Improving a Rapid Alignment Method of Tomography Projections by a Parallel Approach

Abstract

The high resolution of synchrotron cryo-nano tomography can be easily undermined by setup instabilities and sample stage deficiencies such as runout or backlash. At the cost of limiting the sample visibility, especially in the case of bio-specimens, high contrast nano-beads are often added to the solution to provide a set of landmarks for a manual alignment. However, the spatial distribution of these reference points within the sample is difficult to control, resulting in many datasets without a sufficient amount of such critical features for tracking. Fast automatic methods based on tomography consistency are thus desirable, especially for biological samples, where regular, high contrast features can be scarce. Current off-the-shelf implementations of such classes of algorithms are slow if used on a real-world high-resolution dataset. In this paper, we present a fast implementation of a consistency-based alignment algorithm especially tailored to a multi-GPU system. Our implementation is released as open-source.

A modular software framework for the design and implementation of ptychography algorithms

“…Similarly to Zhang et al (2013) ; Tripathi, McNulty & Shpyrko (2014) ; Loetgering et al (2015) ; Dwivedi et al (2018) , it employs a 2D cross correlation signal, calculated at some points in the reconstruction chain. The same principle is used also for CT alignment in Gürsoy et al (2017) ; Guzzi et al (2021c) , as the synthesised projection o j ( x , y ) are inevitably centred ; when confronted with refined estimates ( , XCORR A ), or measured data ( I ( x , y ), XCORR B ), geometrical shifts can then be measured. Figure 2 shows how such position refinement scheme can be introduced in the canvas of a PIE reconstruction algorithm: 2D weighted phase correlation ( Guizar-Sicairos, Thurman & Fienup, 2008 ) here is used to determine the shift between two different estimates o j and (switch in position XCORR A in Fig.…”

“…To do so, a fast subpixel registration algorithm ( Guizar-Sicairos, Thurman & Fienup, 2008 ) has also been implemented via a PyTorch GPU code. This latter element has many uses, for example in CT alignment ( Guzzi et al, 2021c ), or super-resolution imaging ( Guarnieri et al, 2021 ; Guzzi et al, 2018 ). The details are described in the “method section”.…”

A Parameter Refinement Method for Ptychography Based on Deep Learning Concepts

“…Even if in the literature, many solutions have been proposed, in most cases, the problems are typically tackled independently. Partial coherence and mixed-state ptychography have been extensively reviewed (e.g., in [3,27,28]); the same is valid for the position refinement problem [29][30][31][32][33][34][35], with methods that are quite similar to what is performed in, e.g., super-resolution imaging [36] or CT [37]. To understand the importance of positions for ptychography, Figure 4 illustrates a slightly exaggerated condition: note that the entire object computational box changes format, and this is detrimental, especially from the implementation point of view.…”

Automatic Differentiation for Inverse Problems in X-ray Imaging and Microscopy

“…A simple correction model employs serial cross-correlation between projections [ 83 ], also with cosine-stretching, reducing the risk of potentially propagating drifts in the correction [ 84 ]. Even if, in some cases, an algorithm might be able to create a set of landmarks from image features alone [ 85 , 86 , 87 , 88 ], marker-based alignment methods [ 84 , 89 , 90 , 91 , 92 , 93 ] provide the de-facto solution to the problem; at the cost of decreasing the sample visibility, gold-nanobeads are added to the sample solutions, providing a set of landmarks that are relatively easy to detect and track automatically (but manual tracking is often required for a variety of reasons [ 75 , 94 ]). Stretched cross-correlation is still used as a pre-alignment step.…”

“…A different class of algorithms is formalised around the concept of tomography self-consistency , which tries to infer the alignment parameters while reconstructing the volume. These methods can exploit both gradient-less [ 81 , 94 ] or gradient-based optimisation [ 95 , 96 ] to infer the parameters of a model significantly more complex than the one in Equation ( 6 ).…”