Meeting Banner
Abstract #3937

Comparison of Methods for the Analysis of Phase Sensitive Inversion Recovery Images in the Assessment of Myocardial Infarction

Laura Fernandez-de-Manuel1, 2, Haiyan Ding, 13, Maria J. Ledesma Carbayo2, Elliot McVeigh1, Andres Santos2, Aravindan Kolandaivelu4, Daniel A. Herzka1

1Department of Biomedical Engineering, Johns Hopkins School of Medicine, Baltimore, MD, United States; 2Biomedical Image Technologies Lab., ETSI Telecomunicacion, Universidad Politecnica de Madrid and CIBER-BBN, Madrid, Spain; 3Department of Biomedical Engineering, Tsinghua University, Beijing, China; 4Department of Medicine, Cardiology, Johns Hopkins School of Medicine, Baltimore, MD, United States

Two-dimensional PSIR is a well established method of delineating scar in subjects with MI. Quantification of MI has been traditionally based on a threshold of more than 2 standard deviations (+2SD) above the mean of normal myocardial intensity. Newer alternative approaches with improved performance have yet to be applied to 3D PSIR imaging. In this work we compare different scar segmentation methods applied to a dataset of high-resolution 2D and 3D PSIR images from a swine model of MI. Proposed methods are based on Gaussian Mixture Model fitting and the Otsu algorithm, and are compared to the standard (+2SD).

Abstract PDF E-Poster Video

Keywords

account accurate acquisitions animals apex applied appreciated approved array assuming automated avoiding become behaved best biomedical borders breath bright channel china chosen classes classified cohesion comp completely contiguous contours course criterion dark dataset datasets days density dial ding discriminate distances done either empirical engineering error errors established excessive expert facto fibrosis fitting free funding funds gated gating general gold gray help histogram hold imperfect importance improve infarct initial intensity interpolated intervention inversion isolated johns lead manual maria mass matrix maximizes measure measured medicine meters minim ministry misclassification mixture myocardial navigator norm optimal optimum others pare part partial percent pixels poorer post processed processing produced protocol quantification recovery reduction regional remote remove republic require resolution respect resultant scar school science segmentation selected selection sensitive separability sets similarity slices smoother social spoiled step subtle supervised surface surfaces swine system taken tissue typical union user valuated ventricle volume weeks ween yielding