Noise reduction of echocardiography images using Isomap algorithm

Abstract

Medical applications of ultrasound imaging have expanded enormously over the last two decades. De-noising is challenging issues for better medical interpretation and diagnosis on high volume of data sets in echocardiography. In this paper, manifold learning algorithm is applied on 2-D echocardiography images to discover the relationship between the frames of consecutive cycles of the heart motion. By this approach, each image is depicted by a point on reconstructed two-dimensional manifold by Isomap algorithm and similar points related to similar images according to the property of periodic heartbeat cycle stand together. Noise reduction is achieved by averaging similar images on reconstructed manifold. By comparing the proposed method with some common methods and according to qualitative expert's opinions, the proposed method has maximum noise reduction, minimum blurring and better contrast among the similar methods.