Certain investigations on denoising techniques for brain MR images using curvelet transform

Abstract

Recent developments in the computerized analysis of medical images are expected to aid radiologists and other healthcare professionals in various diagnostic tasks of medical image interpretation. In Magnetic Resonance (MR) imaging, the accuracy of diagnosis and assessment of a disease depends mainly on the quality of image acquisition and image interpretation. The focus of this thesis is to extend the application of curvelet transform, for the purpose of image denoising and restoration of brain MR images. In particular, the aim is to develop an efficient artifact free edge preserving medical image denoising method using curvelet transform, assess and compare their performance in such a way to improve the reconstructed brain MR image quality in order to establish an effective CAD. Our novel curvelet-domain filtering procedure overcomes this difficult estimation problem. The objective of this technique is to develop a filtering method to estimate the signal from the magnitude image data. This thesis proposes an improved algorithm in view of the shortcoming of wavelet transform in preserving edges. After applying two dimensional wavelet transform, edge and detailed information are lost appreciably. But in our proposed curvelet transform based algorithm edge and detailed information are retained better than wavelet transform to a level of 40%. Analyzing the PSNR value change tendency of denoisong image among Wavelet, curvelet transform and curvelet improved method. MR image in the situation of and#963; lt 35, PSNR value of curvelet transform improved method denoising is relatively high. With the noise size increasing, PSNR value of curvelet improved method is gradually consistent with curvelet transform, but still surpassed the wavelet transform. CT image in the situation of and#963; and#8804;50, PSNR value of curvelet improved method denoising always surpassed other wavelet methods. newline