An evolutionary algorithm to automate noise reduction in MR images

Noise in MR images is an important concern that undermine their diagnostic accuracy. Wavelet based noise removal has been used in past studies to remove noise without introducing a significant image distortion which is associated with other denoising methods. Wavelet based method involves the selection of a few parameters for optimum denoising. Usually, this selection process is based on human observation which may result in selection bias. To avoid this bias, we propose a purely objective and automatic selection process using an evolutionary algorithm called differential evolution (DE). DE was used to select the optimum set of wavelet parameters for noise reduction and Universal image quality index (UIQI) was used as the objective function within DE. The results showed that this objective selection method provides a means to automate the wavelet based denoising process and improves the SNR by a considerable factor.