Automatic segmentation of ultrasonic image

Automatic segmentation of ultrasonic images is a crucial step in their automatic morphological analysis. The ultrasonic images are used in the diagnosis of human tissues. However, ultrasonic images usually displayed in gray scale suffer from coarse resolution, low contrast and high noise. Interpretation of the ultrasonic image by human eye is difficult and error-prone. These difficulties can be solved by statistical texture measurement. An approach to statistical textural analysis is to extract the textural information from the region of interest by calculating first-order or second-order statistics. Another approach based on fractal geometry has been applied to texture analysis recently. The fractal dimension is an important parameter. Three methods are used to estimate the fractal dimension: i) by determining the average absolute intensity difference of pixel pairs on a surface for various scales; ii) determination from the Fourier power spectrum of the image data; and iii) by determining the number of cubes covered by the surface for various scales. In all three methods, the fractal dimension can be estimated using least-square fitting. In order to assess the effectiveness of the fractal dimension, spatial gray level co-occurrence matrix statistics were also applied for comparison. Image segmentation is achieved using clustering techniques. An unsupervised classification technique is proposed to determine the intrinsic number of clusters in the image data based on a clustering quality parameter. A simple K-means clustering algorithm is applied in the final segmentation.<>