Meeting Banner
Abstract #3444

Adaptive Model for Direct Estimation of Hemodynamic Parameters in MR Perfusion Studies: Comparison and Evaluation Using CT Perfusion and Singular Value Decomposition Technique

Hassan Bagher-Ebadian1, 2, Rajan Jain1, Jayant Narang1, Hamid Soltanian-Zadeh1, 3

1Radiology, Henry Ford Hospital, Detroit, MI, United States; 2Physics, Oakland University, Rochester, MI, United States; 3CIPCE-Department of ECE, University of Tehran, Tehran, Iran

This study investigates feasibility of using a model-trained Adaptive Model to estimate Mean-Transit-Time (MTT) and relative-Cerebral-Blood-Flow (rCBF) in Dynamic-Susceptibility (DSC)-MR perfusion studies. A residue function with an exponential kernel was used to model the T2*-weighted-images with bolus passage. The ANN was trained and validated using a set of central moments and K-Folding-Cross-Validation (KFCV) technique. DSC-MR perfusion and CT-perfusion studies of four patents were analyzed using the ANN and Singular-Value-Decomposition technique. Results imply that the ANN produces accurate and stable hemodynamic maps compared to the SVD technique which can be used as a fast and accurate estimator in DSC perfusion studies.

Abstract PDF Presentation Video

Keywords

accuracy accurate adaptive added addition agent analytical analytically application applied arrival arterial assure audience blood brain central cerebral characterize circulant common computed concentration construct constructed contrast controversial cost coverage decomposition deconvolution delay diagnosis digital direct dynamic edema enhanced error estimating estimation estimator evaluating evaluation exact experimental exponential extracted fast feasibility feature features flow fold folding function generalization generated gold good gray hypothesized identified illustrate illustrated imply important injection insensitive intensity investigates iodinated ionic ionizing issue kernel limited linear maps material materials measure measured medical model modeled modeling moment moments network neural neurologists noise none normalized pages part patients peak performance perfusion phantoms physicists physics physiological plays post potential produced proposed providing quantification quantitative radiation radiology registered reliable remains requires residue resolution role simulated singular slice slices software spacing stable stroke studies susceptibility synthetic table target timing tissue tomography trace tracer trained training transit treatment tumor unlike validated various varying volume white