Segmentation of RT3D ultrasound with implicit deformable models without gradients

This paper presents the implementation and validation of a new 3D deformable model method, based on the Mumford-Shah functional for segmentation of three-dimensional real-time ultrasound. An experiment on 10 patients with primary hypertension was carried out to compare three segmentation methods for quantification of right and left ventricular ejection fraction: (1) manual tracing by an expert cardiologist, (2) 2D parametric deformable model, and (3) 3D implicit deformable model implemented with a level set framework. Deformable model segmentations were performed on denoised data using a (3D+Time) brushlet expansion. The clinical study showed superior performance of the deformable model in assessing ejection fraction when compared to MRI measures. It also showed that the three-dimensional deformable model improved EF measures, which is explained by a more accurate segmentation of small and convoluted ventricular shapes when integrating the third spatial dimension.