Original ArticleAutomatic detection of left and right ventricles from CTA enables efficient alignment of anatomy with myocardial perfusion data
Introduction
In recent years, three-dimensional (3D) fusion of multislice CT angiography (CTA) and myocardial perfusion imaging (MPI) for the assessment of coronary artery disease (CAD) has proven its incremental diagnostic value over the two modalities separately considered.1, 2, 3, 4, 5 MPI with either PET or SPECT is an established method for the assessment of the functional significance of coronary stenosis,6,7 while CTA provides the complementary morpho-anatomical information on the heart structure, the coronary tree configuration, and the location of stenosis, information that is otherwise unavailable to the physician.8,9 By visually fusing these two types of knowledge, clinicians are provided with tools to better interpret complex situations of multiple perfusion defects, to identify the direct cause of abnormalities and conversely, to spare patients unnecessary invasive exams in case of non-critical lesions.4,5,10
A number of technical issues have to be addressed for this integration to become a reliable and efficient tool for everyday use in a clinical environment. In a previous work published by our group,11 an algorithm for the automated alignment of MPI and CTA images obtained from different scanners was presented. Both left (LV) and right (RV) ventricles were manually identified on the two datasets, binary images were created and successively aligned using mutual information principles.12 Since the reference standard remains manual registration, a comparison of automated vs interactive alignments was conducted and no significant differences were observed.
A crucial prerequisite for the proposed alignment algorithm is the availability of myocardial tomograms from the CTA and MPI images. Manual tracings are considered the reference standard, but exhibit the fundamental drawback of being time-consuming and more importantly operator dependent. While tools for automated quantitative MPI are readily available,6,7,13 researchers have been working for decades on efficient methods to segment cardiac CT (or MR).14, 15, 16 Promising methodologies for the automated extraction of complex anatomical structures rely on the use of model-based segmentation algorithms and on their capability of incorporating prior information about the features of the object to be delineated.17,18 Following this approach, a level set region-based algorithm for the automated detection of LV and RV edges was developed.19,20 The segmentation is constrained by local shape priors retrieved from the statistical analysis of a training datasets and was successfully tested on synthetic images as well as on cardiac acquisitions for myocardium extraction in both 3 and 4D.
With the final rationale of proposing a completely automated procedure for 3D image fusion, the aim of this work was twofold: first, the accuracy of the proposed automated edge detection algorithm was assessed in an animal model on a set of contrast-enhanced CTA images; manual segmentation was also performed by an expert user and considered as the reference standard for the myocardium delineation of both LV and RV. Second, perfusion images were aligned to the CTA acquisitions by means of our published second-generation biventricular fusion technique11 using both the automatically segmented and the manually traced myocardial contours. Differences in translations and rotations between the two resulting alignments were estimated, with the registration of the manual tracings taken as the reference standard.
Section snippets
Animal Selection and Preparation
Twelve female Yorkshire cross farm-bred pigs were obtained from North Carolina; their average weight ranged 35-40 kg. Before image acquisitions, acute myocardial infarcts were created in either the left anterior descending artery (LAD) or left circumflex artery (LCX) vascular territories of each animal by means of a previously described angiographic technique.21 The animals were transported to the scanner room and intubated. Appropriate animal care was provided according to norms/standard
CT Automatic Vs Manual Segmentation
Manual and automated segmentations were performed for all 12 CTA acquisitions. In most of the cases, the best systolic phase was selected for the subsequent image processing steps after evaluation of a CTA expert. From the binary images, 3D triangulated meshes were created so that two endocardial and two epicardial surfaces were available for each case. The boundary distance error between automated and hand-traced meshes after alignment in all 12 animals exhibited maximum, minimum, and median
Discussion
Despite the continuous improvement in non-invasive imaging techniques, the reference standard for the clinical assessment of CAD still relies on invasive testing such as coronary arteriography. The usefulness of image fusion of CTA and nuclear perfusion for the creation of a more complete picture of the patient’s heart condition that could avoid invasive procedure may be widely accepted and recognized, but due to the lack of automation for some of the necessary image processing steps, the
New Knowledge Gained
The automated segmentation of the myocardium from CTA acquisitions allows in a single display reliable fusion of anatomical CTA information and physiological myocardial perfusion information.
Conclusions
Our automatic myocardial boundary detection algorithm created left and right ventricular, endocardial and epicardial surfaces from CTA cardiac images that are similar in accuracy compared to corresponding hand-traced surfaces. Moreover, the automated alignment of these automatically detected CTA surfaces with the PET surfaces are similar in accuracy to the automated alignment using hand-traced CTA surfaces. Further improvements to the automated segmentation algorithm will be implemented and
Acknowledgments
This work was supported in part by NIH Grant R01-HL-085417 from NHLBI and by the EMTech Bio Collaborative Grant program.
Conflict of interest
Some of the authors (EG, RF, TF) receive royalties from the sale of the Emory Cardiac Toolbox and have equity positions with Syntermed, Inc., which markets ECTb. The terms of these arrangements have been reviewed and approved by Emory University in accordance with its conflict of interest policies. The remaining authors do not have any conflicts of interest.
Disclosures
None.
References (29)
- et al.
Diagnostic performance of fusion of myocardial perfusion imaging (MPI) and computed tomography coronary angiography
J Nucl Cardiol
(2009) - et al.
Incremental value of combining 64-slice computed tomography angiography with stress nuclear myocardial perfusion imaging to improve non-invasive detection of coronary artery disease
J Nucl Cardiol
(2010) - et al.
The increasing role of quantification in nuclear cardiology: The Emory approach
J Nucl Cardiol
(2007) - et al.
Quantitation in gated perfusion SPECT imaging: The Cedars-Sinai approach
J Nucl Cardiol
(2007) - et al.
Multi-modal volume registration by maximization of mutual information
Med Image Anal
(1996) - et al.
Deformable models in medical image analysis: A survey
IEEE Trans Med Imaging
(1996) - et al.
A review of segmentation methods in short axis cardiac MR images
Med Image Anal
(2011) - et al.
Statistical shape models for 3D medical segmentation: A review
Med Image Anal
(2009) - et al.
Positron emission tomography myocardial perfusion and glucose metabolism imaging
J Nucl Cardiol
(2006) - et al.
Measurements of left ventricle mass in vivo using gated nuclear magnetic resonance imaging
J Am Coll Cardiol
(1986)
Validation of a new cardiac image fusion software for three-dimensional integration of myocardial perfusion SPECT and stand-alone 64-slice CT angiography
Eur J Nucl Med Mol Imaging
Cardiac positron emission tomography/computed tomography imaging accurately detects anatomically and functionally significant coronary artery disease
Circulation
Three-dimensional fusion of coronary arteries with myocardial perfusion distributions: Clinical validation
J Nucl Med
CT of coronary artery disease
Radiology
Cited by (6)
Integrated 3D anatomical model for automatic myocardial segmentation in cardiac CT imagery
2019, Computer Methods in Biomechanics and Biomedical Engineering: Imaging and VisualizationIntegrated 3D anatomical model for automatic myocardial segmentation in cardiac CT imagery
2018, Lecture Notes in Computational Vision and BiomechanicsMultimodality image fusion
2017, Hybrid Imaging in Cardiovascular MedicineHybrid instrumentation versus image fusion: Path to multibrid visualization
2017, Hybrid Imaging in Cardiovascular Medicine
Dr. Tracy L. Faber, this project’s principal investigator, passed away on March 24, 2012.