Analysis of respiration compensation techniques in fMRI

Peter Mckinnis, Purdue University


Functional magnetic resonance imaging data is confounded by many sources of noise and therefore has a low contrast to noise ratio. This has prompted an effort among the fMRI community to find ways to improve the quality of the fMRI signal. One of the avenues chosen has been retrospective digital filters to reduce quasi-periodic respiratory and cardiac induced noise. A host of different filters have been proposed in the literature, however, no satisfactory analysis has been performed to determine their effectiveness. This study analyzes three respiration correction algorithms: RETROICOR, a complex data driven algorithm, and a magnitude only data driven algorithm. To determine the effectiveness of the algorithms, a synthetic signal was overlaid on a portion of a baseline fMRI scan of six subjects at field strengths of 3T and 1.5T. The baseline images were then filtered using each of the various filters. To provide a control, the images were also analyzed unfiltered. Images were assessed for active and inactive voxels using multiple linear regression. The number of true detections and false detections were calculated and used as a quantitative assessment of the filter's performance. Receiver operator curves and the area under the receiver operator curve are presented for each filter. It was found the RETROICOR was the top performing filter when processing scans with a large number volumes. CPX performed the best when there were only a few volumes acquired in a scan. The Magnitude Data driven technique was found to have unpredictable performance with the filter improving the data in some cases and degrading it in other cases.




Talavage, Purdue University.

Subject Area

Biomedical engineering|Medical imaging

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server