The use of wavelet filters for reducing noise in posterior fossa computed tomography images

Abstract

The wavelet shrinkage de-noising algorithm with soft thresholding has been widely used to suppress random noise in images. This paper describes an experimental search for the best wavelets, to reduce Poisson noise in Computed Tomography (CT) scans. Five slices of CT containing the posterior fossa from an anthropomorphic phantom and from patients were selected. As their original projections contain noise from the acquisition process, some simulated noise-free lesions were added on the images. After that, the whole images were artificially contaminated with Poisson noise over the sinogram-space. The configurations using wavelets drawn from four wavelet families, using various decomposition levels, and different thresholds, were tested in order to determine denoising performance. The quality of the resulting images was evaluated by using Contrast to Noise Ratio (CNR), Human Visual System absolute norm (H1) and Structural Similarity Index (SSIM) as quantitative metrics. An assessment of the perceptual image quality was developed by using the jackknife free-response (JAFROC) methodology with three experts. The results of the wavelet filters were compared with other common filters in the frequency domain. Results showed that denoising with wavelet filters improved the quality of posterior fossa region in terms of an increased CNR, without noticeable structural distortions. Wavelet filtering is an alternative to be considered for Poisson noise reduction in image processing of posterior fossa images for head CT.