Research On Ultrasound Image Restoration Approaches

Ultrasound imaging has become one of the most commonly used medicaldiagnosis modes, due to its noninvasive nature, portability, security, low cost, andreal-time imaging. Unfortunately, compared to X-ray, CT and MRI, ultrasoundimaging suffers from its poor resolution, and exhibits coarse speckles. The lowresolution of ultrasound imaging is caused by the convolution between tissuereflectivity and point spread function (PSF) plus the environment noise. In order tosolve the problem, a image restoration processing should be performed. Theresearch of this paper is carried on around the ultrasound image restorationapproach.First, the linear model of ultrasound imaging and the basic working principle ofField II ultrasound simulation platform is analysed and studied. In order to obtaingood signal-to-noise ratio and provide a stable premise for later work, waveletthreshold shrinking method is adopted for denoising of the received RF signal.Several threshold functions are analysed and compared, and the simulation resultsindicate that denosing with half soft threshold function is the best.Then, various PSF estimation methods are studied comparatively. The Field IImethod, the complex cepstrum method and a new generalized homomorphicfiltering method are firstly studied. For the phase wrapping problem exisitinginevitably in the last two methods, the least square phase unwrapping algorithm isadopted. According to the simulation results, with PSF obtained by the Field IImethod as a reference, PSF corresponding to the generalized homomorphic filteringmethod is superior to that obtained by the complex cepstrum method.At last, various classic image restoration methods are studied and analyzed.Wiener filtering is a conventional and effective deconvolution algorithm, and theparameter signal-to-noise ratio has important effect on the deconvolution result. Inthis paper, segmentation Wiener deconvolution method is employed. Consideringthe dynamic variability of signal-to-noise ratio for the whole image, one novelapproach that signal-to-noise ratio of each segment image is estimated andsubsequently used to the corresponding segment image deconvolution is put forward.The simulation results indicate that ultrasound imaging resolution is significantlyimproved, and meanwhile the noise is suppressed effectively, better than the effectof traditional deconvolution just using a fixed signal-to-noise ratio.