Improved Estimation of Cardiac Function Parameters Using a Combination of Independent Automated Segmentation Results in Cardiovascular Magnetic Resonance Imaging.

Jessica Lebenberg,Jean Cousty,Constantin Constantinidès,Frederique Frouin,Christopher Casta,Stephanie Jehan-Besson,Laurent Sarry,Laurent Najman,Patrick Clarysse,Alain Lalande,Muriel Lefort,Alain De Cesare,Elodie Roullot,Christophe Tilmant,Mireille Garreau,Alexandre Cochet,Irene Buvat

doi:10.1371/journal.pone.0135715

Abstract

This work aimed at combining different segmentation approaches to produce a robust and accurate segmentation result. Three to five segmentation results of the left ventricle were combined using the STAPLE algorithm and the reliability of the resulting segmentation was evaluated in comparison with the result of each individual segmentation method. This comparison was performed using a supervised approach based on a reference method. Then, we used an unsupervised statistical evaluation, the extended Regression Without Truth (eRWT) that ranks different methods according to their accuracy in estimating a specific biomarker in a population. The segmentation accuracy was evaluated by estimating six cardiac function parameters resulting from the left ventricle contour delineation using a public cardiac cine MRI database. Eight different segmentation methods, including three expert delineations and five automated methods, were considered, and sixteen combinations of the automated methods using STAPLE were investigated. The supervised and unsupervised evaluations demonstrated that in most cases, STAPLE results provided better estimates than individual automated segmentation methods. Overall, combining different automated segmentation methods improved the reliability of the segmentation result compared to that obtained using an individual method and could achieve the accuracy of an expert.

Highlights

Cardiac Magnetic Resonance Imaging is used more and more frequently in clinical routine to study simultaneously the cardiac anatomy and function
In [2], authors proposed to use Simultaneous Truth and Performance Level Estimation (STAPLE) to define a gold standard segmentation based on two fully-automated algorithms and three semi-automated algorithms requiring manual input, while the present study focuses on improving the accuracy of automated segmentation algorithms by combining them with STAPLE to get a accuracy similar to the one achieved by experts i.e. make it acceptable for clinical routine
MS578 was ranked like M4, this rank being worse than the experts’ ranks but better than the individual methods used to create the combination. These results demonstrate that the left ventricular ejection fraction (LVEF) parameters were more accurately estimated using this combination of segmentation methods than with any of the segmentation methods used in the combination

Summary

Introduction

Cardiac Magnetic Resonance Imaging (cMRI) is used more and more frequently in clinical routine to study simultaneously the cardiac anatomy and function. A series of clinical parameters can be deduced from the acquired scans in cMRI. Among these parameters, the left ventricular ejection fraction (LVEF) remains a major prognostic index for coronary artery diseases assessment. The correct estimation of this parameter requires the accurate measurement of both end-diastolic volumes (EDV) and end-systolic volumes (ESV), providing the stroke volume (SV) and the LVEF. MRI makes these measurements possible with a high accuracy (generally from a series of short-axis cine-MR images), the segmentation of the left ventricle (LV) is still a contemporary issue [1] due to the considerable amount of data that are acquired in a single examination. Semi-automated algorithms that are proposed by commercial image post-processing software are largely used. To assess the performance of these automated segmentation algorithms, the common approach consists in comparing the contours resulting from the automated segmentation with the ones obtained by one or several experts who are known to often outperform automated methods [3]

Objectives

Methods

Results

Discussion

Conclusion