A novel registration and de-noising method for multiple sequence MRI of atherosclerosis

Multiple sequence magnetic resonance imaging (MRI) is often hampered by noise and misalignment. The wavelet de-noising lacks directivity and does not preserve image details, which limits its effectiveness. To solve the problem, an adaptive threshold de-noising method based on wavelet transformation is employed. As to misalignment, a registration method basis of shape context descriptor is proposed to accurately align multiple MR image sequences. In order to achieve better MRI de-noising capabilities, a new threshold criteria that is adaptive to the noise distribution in the high and low sub bands is developed to replace traditional threshold function used in wavelet transform denoising. Multiple sequences MRI registration are performed by identifying edge points with matching shape context firstly. Then, an iterative procedure is invoked to obtain the deformation field, which determines the B-spline interpolated output image. Experimental results show that the de-nosing method removes image noise effectively whilst preserving original information. The registration method can reach an overlap of 96%±0.8 between multi-sequence MR image.