Semi-automated analysis of infarct heterogeneity on DE-MRI using graph cuts
- PDF / 410,149 Bytes
- 3 Pages / 595.28 x 793.7 pts Page_size
- 2 Downloads / 163 Views
TECHNOLOGIST PRESENTATION
Open Access
Semi-automated analysis of infarct heterogeneity on DE-MRI using graph cuts YingLi Lu1, Kim A Connelly2,3*, Yuesong Yang1, Subodh B Joshi2,3, Graham Wright1,4, Perry E Radau1 From 15th Annual SCMR Scientific Sessions Orlando, FL, USA. 2-5 February 2012 Background Two popular methods for determining the threshold values for the infarct core and gray zone on delayed enhancement MR images (DE-MRI) have been proposed previously: full width and half maximum (FWHM) and standard deviation (SD) methods [1]. Major limitations of these methods are:1) three manually drawn contours are needed for endocardial, epicardial and remote myocardium boundaries, which is time consuming and suffers from inter-observer and intra-observer variability; 2) the difficulty in reproducible manual delineation of remote myocardium, is an important contributor to variability in results; and 3) the dependence on the remote region statistics is problematic due to the low SNR of this region [2]. The purpose of this research was to develop a novel algorithm for segmentation of infarct core and gray zone from conventional IR-GRE shortaxis MR images with highly robust and reproducible results comparable to the FWHM analysis while eliminating the requirement for drawing a remote myocardial region. Methods Eleven male patients (age: 63.5±11.8 yr) with known CAD and evidence of LGE and chronic MI had cardiac IR-GRE MR scans with full left ventricle (LV) coverage (7-13 slices). MR imaging was performed on a 1.5 T scanner (CV/i, GE Healthcare) using an 8-channel cardiac coil. DE-MRI was started 10 min after the injection of 0.2 mmol/kg of Gd-DTPA (Magnevist, Berlex). First, the endocardial and epicardial contours were generated from the corresponding SSFP images automatically [3] with papillary muscles included. (42%(54/136) of the contours need manual adjustment. Next, the graph cuts 2 Keenan Research Centre in the Li Ka Shing Knowledge Institute, St. Michael’s Hospital and University of Toronto, Toronto, ON, Canada Full list of author information is available at the end of the article
algorithm [4] was used to segment the infarct: 1) a twoclasses Gaussian Mixture Model (GMM) was created for the myocardial ROI, determined by the endocardial and epicardial contours; 2) each pixel was assigned to the most likely Gaussian component; and 3) a graph was built and the graph cut algorithm finds the optimum classification of healthy and infarcted myocardium. Finally, the segmented infarct is separated into infarct core and gray zone using a threshold of half the maximal signal within the segmented infarct (Fig. 1). Linear regression analysis and Bland-Altman plots were used to compare the FWHM method (requiring manual ROIs for LV and remote myocardium) and our method (Fig. 2).
Results There were excellent correlations of the infarct size (infarct core 1: R^2 = 0.99; gray zone: R^2 = 0.95) derived from our graph cuts method and the manual FWHM method. The Bland-Altman analysis indicated that there was a small overestimation
Data Loading...
