A convolutional neural network for fast upsampling of undersampled tomograms in X-ray CT time-series using a representative highly sampled tomogram

Bellos, Dimitrios; Basham, Mark; Pridmore, Tony; French, Andrew P.

Abstract

We designed a convolutional neural network to quickly and accurately upscale the sinograms of x-ray tomograms captured with a low number of projections; effectively increasing the number of projections. This is particularly useful for tomograms that are part of a time-series, as in order to capture fast-occurring temporal events, tomograms have to be collected quickly, requiring a low number of projections. The upscaling process is facilitated using a single tomogram with a high number of projections for training, which is usually captured at the end or the beginning of the time-series when capturing the tomogram quickly no longer needed. Abstract X-ray computed tomography and, specifically, time-resolved volumetric tomography data collections (4D datasets) routinely produce terabytes of data, which need to be effectively processed after capture. This is often complicated due to the high rate of data collection required to capture at sufficient time-resolution events of interest in a time-series, compelling the researchers to perform data collections with a low number of projections for each tomogram in order to achieve the desired 'frame rate'. It is common practice to collect a representative tomogram with many projections, after or before the time-critical portion of the experiment without detrimentally affecting the time-series to aid the analysis process. In this paper we use this highly-sampled data to aid feature detection in the rapidlycollected tomograms by assisting with the upsampling of their projections, which is equivalent to upscaling the θ-axis of the sinograms. In this paper, we propose a super-resolution approach based on deep learning (termed an upscaling Deep Neural Network, or UDNN) that aims to upscale the sinogram space of individual tomograms in a 4D dataset of a sample, using learnt behaviour from a dataset containing a high number of projections, taken of the same sample occurring at the beginning or the end of the data collection. The prior provided by the highly-sampled tomogram allows the Journal of Synchrotron Radiation research papers 2 application of an upscaling process with better accuracy than existing interpolation techniques. This upscaling process subsequently permits an increase in the quality of the tomogram's reconstruction, especially in situations that require capture of only a limited number of projections, as is the case in high-frequency time series capture. The increase in quality can prove very helpful for the researchers, as downstream it enables easier segmentation of the tomograms in areas of interest, for example. The method itself comprises a convolutional neural network (CNN) which through training learns an end-to-end mapping between sinograms with low and high number of projections. Since datasets can differ greatly between experiments, our approach specifically develops a lightweight network that can easily and quickly be retrained for different types of samples. As part of the evaluation of our technique we present results with different hyperparameter settings, and have tested our method on both synthetic and real-world data. In addition, we have released accompanying real-world experimental datasets in the form of two 80GB tomograms depicting a metallic pin that undergoes corruption from a droplet of saltwater , and also produced and released a new engineering-based phantom dataset, inspired by the experimental datasets.

Citation

Bellos, D., Basham, M., Pridmore, T., & French, A. P. (2019). A convolutional neural network for fast upsampling of undersampled tomograms in X-ray CT time-series using a representative highly sampled tomogram. Journal of Synchrotron Radiation, 26(3), 839-853. https://doi.org/10.1107/s1600577519003448

• Journal Article Type: Article
• Acceptance Date: Mar 11, 2019
• Online Publication Date: May 1, 2019
• Publication Date: May 1, 2019
• Deposit Date: Apr 12, 2019
• Publicly Available Date: Apr 26, 2019
• Journal: Journal of Synchrotron Radiation
• Print ISSN: 0909-0495
• Electronic ISSN: 1600-5775
• Publisher: International Union of Crystallography
• Peer Reviewed: Peer Reviewed
• Volume: 26
• Issue: 3
• Pages: 839-853
• DOI: https://doi.org/10.1107/s1600577519003448
• Keywords: Nuclear and High Energy Physics; Instrumentation; Radiation
• Public URL: https://nottingham-repository.worktribe.com/output/1786007
• Publisher URL: http://scripts.iucr.org/cgi-bin/paper?S1600577519003448
• Additional Information: Publication: Journal of Synchrotron Radiation; Content type: research papers; Article metrics: Available; Peer reviewed: Yes; Review process: Single blind; Received: 6 July 2018; Accepted: 11 March 2019; Published online: 23 April 2019; Copyright: © 2019 Dimitrios Bellos et al.; Licence: Creative Commons Attribution (CC-BY)
• Contract Date: Apr 12, 2019