Conference Paper

Statistical characteristics of portal images and their influence in noise reduction

Book Series cp
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
  • Volumen: 7930 LNCS
  • Número: PART 1
  • Fecha: 23 October 2013
  • Páginas: 386-395
  • ISSN: 03029743 16113349
  • ISBN: 9783642386367
  • Source Type: Book Series
  • DOI: 10.1007/978-3-642-38637-4_40
  • Document Type: Conference Paper
Portal imaging is used in radiotherapy to asses the correct positioning of the patient before applying the treatment. Given the high energy particles used in portal image formation, portal image is intrinsically bound by low contrast and poor spatial resolution. The relevance of portal imaging in radiotherapy treatments and its common use justify efforts to improve its inherent low quality. The knowledge of the statistical properties of both image and noise is essential in order to develop suitable processing algorithms to clean the image. The aim of this paper is to show how the statistical characteristics of the portal images and noise images generated in one of the portal imaging systems most widely deployed, can be exploited to improve the quality of noisy portal images through efficient denoising methods. An ensemble of portal images is used to investigate their statistical characteristics. In the case of noise, a process of averaging and subtraction of the mean is used to extract noise images. The distribution found for the noise is clearly Gaussian, in both the spatial and the wavelet domain. The curves for the noise show a parabolic shape in the semi-log graphs across the different scales, which translates into Gaussian character in the transformed domain. On the other hand, the probability density functions (pdf's) for portal images show large tails. Wavelet thresholding takes advantage of the different statistical features found for noise and signal. In the present work wavelet thresholding is compared to Wiener filtering, and the assesment of the denoised image is carried out by means of the peak signal to noise ratio PSNR and the structural similarity index SSMI. Thresholding the wavelet coefficients of the noisy image gives better denoising results for both figures of merit (PSNR and SSIM) than the Wiener filter in all the analysed cases. Furthermore, the differences between the methods increase as the noise increases. abstract environment. © 2013 Springer-Verlag.

Author keywords

    Indexed keywords

      Funding details