Dynamic deformable models for 3D MRI heart segmentation

Abstract

Automated or semiautomated segmentation of medical images decreases interstudy variation, observer bias, and postprocessing time as well as providing clincally-relevant quantitative data. In this paper we present a new dynamic deformable modeling approach to 3D segmentation. It utilizes recently developed dynamic remeshing techniques and curvature estimation methods to produce high-quality meshes. The approach has been implemented in an interactive environment that allows a user to specify an initial model and identify key features in the data. These features act as hard constraints that the model must not pass through as it deforms. We have employed the method to perform semi-automatic segmentation of heart structures from cine MRI data.

Exploring Boundary Concavities in Active Contours and Surfaces

“…When p i − p i+1 gets below d min , vertex p i+1 is deleted and p i is connected to p i+2 . Active surface remeshing is described in [14,18,7]. While resampling the 2D contour is rather straightforward, remeshing the 3D active surface is carefully performed.…”

“…the points belonging to the contour. From an initial location, these models deform according to evolution methods, such as the variational method [3,1], the greedy algorithm [4,5] or the physics-based method [6,7]. All of them share in common the use of internal terms (energies or forces) to maintain the smoothness of the surface and external terms to drive the vertices towards salient image features.…”

Lesion detection using Gabor-based saliency field mapping

“…1,3,4 Because of this importance, accurate lesion diagnosis is a highly active area of study. In recent years, significant research effort has been devoted to the development of medical-imaging systems for an early detection of brain anomalies.…”

“…For example, in 1 the authors utilize statistical analysis to track the changes of brain lesions across the temporal dimension to assess treatment outcomes. A deformable model is employed in 3 to accurately segment volumetric heart data to aid in visualization. In, 4 the authors use maximal likelihood estimation in conjunction with deformable models to accurately segment tumors in functional images of mice.…”

Using anatomical knowledge expressed as fuzzy constraints to segment the heart in CT images

“…Most heart segmentation methods proposed in the literature deal with the segmentation of internal structures (in particular the left ventricle) as in [1][2][3][4][5][6][7] for instance. However, there is also a need to segment the heart as a whole in order to distinguish its limits and the separations with surrounding structures such as the liver or the aorta.…”