Available online at www.sciencedirect.com ScienceDirect Physics Procedia 69 (2015 ) 343 348 10 World Conference on Neutron Radiography 5-10 October 2014 imars (imaging Analysis Research Software) Jean-Christophe Bilheux*, Hassina Bilheux Oak Ridge National Laboratory, Neutron Sciences Directorate, Oak Ridge, TN 37831, USA Abstract Image processing has become a mainstream capability with commercial software that allow the general public to perfom, for example, photograph enhancement such as noise reduction or deblurring. Scientific imaging data sets often require quantitative image analysis that can only be performed with careful algorithm development. Recently, a number of software packages have been developed with different image processing capabilities. Most of them require some level of algorithm development to be able to perform data analysis. In 2011, the CG-1D neutron imaging beamline joined the Oak Ridge National Laboratory Neutron Sciences general user program and efforts focused on data processing and analysis software development. The image Analysis Research Software (imars) was developed to respond to the CG-1D user demand and provide semi-automated data normalization, processing and analysis tools in the MATLAB framework. New algorithms are often implemented to the imars framework per user request. A brief description of the neutron imaging beamline scientific community will be presented and will be followed by a detailed overview of the imars software. 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 2015 The Authors. Published Elsevier B.V. Selection and peer-review under under responsibility of Paul of Paul Scherrer Scherrer InstitutInstitut. Keywords: data normalization; image processing; software packages; MATLAB * Corresponding author. Tel.: +1-865-406-1704 E-mail address: bilheuxjm@ornl.gov 1875-3892 2015 The Authors. Published by Elsevier B.V. Selection and peer-review under responsibility of Paul Scherrer Institut. Notice: This manuscript has been authored by UT-Battelle, LLC, under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. 1875-3892 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Selection and peer-review under responsibility of Paul Scherrer Institut doi:10.1016/j.phpro.2015.07.048
344 Jean-Christophe Bilheux and Hassina Bilheux / Physics Procedia 69 ( 2015 ) 343 348 1. The CG-1D neutron imaging beamline scientific community The CG-1D imaging beam line (Crow et al. 2011; Bilheux H. et al. 2014; Santodonato et al. these proceedings) has been in the Oak Ridge National Laboratory (ORNL) neutron scattering general user program since January 2011. The beam line is located at the High Flux Isotope Reactor (HFIR). The CG-1D instrument is a white beam imaging beam line that provides cold neutrons (peak at 2.6 Å) with a flux of ~ 1 x 10 7 n/cm 2 /s at the sample position, when configured with a beam collimation of L/D of 400, a spatial resolution of ~ 100 µm at the detector and a field of view of 7.4 cm x 7.4 cm (using a charge coupled device). Neutron radiography is performed within 60 to 90 s, depending on the sample geometry and attenuation. Neutron computed tomography requires hundreds of projections which are acquired over several hours (usually 12-15 h). Recent upgrades include a micro-channel plate (MCP) detector that is capable of fast frame images for cyclic motion samples or high contrast samples such as fluid flow in a rock matrix (µs and ms acquisition modes, respectively). As part of the implementation of the imaging capability, an effort to develop data reduction and analysis has been pursued. The CG-1D user community covers a broad range of scientific areas such as materials research, geosciences, energy, engineering, transportation research, biology, plant physiology, archeology, neutron imaging technology development, as illustrated in Fig. 1. A survey of the community prior experience in image analysis and software requirements showed different levels of image analysis, from beginner to expert levels but also a need for data normalization, processing and analysis techniques. Most importantly, specific needs to extract quantitative information from the 2-dimensional (2D) and 3-dimensional (3D) data sets were clearly expressed, which agrees with the general trend in the neutron-imaging field across the world (Anderson et al., 2009). Fig. 1.CG-1D neutron imaging proposals submitted for the January to July 2014 run cycle. A thorough survey of commercial image processing tools packages, especially in the field of medical imaging, was conducted to evaluate if a software package and appropriate processing and analysis tools were already available in the field. Well-developed software packages such as ImageJ, a freeware, (Rasband et al., 2014) come with functioning image processing capability, but limited analysis capabilities, unless one is capable of coding using the specific language associated with the package. The survey also indicated that MATLAB (Matlab, 2012) offered the highest versatility and could be used for algorithm development, data visualization and analysis, in a friendly and interactive environment. MATLAB was found to be capable to easily handle large matrices (i.e. 2D radiographs). Moreover, the Image Processing toolbox in MATLAB provided standard algorithms for image processing, segmentation, visualization, noise reduction, registration, along with a platform for algorithm development, all of which made the MATLAB package attractive in terms of having some of the image processing basics already available. Another advantage of the MATLAB package is the ability to develop a
Jean-Christophe Bilheux and Hassina Bilheux / Physics Procedia 69 ( 2015 ) 343 348 345 custom-made graphical user interface (GUI) that could be compiled so that one could utilize the GUI at full capacity without having a MATLAB license. 2. imars 2.1 Normalization The design of imars was driven by the motivation to simplify data normalization or reduction and to provide basic data analyses (segmentation, profile, geometry correction, etc.). As illustrated in Fig. 2, the GUI is developed to be intuitive, with options not being available until all requirements are fulfilled, which reduces the margin of freedom but prevents users from being overwhelmed. imars is capable of reading FITS, TIFF and ASCII files, all of which correspond to the formats radiographs are saved by the CG-1D detector suite. The first step in imars is to select the data set with no limit in the number of radiographs that can be normalized (top left corner in Fig. 2). Then, the open beam and dark field images are loaded in the application (top center and right corner in Fig. 2, respectively). At this point, a region of interest outside of the sample can be selected for normalization and to correct for beam fluctuation. Once all these steps are completed, the option to normalize the data set becomes available and all of the raw radiographs are automatically normalized. Raw data set Open Beam data set Dark Field data set Data visualization Normalization Fig. 2. imars and its normalization menu. Normalized data can be processed using the implemented algorithms such as the cylindrical geometry correction tool, registration and/or realignment of samples, etc. 2.2 Combination of radiographs based on signal-to-noise ratio Radiographs can be combined to improve signal-to-noise ratio (SNR), for which three algorithms are available: average, median or sum of images, respectively. The user selects the signal and background areas for the images that are to be combined. imars calculates and displays the SNR, and the user can select the best option by reading the SNR displayed in imars, as illustrated in Fig. 3. Several radiographs of a short time frame are
346 Jean-Christophe Bilheux and Hassina Bilheux / Physics Procedia 69 ( 2015 ) 343 348 recorded at each angle rather than a long exposure of the same amount of time. To minimize the fluctuating noise from neighboring beam lines, the sum of short time radiographs produce higher SNR when added together than the equivalent longer exposure. This is an important processing capability when for low contrast sample radiographs are collected at different angles for further computed tomography. The sum of images can be used to combine short time radiographs (in the milliseconds range for example) as the neutron statistics per image is low. This processing option allows users to acquire data with short time frames and combine radiographs later, if necessary, rather than acquiring data with longer time frames and taking the risk of missing an important time event by undersampling in time. 2.3 Quantitative data analysis Fig. 3. imars and its normalization menu. Several algorithms have been developed to measure the change in transmission/attenuation (2D data) or extract the linear attenuation coefficient (slices from 3D data set). The most straightforward method, which is available in imars is the selection of a region of interest (ROI) from which an average transmission or attenuation coefficient is calculated. Several ROIs can be selected at once and they can be either merged to obtain an average ROI or treated separately. The results can then be exported as a Comma Separated File format (CSV) for further analysis using a tool such as Microsoft Excel or MATLAB. An example of the use of ROIs for quantitative data analysis of water uptake in French parchment can be found in (Herringer et al. these proceedings). Profiles can be utilized to plot the change in transmission/attenuation and/or attenuation coefficient along a line or circle segment for hundreds of radiographs or slices at once, as illustrated in Fig. 4. An illustration of the use of the profile can be found in (Smith et al. these proceedings).
Jean-Christophe Bilheux and Hassina Bilheux / Physics Procedia 69 ( 2015 ) 343 348 347 Fig. 4. Change in transmission along the selection. 3. Conclusion The imars software has been implemented at the CG-1D imaging beamline in order to provide simple and efficient data processing and analysis tools to the user community. Implementation of this capability has enabled the user community to analyze and publish several manuscripts in peer-reviewed journals, which is more timeefficient than relying on the imaging instrument team, as manpower may be limited. The software also includes segmentation techniques (region-growing method, histogram-based, etc.) and registration of the data. Future upgrades include implementation of new algorithms for segmentation, data registration, filtering, etc. Acknowledgements Use of the CG-1D beam line at Oak Ridge National Laboratory s High Flux Isotope Reactor was sponsored by the Scientific User Facilities Division, Office of Basic Energy Sciences, U. S. Department of Energy. This manuscript has been authored by UT-Battelle, LLC, under Contract No. DE-AC05 00OR22725 with the U.S. Department of Energy. The authors would like to thank all of the CG-1D users who helped, by their interactions and feedback, improve the functionality of imars.
348 Jean-Christophe Bilheux and Hassina Bilheux / Physics Procedia 69 ( 2015 ) 343 348 References Anderson, I. S., McGreevy, R., Bilheux, H. S., Neutron imaging and applications: a reference for the imaging community, Springer, 2009. Bilheux, H. Z., Bilheux, J.-C., Bailey, W. B., Keener, W. S., Davis, L. E., Herwig, K. W., Biomedical Science and Engineering Center Conference (BSEC), 2014 Annual Oak Ridge National Laboratory, DOI: 10.1109/BSEC.2014.6867751, July 2014. Crow, L., Robertson, L., Bilheux, H., Fleenor, M., Iverson, E., Tong, W., Stoica, D., Lee, W., Nuclear Instruments and Methods in Physics Research A 634 (2011) S71-S74. Herringer, S. N., Bilheux, H. N., Bearman, G., these proceedings, 2014. MATLAB and Statistics Toolbox Release 2012b, The MathWorks, Inc., Natick, Massachusetts, United States. Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://imagej.nih.gov/ij/, 1997-2014. Santodonato, L., Bilheux, H., Bailey, B., Bilheux, J., Nguyen, P., Tremsin, A., Selby, D., Walker, L., these proceedings. Smith, T., Bilheux, H., Bilheux, J.-C., Yan, Y., High resolution neutron radiography and tomography of hydride zircaloy-4 cladding materials, these proceedings, 2014.