Premium
MO‐F‐CAMPUS‐J‐03: Sorting 2D Dynamic MR Images Using Internal Respiratory Signal for 4D MRI
Author(s) -
Wen Z,
Hui C,
Stemkens B,
Tijssen R,
van den Berg C,
Beddar S
Publication year - 2015
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4925449
Subject(s) - bellows , imaging phantom , signal (programming language) , nuclear medicine , volunteer , magnetic resonance imaging , sagittal plane , computer science , breathing , medical imaging , computer vision , artificial intelligence , nuclear magnetic resonance , physics , medicine , materials science , anatomy , radiology , agronomy , metallurgy , biology , programming language
Purpose: To develop a novel algorithm to extract internal respiratory signal (IRS) for sorting dynamic magnetic resonance (MR) images in order to achieve four‐dimensional (4D) MR imaging. Methods: Dynamic MR images were obtained with the balanced steady state free precession by acquiring each two‐dimensional sagittal slice repeatedly for more than one breathing cycle. To generate a robust IRS, we used 5 different representative internal respiratory surrogates in both the image space (body area) and the Fourier space (the first two low‐frequency phase components in the anterior‐posterior direction, and the first two low‐frequency phase components in the superior‐inferior direction). A clustering algorithm was then used to search for a group of similar individual internal signals, which was then used to formulate the final IRS. A phantom study and a volunteer study were performed to demonstrate the effectiveness of this algorithm. The IRS was compared to the signal from the respiratory bellows. Results: The IRS computed by our algorithm matched well with the bellows signal in both the phantom and the volunteer studies. On average, the normalized cross correlation between the IRS and the bellows signal was 0.97 in the phantom study and 0.87 in the volunteer study, respectively. The average difference between the end inspiration times in the IRS and bellows signal was 0.18 s in the phantom study and 0.14 s in the volunteer study, respectively. 4D images sorted based on the IRS showed minimal mismatched artifacts, and the motion of the anatomy was coherent with the respiratory phases. Conclusion: A novel algorithm was developed to generate IRS from dynamic MR images to achieve 4D MR imaging. The performance of the IRS was comparable to that of the bellows signal. It can be easily implemented into the clinic and potentially could replace the use of external respiratory surrogates. This research was partially funded by the the Center for Radiation Oncology Research from UT MD Anderson Cancer Center.