Filter optimization method for fdk reconstruction algorithm under cl imaging

Abstract

The application relates to the technical field of X-ray CL imaging, in particular to a CL imaging FDK reconstruction algorithm filter optimization method; the method comprises the following steps: taking a filter as a to-be-optimized variable, taking minimization of a square sum of errors of a reconstructed image and a label image as an objective, and establishing a filter optimization model based on a discrete matrix operator form of the FDK reconstruction algorithm under CL imaging; S-L filters are adopted for parameter initialization; a plurality of three-dimensional images are randomly generated as label images, corresponding projection data is generated by using a CL simulation program, a neural network equivalent to the FDK algorithm is constructed, forward calculation is carried out to obtain a reconstructed image, filter parameters are updated through back propagation and a gradient descent method, and the optimized filter is obtained after iteration until loss convergence or a maximum training number is reached. The application is applied to actual CL projection data reconstruction, and the quality of the reconstructed image is improved.

Images