Neural network-based filter for medical ultrasonic images

In this paper. an important class of nonlinear adaptive speckle filter. called "segmentation-based filter", has been used to suppress speckles with few detail lost and edge fuzziness. The initial image is first segmented into regions of different tissue and lesion characteristics using a self-creating and organizing neural network (SCONN) based on fractal features. Then each of the segmental regions is processed by a different filter parameter. SCONN is a modified self-organizing neural network (SONN). which can search for an optimal number of output nodes automatically and has no dead center nodes and boundary effect. Experimental results of several sectional ultrasonic images show that our method can filter the medical ultrasonic images efficiently and proved to be superior to traditional filters.