Registration of Fresnel diffraction patterns for X-ray phase nanotomography
Résumé
X-ray phase contrast imaging is a relatively new imaging modality which offers several orders of magnitude higher sensitivity compared to attenuation imaging [1] [2]. This is especially interesting for the imaging of soft tissue [3]. The X-ray phase cannot be measured directly, however. A simple way of achieving phase contrast is to let the beam propagate in free space after the sample. Several phase retrieval algorithms from such images have been developed [4]. Most of these rely on linearisation of the Fresnel diffraction formula to yield linear filtering based algorithms. The phase maps can be fed to a tomographic reconstruction algorithm to yield a 3D reconstruction of the refractive index decrement distribution [5].Information transfer from phase shift to contrast in the low spatial frequency range is weak, which makes phase retrieval from Fresnel diffraction patterns sensitive to low frequency noise [6]. Several approaches to overcome this limitation have been proposed [6-9]. One particularity of the CTF-based methods [6, 8, 9] is that they allow the use of several phase contrast images taken at different sample-to-detector distances, yielding a better coverage of the phase information in the frequency domain. Indeed, using several distances seems necessary to achieve good image quality in high-resolution imaging [10]. The images at different distances will be out of alignment due to instrumental imprecisions, drift and vibrations. This problem has only been mentioned in passing in literature, despite being non-trivial. It is a registration problem where at lower resolution the contrast is reasonably similar, and at higher resolution the contrast can be very different, between the different distances. To our knowledge, this problem has so far been handled using correlation based methods. In our experience, correlation works well at relatively low resolutions and in the absence of strong artifacts, but fail more often as resolutions become higher. Therefore, the aim of this work is to investigate the use of different registration algorithms for the alignment of Fresnel diffraction patterns acquired at different distances. The influence of misalignments on the retrieved phase maps and 3D reconstructions is investigated using simulated data. Then, different registration algorithms are evaluated using experimental phase nano-CT data from the new Nano-Imaging beamline ID16A at the ESRF, Grenoble, France, and compared to the previously used correlation-based method. This work was supported by the ANR grant ANR-14-CE35-0030-01, the Région Rhône-Alpes, and LabEx PRIMES (ANR-11-LABX-006”). We thank the ESRF for support through the LTP MD830.References[1]R. Boistel et al., Plos one 6, e22080 (2011).[2]I. Zanette et al., Physica Status Solidi A 208, 2526 (2011).[3]J. P. Guigay et al., Opt. Lett. 32, 1617 (2007).[4]M. Langer et al., Med. Phys. 35, 4556 (2008).[5]P. Cloetens et al., Appl. Phys. Lett. 75, 2912 (1999).[6]M. Langer, P. Cloetens, F. Peyrin, IEEE Trans. Imag. Process. 19, 2428 (2010).[7]D. Paganin et al., J. Microsc. 206, 33 (2002).[8]M. Langer et al., Opt. Lett. 37, 2151 (2012).[9]M. Langer et al., Phil. Trans. R. Soc. A 372, 20130129 (2014).[10]M. Langer et al., Plos one 7, e35691 (2012).