Meeting Banner
Abstract #3635

Repeatability of Independent Component Analysis Applied to Dynamic Susceptibility Contrast MRI in Newly Diagnosed Brain Tumor Patients with Two Baseline Imaging Scans

Peter S. LaViolette1, Mitchell Daun2, Melissa A. Prah1, Kourosh Jafari-Khouzani3, Pavlina Polaskova3, Elizabeth R. Gerstner4, Steven M. Stufflebeam3, Kathleen M. Schmainda1

1Radiology, Medical College of Wisconsin, Milwaukee, WI, United States; 2Neurology, Medical College of Wisconsin, Milwaukee, WI, United States; 3Radiology, Massachusetts General Hospital, Charlestown, MA, United States; 4Neurology, Massachusetts General Hospital, Boston, MA, United States

Independent component analysis applied to DSC perfusion data separates arterial and venous voxels based on the temporal dynamics of contrast agent perfusion. This study varies the number of ICA components modeled to determine which number generates the most repeatable arterial and venous maps in 31 patients who underwent two identical DSC imaging sessions days apart. We compare the repeatability of the resulting arterial and venous maps both across different numbers of components modeled, and between the simultaneously acquired spin-echo and gradient-echo acquisitions. We find that modeling 3 components results in highly repeatable arterial and venous maps, and that ICA maps generated from GE data are more repeatable than SE.

Abstract PDF Presentation Video

Keywords

acquisition acquisitions administered advancing agent apart applied applying arterial audience best black brain brought categorization characteristics clinicians college combined combining component components consisted contrast corrected correction courses created creation dataset days demonstration derived determine diagnosed dotted driven dual dynamic equation explores funding general generally gradient greater healthier highly hospital human identical implemented independent index indicates injection interpolation leakage library likely maps maximally maximized measure measures medical melodic minimize modeled modeling months motion multivariate nearest neighbor neurology newly noise onset opportunity overall overlap overlapping patient patients perfect perfusion peter planar platform population previous prior probabilistic processed processing protocol radiology rapid recruited relationship removing repeatability repeatable repeated represent represented scanned scientists separate separately session sessions significantly simultaneously smith sorted space spin statistically studies surgery susceptibility system takes target temporal thresholded tool trans transformation treatment tumor tumors twice unique utility venous versus visit visualize visually workshop