GA - Based Adaptive Wavelet Denoising of Low-Dose Medical Images: Application to EMR Tomograms

Abstract

Medical images, acquired with low exposure to radiation or after administering low-dose of imaging agents, often suffer due to noise arising from physiological sources and from acquisition hardware. The noise can be detrimental to the correct diagnosis well as for the computation of quantitative functional information. To overcome these deficiencies, in this paper we present a genetic algorithm-based wavelet domain denoising technique. The proposed technique optimizes the tradeoff between signal-to-noise ratio (SNR) and resolution. The SNR and Liu's error factor form the basis of an objective function to optimize threshold for each subband across the scales. Based on an object- oriented approach, the proposed algorithm is developed as reusable software component using Java. Evaluation using simulated Shepp-Logan head phantom and real EMR images of phantoms and live animals, and comparison with other state of art wavelet-based denoising methods show the GA-based approach to be superior in terms of visual quality as well as quantitative metrics such as PSNR, RMSE and Liu's error factor, F(I).