Selected presentation abstracts, with an informal introduction from the speakers.
TIMBIR - A method for high temporal resolution tomographic reconstructions
Advanced Photon Source, beamline 2-BM
It all started in the year 2011. For the first time in my life, I experienced the winter snow and ice, and with it came the window frost - those beautiful intricate 2-dimensional structures that form on the insides of window panes of houses and cars. I was intrigued and amazed at just how nature managed to make such a simple phenomenon of vapour condensation so pretty. Little did I realize that a good part of my PhD will be dedicated to imaging such structures albeit in metallic samples.
In the year 2012, I started working on my PhD with Prof. Charles Bouman at Purdue University. We met Prof. Peter Voorhees from Northwestern University who was trying to observe the growth of metallic dendrites. The basic idea was to heat a metallic sample until it becomes a liquid. It is then cooled at a constant rate until solids appear in the form of dendrites. The kinetics of dendritic growth and morphology have profound technical implications. The objective was to image these dendrites at sufficient resolution to make quantitative claims about the physics of dendritic growth. Using the best state of the art techniques available only at a synchrotron, they could only image at a temporal resolution of around 20 seconds. In contrast, we will need temporal resolution of well under 2 seconds to even observe dendritic growth. We knew it was time to go back to the drawing board and come up with a radical new imaging technique.
We designed a new method called TIMBIR that stands for time interlaced model-based iterative reconstruction. TIMBIR is the synergistic combination of two innovations – a new interlaced view sampling method, and a novel 4D model-based iterative reconstruction (4D-MBIR) algorithm. In interlaced view sampling, the view angles in multiple 180-degree rotation cycles are interlaced with respect to each other. The 4D-MBIR algorithm is used to reconstruct the interlaced data at any desired temporal resolution. 4D-MBIR uses a prior model to enforce sparsity across space and time and a noise model to model the effects of noise in the data.
Next, it was time to do some real experiments at the APS at Argonne National Lab. We were allocated three days of beam time at the 2-BM beamline. It was my first time at a synchrotron. I stood in awe appreciating the engineering marvel that surrounded me. I don’t recall sleeping for more than a total of 6 hours in the next 3 days. Two days passed with no useful data. In the first day, the beamline scientists helped us fix hardware issues. For the next two days, we tried to synchronize our data acquisition with the time frame of dendritic solidification. It was just a few hours before the end of beam time when we finally successfully managed to image the 3D morphology of growing dendrites in real time. We achieved the impossible and successfully imaged the formation of dendrites at time steps of 1.8 seconds and at a resolution of 0.65 micro-meters.
Dentrite growth
Abstract
TIMBIR – A METHOD FOR HIGH TEMPORAL RESOLUTION TOMOGRAPHIC RECONSTRUCTIONS
K. A. Mohan1, S. V. Venkatakrishnan2, J. W. Gibbs3, E. B. Gulsoy4, X. Xiao5, M. De Graef6, P. W. Voorhees4, and C. A. Bouman1
1. Electrical & Computer Engineering, Purdue University, West Lafayette, IN, USA
2. Lawrence Berkeley National Lab, Berkeley, CA, USA
3. Los Alamos National Lab, Los Alamos, NM, USA
4. Material Science and Engineering, Northwestern University, Evanston, USA
5. Argonne National Lab, Lemont, IL, USA
6. Materials Science and Engineering, Carnegie Mellon University, Pittsburgh, PA, USA
Synchrotron X-ray computed tomography (SXCT) is increasingly being used for time-space 4D imaging of dynamic physical processes. However, the high data rate requirements imposed by the conventional reconstruction methods limit the temporal resolution of 4D reconstructions. In a synchrotron beamline, the data rate is typically limited by factors such as the camera frame rate and data transfer bandwidth. The temporal resolution of conventional methods is insufficient to observe the 3D morphology of important physical processes such as dendritic solidification in metallic alloys [1,2]. Furthermore, measurement non-idealities cause ring and streak artifacts in the reconstruction. These artifacts are typically corrected using low-pass filtering methods that also reduce the spatial resolution of reconstructions.
We present the method of time-interlaced model-based iterative reconstruction (TIMBIR) [1] that enables us to perform real-time imaging of extremely fast physical processes. TIMBIR is the synergistic combination of two innovations. The first innovation is a novel data acquisition method called interlaced view sampling that distributes the angular views of the object more evenly across time. The second innovation is a new 4D model-based iterative reconstruction (MBIR) algorithm that takes advantage of the interlacing of views to produce a synergistic improvement in spatial and temporal resolution of reconstructions. We also correct for ring and streak artifacts by modeling the cause of these artifacts in the algorithm.
To validate the performance benefits obtained using TIMBIR, we performed dendritic growth experiments [1,2] in an Al-Cu alloy using synchrotron X-rays. In the conventional approach, we did a filtered back projection reconstruction of data where the adjacent angles were equally spaced with respect to each other (progressive views). The reconstruction using the conventional method and TIMBIR is shown in Fig. 1. The two experiments were performed at the same maximum allowable data rate. Using TIMBIR, we were able to reconstruct the dendritic growth at a resolution of 0.25 seconds and a temporal rate that is 32 times greater than the conventional method [1].
Figure 1: Comparison of TIMBIR with the conventional method. The first row shows a 2D slice at a fixed point in time. The second row shows a 1D slice as a function of time.
References
- K. A. Mohan et al., "TIMBIR: A Method for Time-Space Reconstruction from Interlaced Views," IEEE Transactions on Computational Imaging, 1-2, p. 96-111 (2015).
- J. W. Gibbs, et al., “The three-dimensional morphology of growing dendrites,” Nature Scientific Reports, 5, 11824 (2015).
- Project was supported by an AFOSR/MURI grant #FA9550-12-1-0458 and use of the Advanced Photon Source was supported by the U.S. DOE under Contract No. DE-AC02-06CH11357
Kadri Aditya Mohan
I am a PhD candidate at the department of electrical and computer engineering at Purdue University. I design and implement model-based reconstruction algorithms for various imaging modalities such as X-ray absorption and phase contrast tomography, vector field electron tomography, etc. Model-based algorithms have been widely used in the field of medical imaging to greatly reduce dose, decrease noise, and improve reconstruction quality. However, these algorithms are yet to be widely adopted by the materials and biological imaging communities at synchrotron light sources.
In my spare time, I like to play tennis or squash. I also like to go swimming.
Diamond Light Source is the UK's national synchrotron science facility, located at the Harwell Science and Innovation Campus in Oxfordshire.
Copyright © 2022 Diamond Light Source
Diamond Light Source® and the Diamond logo are registered trademarks of Diamond Light Source Ltd
Registered in England and Wales at Diamond House, Harwell Science and Innovation Campus, Didcot, Oxfordshire, OX11 0DE, United Kingdom. Company number: 4375679. VAT number: 287 461 957. Economic Operators Registration and Identification (EORI) number: GB287461957003.