A novel extension to non-local means algorithm: Application to brain MRI de-noising

摘要

Image de-noising is an essential intermediate step in several medical applications related to brain MRI. The noise present in brain MRI degrades the performance of computer-aided analysis of these images. Therefore, the noise should be removed prior to subsequent processing. Non-local means (NLM) is a classical de-noising algorithm, which has been successfully applied for de-noising of brain MRI. In this work, a variant of traditional NLM, called improved adaptive non-local means (IANLM), has been extended for application to brain MRI. The new algorithm is named as extended non-local means (ExNLM) algorithm. To be precise, the IANLM algorithm is adapted to Rician noise inherently present in MR images by using a Ricain bias correction procedure. Moreover, a wavelet coefficient mixing procedure is proposed which exploits valuable information in different sub bands of over- and under-smoothed images, obtained by using the IANLM algorithm with different set of parameters. Finally, optimization of the proposed ExNLM algorithm is performed for brain MR images de-noising. Different variants of the proposed algorithm have been validated on a simulated brain MRI dataset. Quantitative and qualitative results indicate that the proposed algorithm suppresses the noise in brain MR images effectively, and outperforms several contemporary methods of de-noising.