Employing non-local temporal self-similarity across the entire time domain in 4D CT reconstruction

Orateur : Daniil KAZANTSEV (The School of Materials, MXIF, The University of Manchester, UK)

There are many cases where one needs to limit the x-ray dose, or the number of projections, or both, for high frame rate (fast) imaging. Normally it improves temporal resolution but reduces spatial resolution of the reconstructed data. The redundancy of information in the temporal domain (e.g. arising from the self-similar structures) can be employed to improve spatial resolution. In our work, we propose a novel regularizer for iterative reconstruction of time-lapse computed tomography. The penalty term is based on non-local means (weighted graphs) and selectively employs all (not only adjacent time frames) available temporal information to improve the spatial resolution. We also present a new approach using additional structural information (high resolution pre-scan) with the proposed regularizer. This makes the method a flexible tool of achieving a desirable trade-off between the spatial and temporal resolution.

Télécharger dans mon agenda