Adjoint pairs of image warping operators for motion modeling in 4D-CT

In 4D-CT, a 4D image (3D+time) needs to be reconstructed from projection data of a moving object. A simple approach is to acquire a number of consecutive 3D-CT scans (called subscans) and to then reconstruct them separately. When the object is moving rapidly, or a high time resolution is desired, a short acquisition time is needed for the subscans, leading to a low number of projections and a strongly underdetermined system to solve. Recently, a number of methods have been proposed to alleviate this issue using image warping operators. By warping an image from a certain time frame to a neighboring time frame, the image can be compared to the projection data of that neighboring time frame, reducing underdetermination. The motion that is used for the warp can be either estimated in advance, or it can be reconstructed jointly with the images. To invert a model that incorporates warping operators, the adjoints of these warping operators are needed. The current methods approximate these adjoint operators by warping operators for an approximate inverse flow. We propose to use the exact adjoints instead, which leads to faster convergence, less computation and less memory requirements.

Jaar van publicatie:2020

Trefwoorden:P3 Proceeding

Toegankelijkheid:Open