Meeting Banner
Abstract #1755

Reducing Inter-Observer Variability in DCE-MRI Using Semi-Automatic Lesion Segmentation and Histogram Analysis - Comparison to Manual Region of Interest Placement.

Tobias Heye1, Elmar M. Merkle2, Caecilia S. Reiner1, Matthew S. Davenport3, Jeff J. Horvath1, Sebastian Feuerlein4, Steven R. Breault1, Peter Gall5, Mustafa R. Bashir1, Brian M. Dale6, Attila Kiraly7, Daniel T. Boll1

1Department of Radiology, Duke University Medical Center, Durham, NC, United States; 2Klinik fr Radiologie und Nuklearmedizin, Universittsspital Basel, Basel, Kanton Basel, Switzerland; 3Department of Radiology, University of Michigan Health System, Ann Arbor, MI, United States; 4Department of Radiology and Medical Imaging, University of Virginia Health System, Charlottesville, VA, United States; 5Imaging & Therapy Division, Healthcare Sector, Siemens AG, Erlangen, -, Germany; 6MR R&D Collaborations, Siemens Healthcare, Morrisville, NC, United States; 7Corporate Research and Technology, Siemens Corporation, Princeton, NJ, United States

Inter-observer variability is a considerable factor of measurement variation in any quantitative imaging approach in particular multi-center studies. Many of the contributors to the overall variation in DCE-MRI results are technical or mathematical in nature and their influence is systematic, thus allowing for estimation, correction and optimization of the error. However, unlike these more predictable sources, the observer imparts a more variable contribution to the overall error in the entire process of DCE-MRI. We could shown that a guided measurement method can reduce inter-observer variability significantly (relative reduction by 42.5%) compared to manual ROI placement (16.4% versus 28.5%, respectively).

Abstract PDF Poster PDF

Keywords

able additionally address aimed among angiography applied approaches assisted auto automatic axial background benefit boll category causing clinical combination comparable confinement consisted context control controlled corporate dale davenport decrease dedicated defined degrees delineated description deviation diagnostics directed displayed division done duke encompassing environment every excluded expressed female fibroid fibroids gall graphically guide health histogram identification impact implementation importance improving indicates injection instructions inter intra issue largest least lesion lesions likelihood location made manual manually mapping maps marked material medical model noise observer observers overall pair parametric particular peak peter placed placement plainly position post potential previous previously principles processed processing radiology random reduce reducing reduction refinement remains removed repeated represented reproducibility resolved root section sector segmentation selected selecting semi semiautomatic setting significantly since slice software sources square studies subject system table tagging targeted technology therapy thereby thresh threshold tissue tool trajectories transferred trials twist user uterine variability variable variation wise years