- Poster presentation
- Open Access
An automatic segmentation for improved visualization of atrial ablation lesions using magnetic resonance imaging
© Karim et al; licensee BioMed Central Ltd. 2011
- Published: 2 February 2011
- Atrial Fibrillation
- Leave Atrial
- Paroxysmal Atrial Fibrillation
- Manual Interaction
- Atrial Wall
Delayed-enhancement (DE) MRI is an effective technique for imaging left atrial (LA) ablation lesions following radio-frequency ablation for atrial fibrillation (AF). Existing techniques for segmentation/visualization of lesions require manual interaction of an expert user making them prone to high observer variability. Oakes et al. applied thresholding on the intensity histogram of the manually-outlined atrial wall. Peters et al. used maximum intensity projections (MIPs) to generate volume lesion visualizations for expert-user interpretation. Knowles et al. employed MIP followed by user-interactive thresholding for lesion surface visualization.
The aim of our work was to develop a fully-automated approach for LA lesion segmentation/visualization.
Five patients with paroxysmal AF (2 male, average age 55 years, LA size 4.2±0.5cm) underwent pulmonary vein (PV) isolation achieving successful electrical isolation of all PVs. Imaging was performed 6-months post-ablation on a Philips 1.5T Achieva with scans including respiratory-navigated and cardiac-gated whole-heart 3D-SSFP and inversion recovery-prepared MRI for visualization of Gd-DTPA DE (complete LA coverage, resolution of 1.3×1.3×2mm3).
The endocardial cavity of the LA was segmented from the whole heart SSFP-MRI scan. The segmented LA was registered to the DE-MRI scan using DICOM header data. The atrial wall was approximated by a ±3mm thick region from the endocardial surface. For each voxel in the wall, the probability of it being labelled as scar or healthy was derived from trained classifiers. For the healthy-tissue class, a mixture of Gaussian distributions was used to model the observed atrial wall tissue, with parameters computed using the Expectation-Maximization algorithm. For the scar-tissue class, the classifier was trained on prior segmentations. The DE-MRI images were segmented using the proposed algorithm and by an expert using the approach in .
The number of cardiac surface vertices classified into the healthy and scar tissue categories based on semi-automated and automatic segmentation. There was no statistical difference in the methods (p=0.36 paired t-test).
Automated algorithm (# surface vertices)
Expert-operated semi-automated (# surface vertices)
% scar using automated algorithm
% scar using expert-operated semi-automated
This study demonstrates a fully-automatic and rapid segmentation/visualization of post-ablation LA DE-MRI that will minimise the observer variability that is seen with existing approaches that require expert manual interactions.
- Oakes, et al: Circulation. 2009, 119 (13): 1758-67. 10.1161/CIRCULATIONAHA.108.811877.PubMed CentralView ArticlePubMedGoogle Scholar
- Peters, et al: Radiology. 2007, 243 (3): 690-5. 10.1148/radiol.2433060417.View ArticlePubMedGoogle Scholar
- Knowles, et al: IEEE Trans.Biomed.Eng. 2010, 57 (6): 1467-75. 10.1109/TBME.2009.2038791.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (<url>http://creativecommons.org/licenses/by/2.0</url>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.