Browse Prior Art Database

Method and System for Utilizing an Algorithm for Automatic Segmentation of Anatomical Structures in Four-Chamber View Echocardiogram Images

IP.com Disclosure Number: IPCOM000241431D
Publication Date: 2015-Apr-25
Document File: 8 page(s) / 369K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method and system is disclosed for providing an algorithm for automatic segmentation of anatomical structures in four-chamber view echocardiogram images.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 47% of the total text.

Page 01 of 8

Method and System for Utilizing an Algorithm for Automatic Segmentation of Anatomical Structures in Four-Chamber View Echocardiogram Images

Cardiac ultrasound or echocardiogram is a popular screening tool for many cardiac diseases ranging from valvular diseases, cardiomyopathies, to infarctions and aneurysms. Typically, the echocardiograms represent imagery taken from multiple viewing angles suitable for illustrating various anatomical structures and anomalies. A popular viewpoint to use

for diagnosis is an apical 4-chamber view where heart muscle, valves and chambers are clearly visible. Automatic

interpretation of such structures and anomalies requires good region segmentation algorithms that partition echocardiogram images into regions roughly corresponding to important anatomical structures. Such region segments are a good basis for subsequently building a cardiac atlas when the region segments are generated across a large collection of similar viewpoints.

Disclosed is a method and system for providing an algorithm for automatic segmentation of anatomical structures in four-chamber view echocardiogram images. In accordance with the method and system, a two-pass segmentation algorithm is used for segmentation. First, an image to identify heart muscle (bright) and chamber regions (dark) is processed by adapting an edge weighted centroidal Voronoi tessellation (AEWCVT) algorithm. In echocardiogram images, bright regions usually represent heart tissues, whereas the dark regions often correspond to chambers (e.g., ventricle and atrium). The recovery of anatomical structures is easier once the chambers are separated from the heart tissue. The bright and dark regions are partitioned into approximately convex regions using a new convexity pursuit segmentation algorithm.

An original echocardiogram image overlapping with a sector region of interest (ROI) is illustrated in fig. 1. Initially, ROI containing anatomical structures is detected. An ultrasound scan sector in an image is usually bounded by dominant

lines as shown in figure 1.

1


Page 02 of 8

Figure 1

Using a Hough transform in polar coordinates (r, t) on the edge-filtered image, dominant orientation with maximum hits in the Hough transform. Thereafter, the region of interest is recovered as a region bounded by these intersecting lines

within the maximum possible radius of the sector.

A 2D gray scale image is regarded as a function u defined on a domain Ω⊂ R2, where, values of u represent the intensity

of the pixels. The image is defined as a set D = {(i, j) |i = 1. . . I, j =1. . . J}, where I and J are height and width of the

image, respectively. Given L number of clusters, the EWCVT algorithm groups each pixel into a cluster by minimizing the following objective function

2


Page 03 of 8

where

are cluster centers, and

                                     is segmentation corresponding to each cluster. is a neighborhood region with size ω centered at pixel (i, j). X(i, j) (i′, j′) is an indicator function wh...