Evaluation of Active Appearance Models for Cardiac MRI

Segmentation of the left ventricular endocardium from magnetic resonance images by using different statistical shape models

“…A potential limitation in previous SSM applications [6] to segment the LV from CMR images arises from the fact that the applied 3D shape model used a training set built from manual tracing of 2D SAX CMR slices with anisotropic resolution (1.5 x 1.5 mm planar resolution vs. 6-8 mm slice thickness) subsequently interpolated along the third dimension [7]. In this way, detailed anatomical information along the LV long-axis (typically at LV base and apex) cannot be included in the SSM.…”

3D right ventricular endocardium segmentation in cardiac magnetic resonance images by using a new inter-modality statistical shape modelling method

“…Training the 3D SM on samples derived from manually traced 2D contours on images with anisotropic resolution (slab thickness several times larger than the planar resolution) and subsequently interpolated along the third dimension [24,25], reduce the resolution of the SM. On the contrary, the use of training samples derived from 3D surfaces obtained from 3DE data avoided the need of interpolating and allowed to increase the resolution of the SM; -free from morphological artefacts: training samples are derived from 3DE images that are intrinsically 3D so avoiding potential slice misalignments due to breathing-related motion that could affect the generation of the SM with inaccuracies, such as stair-case aliasing [26,27]; -with detailed anatomical information of basal and apical regions: a type of information usually not included in the SM [28] due to the problematic visualization of these parts in the 2D CMR SAX images.…”

Right Ventricular Endocardial Segmentation in CMR Images using a Novel Inter:Modality Statistical Shape Modelling Approach

“…This model-based technique have been successfully applied to segment left ventricular (LV) endocardium and epicardium in CMR images [2] but few works used a model-based technique to detect the RV endocardium. A potential limitation in previous SSM applications to segment the ventricles from CMR images arises from the fact that the 3D SSM was built from a training set of 2D manual tracings on CMR SAX slices (with anisotropic resolution) interpolated along the third dimension [3]. In this way, detailed anatomical information at RV basal and apical level cannot be included in the SSM.…”