Region of interest labeling of ultrasound abdominal images using hausdorff distance

Abstract

This paper presents a two stage approach to segment the ultrasound abdominal images. During first stage, images are segmented into different regions which are stored in the database along with their statistical properties. In second stage, different segmented regions in the database are used for Region of Interest labeling of ultrasound images using hausdorff distance. The quality of ultrasound images are strongly affected by the presence of speckle noise. As Speckle noise is multiplicative in nature, so homomorphic filtering is found to be best suited to reduce such a noise from the images. To segment out the region for the learning purpose, improved marker-controlled watershed segmentation algorithm is used. But this normally results in over segmentation. This over segmentation problem is solved by defining the internal markers on the image using morphological operations. All the regions segmented out in this step are stored in the database along with their statistical properties. Finally, a template matching approach is proposed to match the regions of a given ultrasound image with already stored template regions using hausdorff distance. If a region in the image matches with the template, it is appropriately labeled as per the other details stored in the database. It is observed that the efficiency of whole system depends upon the efficiency of learning stage. Although the computational complexity of system is very high during learning stage, but it is relatively very less during implementation stage. The overall efficiency of the system is found to be very good in terms of precision and recall parameters. © Springer Science+Business Media B.V. 2008.