was then performed on the PCA projections to extract slowly varying components, and frequency analysis was employed on the SFA output signals to acquire the motion phases. For each case, vessel center location was annotated by experts and served as ground truth (GT). A sub-sequence lasting 100 seconds was selected as training data, while the remaining frames serveed as testing data. We related the motion phase in testing frames with the most similar motion phase in training frame. We then mapped the pre-defined target location to testing data in order to make fast spatial alignment, which allows us to utilized a fast template matching (TM) within a small search region to improve the tracking accuracy.
Results: The average tracking error between expert’s annotation and the location extracted from the indexed training frame is 1.3 ± 1.3 mm in the x-direction and 0.7 ± 1.1 mm in the y-direction. Adding a fast TM procedure within a small search region reduces the error to 0.8 ± 1.5 mm and 0.5 ± 1.3 mm. Furthermore, the tracking time per frame is 15 milliseconds, which is well below the frame acquisition time.
Conclusion: We propose a fast approach for tracking respiration induced landmark motion using 2D ultrasound data with high accuracy and low runtime. For the future work, our proposed real-time US tracking approach can be extended to 3D and integrated into the current ultrasound guided radiotherapy workflow.
Funding Support, Disclosures, and Conflict of Interest: This research was partially funded by the National Natural Science Foundation of China (NO.61471226), Natural Science Foundation for Distinguished Young Scholars of Shandong Province (NO. JQ201516).We also thank the supporting of Taishan scholar project of Shandong Province.
Contact Email: