Integrated 3D anatomical model for automatic myocardial segmentation in cardiac CT imagery

ABSTRACT

Segmentation of epicardial and endocardial boundaries is a critical step in diagnosing cardiovascular function in heart patients. The manual tracing of organ contours in computed tomography angiography (CTA) slices is subjective, time-consuming and impractical in clinical setting. We propose a novel multidimensional automatic edge detection algorithm based on shape priors and principal component analysis (PCA). We have developed a highly customised parametric model for implicit representations of segmenting curves (3D) for left ventricle (LV), right ventricle (RV) and epicardium (Epi) used simultaneously to achieve myocardial segmentation. We have combined these representations in a region-based image modelling framework with high-level constraints enabling the modelling of complex cardiac anatomical structures to automatically guide the segmentation of endo/epicardial boundaries. Test results on 30 short-axis CTA datasets show robust segmentation with error (mean$\pm$std mm) of (1.46$\pm$0.41), (2.06$\pm$0.65) and (2.88$\pm$0.59) for LV, RV and Epi, respectively.

KEYWORDS:

• Myocardial segmentation
• shape analysis
• principal component analysis
• active contours
• cardiac tomographic angiography (CTA)

Acknowledgments

This work was funded in part by a seed grant from the Coulter Foundation (Cou 2017), National Science Foundation (NSF) grant number CCF-1526848, National Institutes of Health (NIH) grant number R01 HL143350 and Army Research Office grant number ARO W911NF-18-1-0281.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Since our models are defined directly in 3D, the term ‘region’ means a volume.

2. Since we are working in 3D images, the zero level set of their 3D SDF representations, $\hat{S}$, is a 3D surface and $\hat{P}$ refers to the collection of 3D zero crossing points on the surface.

3. However, this approximation has little consequence since we depend on the zero level set and not the values of the entire function itself.

4. In contrast to regular Chan–Vese, overlap may occur during evolution, but the explicit overlap penalty introduced in Section 5.3 ensures it does not occur in the final result.

5. This kind of total derivative structure is generally referred to as the material derivative when the derivative is taken in time.

Additional information

Notes on contributors

N. Dahiya

N. Dahiya obtained his BE in Mechanical Engineering from Delhi College of Engineering, India, MS in Mechanical Engineering from UW-Madison, and MS in Electrical and Computer Engineering from Georgia Institute of Technology.He is currently pursuing his PhD in Mechanical Engineering from Georgia Institute of Technology under the guidance of Prof. Anthony Yezzi. Navdeep has over 10 years of research experience in the field of Computer Vision and Image Processing using traditional, Variational as well as Deep Learning based techniques. His current PhD research is focused on developing automated methods for Cardiac image segmentation in both CT and MRI imagery.

A. Yezzi

Dr. A. Yezzi holds the position of Julian Hightower Chair Professor within the School of Electrical and Comptuer Engineering at Georgia Institute of Technology where he directs the Laboratory for Computational Computer Vision. He has over twenty years of research experience in shape optimization via geometric partial differential equations. He obtained his Ph.D. in Electrical Engineering in December 1997 from the University of Minnesota with a minor in mathematics.  After completing a postdoctoral research appointment at Massachusetts Institute of Technology, he joined the faculty at Georgia Tech in August 1999.Dr. Yezzi's research lies primarily within the fields of image processing and computer vision with a particular emphasis on medical imaging and 3D surface reconstruction. He has consulted for a number of companies including GE, 3M, MZA, Philips, Picker, and VTI. His work spans a wide range of image processing and vision problems including image denoising, edge-detection, segmentation, shape analysis, multi-frame stereo reconstruction, visual tracking, and registration.  Some central themes of his research include curve and surface evolution, differential geometry, partial differential equations, and shape optimization.

M. Piccinelli

Dr. M. Piccinelli graduated at the Politecnico di Milano in Italy and obtained her PhD in Bioengineering at the Eindhoven University of Technology in The Netherlands. Her research areas focus on the development of software tools for the diagnosis and treatment of vascular and cardiovascular diseases from a variety of imaging modalities including Positron Emission Tomography (PET), Single-Photon Emission Computed Tomography (SPECT), invasive angiography, Computed Tomography (CT) and Magnetic Resonance (MR).

E. Garcia

Dr. E. Garcia is a past president of the American Society of Nuclear Cardiology, founding president of the American Society of Nuclear Cardiology Foundation, past president of the Institute for Clinical PET, past president of the Cardiovascular Council of the Society of Nuclear Medicine.He was selected by Medical Imaging Magazine as an industry top 10 Nuclear Medicine Researcher, designated by the Better World Project for his leadership role in starting a successful company, spun off from academic research that has changed the world and named Game Changer by Emory University School of Medicine for his role in the development of tools for the objectification of image interpretation. He was also named to the Council of Distinguished Investigators of the Academy of Radiology Research. He is a co-author of more than 330 peer reviewed articles and 60 book chapters and co-editor of 10 books.