Dominika Kruk1, Arnaud Boucher1, Alain Lalande1,2, Alexandre Cochet1,2, and Tadeusz Sliwa1
1Le2i FRE2005, CNRS, University of Burgundy, Auxerre, France, 2Service de Spectroscopie-RMN, CHU de Dijon, Dijon, France
Synopsis
Previous approaches on
left ventricle segmentation from DE-MRI have focused on the extraction of
myocardium or just diseased region in short axis orientation. However these
studies are not well suited for the segmentation of non-diseased myocardium
region on DE-MRI. This paper presents a novel semi-automatic segmentation method
of non-diseased myocardium region segmentation on diseased patient heart from
DE-MRI based on watershed algorithm with shape priors application. To assess
the results segmented images were compared with gold standard images performed
by an experienced user. Novel solution for DE-MRI segmentation has been
proposed, which brought promising results of measured parameters.
1. Introduction
Cardiac MR (CMR) provides
information about structure, perfusion and contractile function of the heart.
One of the MR techniques used to assess the viability of myocardium is
DE-MRI1.
To obtain
quantitative results on the myocardium region, the application of segmentation
method will be necessary.
Previous approaches on left
ventricle segmentation from DE-MRI have focused on the extraction of myocardium2-6 or just diseased region7-11. However these studies did not take into
account the segmentation of non-diseased myocardium region from DE-MRI. The
segmentation of non-diseased myocardium from DE-MRI, has some useful
applications. For instance it can simplify the PET-MR registration process,
because the strongest PET signal is expected on the areas of a healthy part of
myocardium. Our method can be also used to differentiate the non-diseased and
diseased part on cine-MRI. Several studies have been carried out on the
segmentation of myocardium from cine-MRI by using the combination of shape
priors and with usual segmentation method12,13. Shape priors
significantly improves the results of myocardium segmentation on cine-MRI.
The core problem of
the left ventricle segmentation is the complex geometry of the myocardium,
which is presented in various plane. However, the similarity between the MR
signals on the myocardium and its surrounding region induces a low signal to
noise ratio coefficient of DE-MRI. To deal with this problem, we propose a
semi-automatic segmentation method for DE-CMR images dedicated to detect non-diseased
myocardium part. Our method is based on watershed algorithm and shape priors
application. 2. Methods
Vincent and Soilles
presented a technique based on the watershed algorithm, which is applied to the
gradient magnitude image data and produces the small volume primitives14, 15. The output of watershed segmentation is the image, which represents the
homogeneous areas of gray level intensities, instead of every pixel
representation. Watershed approach was selected in order to obtain the map of
homogeneous areas on the image. Shape priors was applied to select the regions
of pixels intensities which include non-diseased myocardium part.
To obtain a left ventricle
segmentation in 3D, user has to position the points on two and four cavity 2D DE-MR
images to define the shape of the myocardium. The points are projected to 3D
short axis images. For each short axis image we obtained four points which are
used to define the shape of myocardium as an ellipse(see Fig.1). The gray level
information is used as a reference value to choose the non-diseased parts of
myocardium and excluded diseased regions, which were usually included in the
applied shape (Fig.2).
Results
The segmentation process was performed on 160 short
axis images. The images were acquired from 20 patients with different cardiac
disorders (myocardial infraction, hypertrophic cardiomyopathy and normal
heart). All patients underwent CMR in 1.5T magnet with late gadolinium
enhancement (LGE) sequence in Centre Hospitalier Universitaire in Dijon.
The quantitative analysis were performed to assess the
difference between the gold standard performed by experienced user and segmented
image obtained by using our method (see fig.2).
To assess the results the DC and RMSD were computed.
The value of RMSD was equal 1.3±0.25mm and the value of DC was equal 0.7±0.11.
The best value of these parameters was obtained at mid
level (RMSD=1.16 ±0.18mm, DC = 0.76± 0.09), and base level (RMSD=1.3±0.29mm,
DC=0.71±0.14) of myocardium. At apical level (RMSD=1.43±0.29mm, DC=0.63±0.12) the
segmentation is less efficient due to the lower contrast between tissue in this
part of myocardium (see fig.3 and 4).
Slightly low value of DC occurred in
blurred and almost not visible edges between myocardium and tissues around. The
value of RMDS is relatively low so it shows that the contour obtained by our
method is close to the contour plotted by an expert.
Discussion and Conclusion
Cardiac
MR Images are usually blurred with almost invisible boundary between lung and
myocardium. Automatic detection of the regions of myocardium based on shape
priors greatly simplifies the segmentation process. The segmentation of the
images which represents different cardiac diseases, often brought good value of
measured parameters. This method of non-diseased myocardium
segmentation will be applied as one of the step in our 3D PET-MR registration method and to cine-MRI
segmentation. Segmented images can be used to build 3D model of the
non-diseased myocardium region. In conclusion, we have proposed a novel solution
for non-diseased myocardium region segmentation from DE-MRI, which brought promising
results.Acknowledgements
No acknowledgement found.References
- Kramer,
Christopher M., et al. "Standardized
cardiovascular magnetic resonance (CMR) protocols 2013 update." Journal of
Cardiovascular Magnetic Resonance 15.1 (2013): 91.
- Chenoune, Y., et
al. "Methodology for jointly assessing myocardial infarct extent and
regional contraction in 3-D CMRI." IEEE Transactions on Biomedical
Engineering 59.9 (2012): 2650-2659.
- Ciofolo, Cybele,
et al. "Automatic myocardium segmentation in late-enhancement MRI."
Biomedical Imaging: From Nano to Macro, 2008. ISBI 2008. 5th IEEE International
Symposium on. IEEE, 2008.
- Dikici, Engin,
et al. "Quantification of delayed enhancement MR images."
International Conference on Medical Image Computing and Computer-Assisted
Intervention. Springer, Berlin, Heidelberg, 2004.
- Wei, Dong, et
al. "Myocardial segmentation of late gadolinium enhanced MR images by
propagation of contours from cine MR images." International Conference on
Medical Image Computing and Computer-Assisted Intervention. Springer, Berlin,
Heidelberg, 2011.
- Xu, Robert S.,
et al. "Myocardial segmentation in late-enhancement MR images via
registration and propagation of cine contours." Biomedical Imaging (ISBI),
2013 IEEE 10th International Symposium on. IEEE, 2013.
- Detsky, Jay S., et al. "Reproducible
classification of infarct heterogeneity using fuzzy clustering on multicontrast
delayed enhancement magnetic resonance images." IEEE transactions
on medical imaging 28.10 (2009): 1606-1614.
- Hennemuth, Anja, et
al. "A comprehensive approach to the analysis of contrast enhanced cardiac
MR images." IEEE Transactions on Medical Imaging 27.11
(2008): 1592-1610.
- Karim, Rashed, et
al. "A method to standardize quantification of left atrial scar from
delayed-enhancement MR images." IEEE journal of translational
engineering in health and medicine 2 (2014): 1-15.
-
Lu, Yingli, et al.
"Automated quantification of myocardial infarction using graph cuts on
contrast delayed enhanced magnetic resonance images." Quantitative
imaging in medicine and surgery 2.2 (2012): 81
-
Tao, Qian, et al.
"Automated segmentation of myocardial scar in late enhancement MRI using
combined intensity and spatial information." Magnetic Resonance in
Medicine 64.2 (2010): 586-594
- Mahapatra, Dwarikanath. "Cardiac image segmentation from cine cardiac MRI using graph cuts and shape priors." Journal of digital imaging 26.4 (2013): 721-730.
- Alessandrini, Martino, et al. "Using a geometric formulation of annular-like shape priors for constraining variational level-sets." Pattern Recognition Letters 32.9 (2011): 1240-1249.
-
Grau, Vicente, et al. "Improved watershed
transform for medical image segmentation using prior information." IEEE
transactions on medical imaging 23.4 (2004): 447-458.
-
Vincent, Luc, and
Pierre Soille. "Watersheds in digital spaces: an efficient algorithm based
on immersion simulations." IEEE
Transactions on Pattern Analysis & Machine Intelligence 6 (1991): 583-598.