A 3-D Active Contour Method for Automated Segmentation of the Left Ventricle From Magnetic Resonance Images

• Published in: IEEE transactions on biomedical engineering Vol. 64; no. 1; pp. 134 - 144
• Main Authors: Hajiaghayi, Mahdi, Groves, Elliott M., Jafarkhani, Hamid, Kheradvar, Arash
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Active contour method; Active contours; Algorithms; automated; Automation; cardiac MRI; Cardiovascular diseases; Computational geometry; Contour matching; Convexity; Coronary artery disease; Heart; Heart diseases; Heart Ventricles - anatomy & histology; Heart Ventricles - diagnostic imaging; Humans; Image edge detection; Image Enhancement - methods; Image Interpretation, Computer-Assisted - methods; Image processing; Image reconstruction; Image segmentation; Imaging, Three-Dimensional - methods; Interpolation; Machine Learning; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Medical imaging; Muscles; Myocardium; NMR; Nuclear magnetic resonance; Pattern Recognition, Automated - methods; Reproducibility of Results; segmentation; Sensitivity and Specificity; Statistical methods; Statistics; three dimensional; Three dimensional models; Three-dimensional displays; Ventricle; volumetric
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/TBME.2016.2542243

Objective: This study's objective is to develop and validate a fast automated 3-D segmentation method for cardiac magnetic resonance imaging (MRI). The segmentation algorithm automatically reconstructs cardiac MRI DICOM data into a 3-D model (i.e., direct volumetric segmentation), without relying on prior statistical knowledge. Methods: A novel 3-D active contour method was employed to detect the left ventricular cavity in 33 subjects with heterogeneous heart diseases from the York University database. Papillary muscles were identified and added to the chamber using a convex hull of the left ventricle and interpolation. The myocardium was then segmented using a similar 3-D segmentation method according to anatomic information. A multistage approach was taken to determine the method's efficacy. Results: Our method demonstrated a significant improvement in segmentation performance when compared to manual segmentation and other automated methods. Conclusion and Significance: A true 3-D reconstruction technique without the need for training datasets or any user-driven segmentation has been developed. In this method, a novel combination of internal and external energy terms for active contour was utilized that exploits histogram matching for improving the segmentation performance. This method takes advantage of full volumetric imaging, does not rely on prior statistical knowledge, and employs a convex-hull interpolation to include the papillary muscles.