A frequent problem in the field of image processing nowadays is to obtain images close to ideal images from noisy ones.
In particular, the inefficiency of detectors in the field of diagnosis based on images has made this problem even more serious. The most common feature of noise in medical images is two: the probability distribution is iterative, and the noise is non-linearly changed through the reconstruction process, resulting in a lower resolution of the resulting image data.
The edge-preserving filter does not degrade the image resolution as it does not smooth the image data, but removes noise by suppressing only noise signals. Therefore, it is widely used in image diagnosis with high accuracy for tumors or patient organs. In these applications, filters that smooth the boundaries and displacements lose diagnostic significance because they increase the ambiguity between the critical parts of diagnostic significance and reduce the accuracy of images.
There are two problems with using a boundary-preserving filter.
First, almost all edge-preserving filters can be applied only to typical Gaussian noise filters. Therefore, these types of filters cannot be applied to image data such as low-dose CT image noise. Because the pixel values in these images follow the Poisson distribution, noise variance is not the same throughout the image.
Second, when using these types of filters, certain information about the noise variance at any location should be available for local or global parts of the image.
On the basis of the analysis of the path characteristics rather than noise, Ri Hwi Song, a researcher at the Faculty of Biology and Medicine Engineering, has proposed a sinogram estimation method based on various optimization methods and a corresponding algorithm to recover the ideal image by maximally preserving information only in the sinogram data distorted by various factors and removing noise.
First, he analyzed the path characteristics of the sinograms in low-dose CT images and used optimization methods to optimize the information content of the sinograms in low-dose CT images. On this basis, he proposed a method to remove the noise components and a corresponding algorithm for their implementation. Then, he analyzed the results of sinogram refinement using the optimization method, comparing the image obtained through image reconstruction with the ideal CT image data, and proposed some methodologies to speed up the algorithm.
Qualitative analysis of this image quality improvement method shows that the sinogram estimation algorithm for low-dose CT using the optimization method can effectively remove noise while preserving both image structure and boundary even in the presence of large noise and non-stationary noise as in low-dose CT. He has also confirmed that decreasing the operation time of the sinogram estimation algorithm using the optimization method could lead to the possibility of drastically decreasing the dose in CT.
You can find the details in his paper “Sinogram restoration based on shape property in computed tomography” in “Informatics in Medicine Unlocked” (SCI).
© 2021 Kim Chaek University of Technology