Filtering Method Of Medical Ultrasonic Image Based On Independent Component Analysis

Ultrasonic imaging has been widely applied in clinic diagnoses because of its real-time visualization, non-invasiveness, low price and convenience. However, it suffers from severe speckle noises. The quality of medical ultrasonic image should be improved for image analysis. Image filter is the common-used method to improve the quality. Until now, a lot of filtering algorithms have been proposed, including median filter, wavelet transform based and anisotropic diffusion based filters. The quality of the image is improved to some extent, and to satisfy some determinate applications. However, they do not solve the defects of the ultrasound medical image radically. According to the ultrasonic medical image noise model put forward by Jain, it can be deduced that speckle is multiply mixed with tissue structure signal, so ultrasonic medical image de-noising is a complex and difficult work. Independent component analysis (ICA) is a new signal decomposing technique developed in recent years. Based on the statistical independence between com-ponents, the observed mixture signal can be decomposed without any prior infor-mation about the components. Ultrasound medical image is nonlinearly mixed with some unknown echo signals which can be regarded as independent components. In this point, independent component analysis offers a means to decompose the components of ultrasound medical image. In this thesis, a new filtering method is proposed based on ICA. Since up to now, nonlinear ICA theory and algorithm is not fully developed, so for the study of ICA's application in ultrasonic medical image filtering, linear ICA theory and algorithm is utilized in this thesis. It is necessary to transform the multiplicative model to additive model based on Jain ultrasonic speckle noise model firstly for the application of ICA algorithm. The filter procedure can be described as following steps: firstly, Jain's speckle noise model is employed on the image; secondly, linear ICA algorithm is used to decompose the image into independent components; and finally, post-process algorithm is introduced to get the de-noised image. The filter is validated by applying it to ultrasonic medical images. Experimental result shows that the image detail is maintained ideally during filtering. With three ultrasonic medical image quality evaluation criterions, the performance of the filter is evaluated quantitatively. But comparing with the wavelet based filter, the proposed filter is not so ideal in depressing noise. Further study and effort is needed to improve the performance. The results of this study can be used in the analysis of ultrasonic radio frequency (RF) signal, and improve the quality of the image radically. The filter based on ICA is to filter the image from a completely new point of view. It's quite different from other filters reported.