1Department of Electrical Engineering, Faculty of Science and Technology, Keio University, 2Institute of Information Sciences and Electronics, University of Tsukuba, 3Department of Obstetrics and Gynecology, School of Medicine, Keio University
Extraction of organ boundaries from ultrasonic echo images is essential to obtain a hign quality three-dimensional (3 D) image in 3 D display techniques. Automatic detection of the boundaries is difficult due to poor intensity contrast, echo drop-outs, and speckle noisy nature of the images. In this paper, we present a new approach that employs fuzzy reasoning techniques to automatically extract the boundaries of ultrasonic echo images. In the proposed method, each voxel is classified as having one of three attributes, liquid (e. g., amniotic fluid, blood, etc.), boundary, and softtissue. Two feature parameters are used as the input to a fuzzier unit. These are the intensity local mean, and the distance between the center of the reference region and the center of gravity of intensities in the 5×5×5 voxels reference region. In our approach, the images are first normalized to avoid excessive brightness or darkness of the whole echo images. Then the feature parameters are calculated and input into the fuzzier unit consisting of a fuzzification part and a fuzzy inference engine. After this step, the grades or possibilities of each voxel attributable to the liquid, boundary and softtissue, are generated. Thereafter, the generated grades are input into a defuzzier unit. Our defuzzier unit differs from other conventional fuzzy logic controllers, in thslt it employs relaxation techniques, and consists of several logical rules as local constraints. After defuzzification, the attribute of each voxel whether it be liquid, boundary or softtissue, is determined. The experimental results of our approach for the clinical data indicate that organ boundaries are pertinently extracted. The proposed approach may be expected as an effective boundary extraction method for the future for obtaining a high quality 3 D image of each part of an organism from ultrasonic echo images.