Validation of semi-automatic segmentation of the left atrium

M. E. Rettmann, D. R. Holmes, J. J. Camp, D. L. Packer, R. A. Robb

Research output: Chapter in Book/Report/Conference proceedingConference contribution

11 Scopus citations

Abstract

Catheter ablation therapy has become increasingly popular for the treatment of left atrial fibrillation. The effect of this treatment on left atrial morphology, however, has not yet been completely quantified. Initial studies have indicated a decrease in left atrial size with a concomitant decrease in pulmonary vein diameter. In order to effectively study if catheter based therapies affect left atrial geometry, robust segmentations with minimal user interaction are required. In this work, we validate a method to semi-automatically segment the left atrium from computed-tomography scans. The first step of the technique utilizes seeded region growing to extract the entire blood pool including the four chambers of the heart, the pulmonary veins, aorta, superior vena cava, inferior vena cava, and other surrounding structures. Next, the left atrium and pulmonary veins are separated from the rest of the blood pool using an algorithm that searches for thin connections between user defined points in the volumetric data or on a surface rendering. Finally, pulmonary veins are separated from the left atrium using a three dimensional tracing tool. A single user segmented three datasets three times using both the semi-automatic technique as well as manual tracing. The user interaction time for the semiautomatic technique was approximately forty-five minutes per dataset and the manual tracing required between four and eight hours per dataset depending on the number of slices. A truth model was generated using a simple voting scheme on the repeated manual segmentations. A second user segmented each of the nine datasets using the semi-automatic technique only. Several metrics were computed to assess the agreement between the semi-automatic technique and the truth model including percent differences in left atrial volume, DICE overlap, and mean distance between the boundaries of the segmented left atria. Overall, the semi-automatic approach was demonstrated to be repeatable within and between raters, and accurate when compared to the truth model. Finally, we generated a visualization to assess the spatial variability in the segmentation errors between the semi-automatic approach and the truth model. The visualization demonstrates the highest errors occur at the boundaries between the left atium and pulmonary veins as well as the left atrium and left atrial appendage. In conclusion, we describe a semi-automatic approach for left atrial segmentation that demonstrates repeatability and accuracy, with the advantage of significant time reduction in user interaction time.

Original languageEnglish (US)
Title of host publicationMedical Imaging 2008 - Physiology, Function, and Structure from Medical Images
DOIs
StatePublished - Jun 2 2008
EventMedical Imaging 2008 - Physiology, Function, and Structure from Medical Images - San Diego, CA, United States
Duration: Feb 17 2008Feb 19 2008

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
Volume6916
ISSN (Print)1605-7422

Other

OtherMedical Imaging 2008 - Physiology, Function, and Structure from Medical Images
CountryUnited States
CitySan Diego, CA
Period2/17/082/19/08

Keywords

  • Image-guided ablation
  • Left atrial fibrillation
  • Left atrial segmentation
  • Left atrium

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Biomaterials
  • Atomic and Molecular Physics, and Optics
  • Radiology Nuclear Medicine and imaging

Fingerprint Dive into the research topics of 'Validation of semi-automatic segmentation of the left atrium'. Together they form a unique fingerprint.

  • Cite this

    Rettmann, M. E., Holmes, D. R., Camp, J. J., Packer, D. L., & Robb, R. A. (2008). Validation of semi-automatic segmentation of the left atrium. In Medical Imaging 2008 - Physiology, Function, and Structure from Medical Images [691625] (Progress in Biomedical Optics and Imaging - Proceedings of SPIE; Vol. 6916). https://doi.org/10.1117/12.773097