Directional Representations of 4D Echocardiography for Temporal Quantification of LV Volume

Abstract

Abstract. Real-time acquisition via four-dimensional (3D plus time) ultrasound obviates the need for slice registration and reconstruction, leaving segmentation as the only barrier to an automated, rapid, and clinically applicable calculation of accurate left ventricular cavity volumes and ejection fraction. Speckle noise corrupts ultrasound data by introducing sharp changes in an image intensity profile, while attenuation alters the intensity of equally significant cardiac structures, depending on orientation with respect to the position of the ultrasound beam. These properties suggest that measures based on phase information rather than intensity are appropriate for denoising and boundary (surface) detection. Our method relies on the expansion of temporal volume data on a family of basis functions called Brushlets. These basis functions decompose a signal into distinct patterns of oriented textures. Projected coefficients are associated with distinct "brush strokes" of a particular size (width) and orientation (direction). Brushlet decompositions are invariant to intensity (contrast range) but depend on the spatial frequency content of a signal. Preliminary results of this directional spacefrequency analysis applied to both phantoms and clinical data are presented. The method will be used to clinically evaluate 4D data and to extract and quantify heart LV volumes.

LV volume quantification via spatiotemporal analysis of real-time 3-D echocardiography

“…These examples also showed the superiority of 4-D brushlet denoising over 3-D for 4-D data sets. We have previously shown that we can characterize and isolate features of interest in LV volumes by selection of specific brushlet coefficients [64]. Since decomposition on a brushlet basis can efficiently isolate directional features at specific frequencies, preprocessing of RT3D volumes via thresholding of lower frequency brushlet coefficient can assist segmentation by removing noise components and enhancing anatomical features.…”

BM3D-based ultrasound image denoising via brushlet thresholding

“…To collaboratively filter the volumes of grouped blocks, we use brushlet denoising [13]. Specifically, we decompose the volume in the Fourier domain, compute brushlet coefficients, and eliminate high energy coefficients in the lower frequencies via an adaptive thresholding.…”

Wavelets in Temporal and Spatial Processing of Biomedical Images

“…A new spatio-temporal directional analysis tool called the brushlet, first introduced by Meyer & Coifman in 1997 (16), has been shown to be remarkably effective in the analysis of ultrasound data. Angelini et al (96) developed directional denoising and segmentation in three dimensions for feature extraction, by identifying efficient brushlet projection coefficients within sets of redundant articulated (orientation-rich) bases. An example of a set of coefficients for an overcomplete transform in 12 different directions is displayed in Figure 12.…”