Digital Industrial Radiology and Computed Tomography (DIR 2015) 22-25 June 2015, Belgium, Ghent - www.ndt.net/app.dir2015 More Info at Open Access Database www.ndt.net/?id=18082 Process Control for Computed Tomography using Digital Detector Arrays David Allen FRY 1, Christopher Jason STULL 1, Brandon Michael LATTIMORE 1 1 Nondestructive Testing & Evaluation Group, Los Alamos National Laboratory; Los Alamos, New Mexico USA Phone: +1 5056652916, e-mail: dafry@lanl.gov Abstract Los Alamos National Laboratory has recently completed the qualification of a computed tomography process for the evaluation of small parts. As part of the qualification, ASTM International Digital Radiography (DR) and Computed Tomography (CT) standards were used to develop a process control program. ASTM E2737 and E2698 were used to create a series of baseline tests for the individual DR images that make up a CT data set. Spatial resolution, contrast sensitivity, detector offset, and lag are performed initially and then repeated on a weekly basis to ensure the digital detector array is operating within tolerance. ASTM E1695 spatial resolution and contrast sensitivity tests are used to assess CT baseline and stability. The complexity of these tests, in addition to the frequency at which they are to be executed, necessitated the development of software to automatically analyze the process control data. To this end, Graphical User Interfaces (GUIs) were written using the scripting language, Python. It is the intent of the authors to make this software available to the ASTM community at a later date. Keywords: Computed tomography, digital radiography, process control 1. Introduction Los Alamos National Laboratory (LANL) has been doing Research & Development on digital radiography (DR) and computed tomography (CT) for many years. Recently, DR/CT processes have needed to be qualified for production use. All system variables must be addressed, established, and documented. Part of the qualification is performing a system baseline characterization. Periodic stability tests comparing results to the baseline are then performed to ensure the system is operating within acceptable limits. LANL produces components for the United States government s National Nuclear Security Agency (NNSA). In the NNSA environment, the high level requirements document for DR/CT is RMI T097 Qualification of Digital Radiographic Imaging Techniques [1]. The high level requirements have been incorporated into a LANL Design Agency specification SS6K0461 General Specifications for Digital Radiography of Los Alamos National Laboratory-Designed Parts and Assemblies [2]. Both of these documents rely on implementation of ASTM International standards E1695 [3] and E2737 [4] to satisfy requirements. Table 1 compares the requirements and standards. We aim to make these tests as easy and quick as possible. Currently we have agreed with the Design Agency to perform weekly performance checks. For our 10X magnification CT system based on DDA technology, we perform both DDA and CT checks. We have developed automated and semi-automated methods for analysis of the performance test images.
2. Terminology CNR Contrast to Noise Ratio CR Computed Radiography CSa achievable Contrast Sensitivity (see ASTM E2597 [5]) DDA Digital Detector Array DDL Digital Driving Level ERF Edge Response Function LSF Line Spread Function MTF Modulation Transfer Function PV Pixel Value SRb Basic Spatial Resolution (see ASTM E2597) isrb - Interpolated Basic Spatial Resolution (see ASTM E2597) SNR Signal to Noise Ratio Table 1 Requirement T097 SS6K0461 ASTM/DDA Standard Phantom Spatial X X E2737 E2002 Gauge Resolution/ or line pair MTF/SRb gauge Contrast Sensitivity/ CNR Signal-to-Noise (SNR) X X E2737 Penetrameter/ Shim X X E2737 Penetrameter/ Shim ASTM/CT Standard E1695 E1695 Phantom Uniform Disk Uniform Disk Measure isee SRb Python GUI CT MTF Python GUI CNR Python GUI CT CDF Python GUI or isee SNR Lag X X E2737 None Python GUI Lag Dynamic Range X X E2737 Penetrameters/ Python GUI CNR Shims Dimensional Uniformity X X E2445 CR Phantom isee Distance measure Detector Degradation X X E2737 None Python GUI or isee Display Brightness X X E2737 SMPTE RP133 Calibrated light meter Display Contrast X X E2737 SMPTE RP133 Visual 0% DDL against 5% DDL and 95% DDL against 100% DDL Display Spatial Resolution X X E2737 SMPTE RP133 Visual alternating 1% & and 100% contrast lines
3. DDA Checks ASTM International, E2698 Standard Practice for Radiological Examination Using Digital Detector Arrays [6] calls for tests for monitoring the DDA performance over time shall be performed in accordance with ASTM International, E2737 Standard Practice for Digital Detector Array Performance Evaluation and Long-Term Stability. 3.1 Spatial Resolution An image of the ASTM E2002 Gauge [7] with both vertical and horizontal orientations is captured with the gauge directly on the DDA face. ISee! Professional software [8] contains a modulation calculator based on an averaged line profile through the E2002 gauge see Figures 1 and 2. The line pair gauge is captured and analyzed for the largest pair with < 20% modulation (SRb) or interpolated to find the exact 20% modulation frequency (isrb). Figure 1: ISee! averaged line profile through ASTM E2002 duplex wire gauge (horizontal profile shown)
Figure 2: ISee! measurement of modulation of a wire pair. Three measurement points are user selected (white dashed lines) and the modulation reads in the upper right corner (arrow). 3.2 Contrast Sensitivity and Dynamic Range Images of appropriate material and thickness hole-type penetrameters on a shims is captured with the same parameters as production imaging for the thinnest and thickest sections to be imaged. Our Python CNR routine is used to analyze the images. The user selects the Figure 3: CNR determined from signal and noise inside and outside a penetrameter hole penetrameter hole for which CNR is desired and the routine automatically calculates CNR based on the region between two regions of interest around the hole with sides 2X and 4X the hole diameter and the hole see Figure 3. From CNR, achievable Contrast Sensitivity (CSa) can be calculated. CNR and/or CSa is calculated for both the thinnest and thickest regions of a given part, or an acceptable range of part thickness is determined with minimum CNR/CSa.
3.3 Signal-to-Noise Ratio (SNR) A graph of SNR vs. Exposure is generated from flat field images, see Fig. 4. A quick check of just one point, such as 90% of saturation, on the line is sufficient for the periodic tests. The slope of the line also gives the ASTM E2737 Efficiency of the detector. SNR vs Exposure 1E33 Technique 250 200 SNR = 2256.4*sqrt(E) - 0.2905 SNR 150 100 50 0 SNR = 1053.2*sqrt(E) - 9.139 0 0.05 0.1 0.15 SQRT(mGy/frame) 1 frame 8 frame Linear (1 frame) Linear (8 frame) Figure 4: SNR vs. square root of exposure 3.4 Lag A graph of signal vs. time is generated from a series of images before and after the XGD is shut down, see Fig. 5. The user selects a region of interest and the software builds a table of average signal vs. time for a sequence of images taken with the CT acquisition software with no object.
% of 1st frame PV 150 100 50 0 Shutdown 1E33 Technique (3.0 fps) 0 2 4 6 8 10 12 14 Time (seconds) 1E33 Shutdown Shutdown 1E33 Technique (3.0 fps) % of 1st frame 8 6 4 2 0 y = 5.0495x -0.512 0 5 10 15 20 25 30 35 Shutdown Power (Shutdown) Frames after Transition Figure 5: Lag measurement 3.5 Dimensional Uniformity An ASTM E2445 [9] or similar phantom, see Fig. 6, is measured to check dimensional features are uniform. Capturing this phantom image obtains the E2002 gauges for the spatial resolution check.
3.6 Detector Degradation Figure 6: LANL CR Phantom used for dimensional uniformity An Offset image is analyzed for signal level and SNR, and visually for any pattern. An increase in average offset PV or decrease in SNR signals degradation. 4. CT Checks ASTM E1570 Standard Practice for Computed Tomographic (CT) Examination [10] calls for initial and periodic system performance measurement. These requirements can be satisfied using the methods of ASTM 1695 Standard Test Method for Measurement of Computed Tomography (CT) System Performance.
4.2 Contrast Resolution An appropriate material and diameter uniform disk is scanned with the same parameters as production imaging. For an individual reconstruction slice, extracted from this data set, the user selects a circular region within the disk of material large enough to supply a statistically significant sample of pixels, but small enough to avoid a common reconstruction artifact referred to as cupping. Cupping is an artifact arising primarily from beam hardening (but also from internal scatter), resulting in reduced attenuation values near the centers of uniform cross-sections. These reduce attenuation values imply that the density in the interior region is less than that of the exterior region which is, of course, not the case for uniform cross-sections. Figure 7: Example tile selection from reconstruction slice of uniform disk. After the circular region is selected, the software draws a square, inscribed within the circle is drawn (see Figure 7). That square is then progressively subdivided into square tiles, until the tile size (i.e. edge length of the tile, in pixels) is equal to one pixel. The image below presents an example of this operation for an edge length of the initial square equal to 256 pixels.
Figure 8: Example tiling scheme Having performed this operation, the following mathematical operations are conducted for each set of tiles of a given size (e.g. for all 64 pixel by 64 pixel tiles, then for all 32 pixel by 32 pixel tiles, and so on). Example output from the Python-based GUI is given in Figure 9. 1) Calculate the mean attenuation of each of the tiles. 2) Calculate the standard deviation of the values computed in 1), in order to obtain the standard error in the mean note that this operation yields on values for each set of tiles. 3) Express the standard error in the mean as a percentage of the mean of the values computed in 1) and multiply by 3, which corresponds to a 50% false-negative rate (i.e. the case of threshold detectability).
4.1 Spatial Resolution Figure 9: Contrast resolution plot from Python-based GUI As for the contrast resolution system test, the spatial resolution system test begins with a reconstruction slice of a disk of uniform material. This system test, however, is concerned with the edge response function, and so, the user draws two circles that bracket the edge of the reconstruction slice of the disk; this is illustrated in the figure below. The pixels that fall between those two circles are then sorted and binned, according to their distances from the center of mass of the disk. These binned pixels are then averaged to obtain a single value for each bin. The Edge Response Function (ERF) is then computed as a piecewise cubic polynomial fitted to the averages computed previously using a set (determined by the number of points used to fit the cubic polynomial) of consecutive bins; performing this fit for multiple sets of consecutive bins yields the piecewise component. It is clear that the calculation (or placement) of the center of mass of the disk is important, as asymmetries can cause substantially erroneous results. Therefore, despite its apparent trivial nature, it is worth pointing out specifically that care must be taken in calculating (or placing) the center of mass.
Note the asymmetry of the circular region due to misplacement of the center of mass. Figure 10: Center of mass calculation (or placement) error. Having computed the ERF, the Line Spread Function (LSF) and Modulation Transfer Function (MTF) are reasonably straightforward to obtain. The procedure to compute the LSF begins by fitting a piecewise cubic polynomial to the ERF, essentially smoothing the ERF. For each piecewise fit, the derivative is evaluated at the center of the piecewise window. The ensemble of these derivative evaluations is then normalized by the maximum to arrive at the LSF. Finally, the MTF is calculated simply as the amplitude of the Fourier Transform, normalized to unity at the zero frequency. Example output from the Python-based GUI is given in Figure 11.
Figure 11: Spatial resolution plots from Python-based GUI. 5. Display Checks All display checks are based on a display of the SMPTE RP133 test pattern (Fig. 9) [11] per the requirements of ASTM E2698: The image display shall meet the following requirements as a minimum. Alternate image displays or requirements may be used with Cognizant Engineering Oraganization approval. The minimum brightness as measured off the image display screen at maximum Digital Driving Level (DDL) shall be 250 cd/m 2. The minimum contrast as determined by the ratio of the screen brightness at the maximum DDL compared to the screen brightness at the minimum DDL shall be 250:1. The image display shall be capable of displaying linear patterns of alternating pixels at full contrast in both the horizontal and vertical directions without aliasing. The display shall be free of discernable geometric distortion. The display shall be free of screen flicker, characterized by high frequency fluctuation of high contrast image details.
The image display shall be capable of displaying a 5 % DDL block against a 0 % DDL background and simultaneously displaying a 95 % DDL block against a 100 %background in a manner clearly perceptible to the user. The monitor shall be capable of discriminating the horizontal and vertical low contrast (1 %) modulation patterns at the display center and each of the four corner locations. A calibrated brightness meter measures the 100% and 0% DDL blocks on the test pattern to get the maximum brightness and contrast ratio. 6. Future Work Figure 12: SMPTE RP133 Test Pattern [12] We intend to continue the automation process in order to increase the speed to the test process including generation of an automated report. Grouping the tests under one GUI will also help the user complete the tests. Computed radiography (CR) tests can also be added to the suite (see ASTM E2445). Many of the same or similar tests are shared between DDA and CR.
Acknowledgements This article is a work performed under the auspices of the United States Department of Energy. Los Alamos National Laboratory is operated by Los Alamos National Security, LLC for the National Nuclear Security Administration of the U.S. Department of Energy under contract DE-AC52-06NA25396. Approved for public release (LA-UR-15-23923); distribution is unlimited. References 1. National Nuclear Security Agency, T097 Qualification of Digital Radiographic Imaging Techniques 2. Los Alamos National Laboratory, SS6K0461 General Specifications for Digital Radiography of Los Alamos National Laboratory Designed Parts and Assemblies 3. ASTM International, E1695 Standard Test Method for Measurement of Computed Tomography (CT) System Performance 4. ASTM International, E2737 Standard Practice for Digital Detector Array Performance Evaluation and Long-Term Stability 5. ASTM International, E2597 Standard Practice for Manufacturing Characterization of Digital Detector Arrays 6. ASTM International, E2698 Standard Practice for Radiological Examination Using Digital Detector Arrays 7. ASTM International, E2002 Standard Practice for Determining Total Image Unsharpness in Radiology 8. Vision in X Industrial Imaging, isee! Professional User Manual 9. ASTM International, E2445 Standard Practice for Performance Evaluation and Long-Term Stability of Computed Radiography Systems 10. ASTM International, E1570 Standard Practice for Computed Tomographic (CT) Examination 11. The Society of Motion Picture and Television Engineers, RP133 Specifications for Medical Diagnostic Imaging Test Pattern for Television Monitors and Hard-Copy Recording Cameras 12. Rich Franzen's PNG Gallery, http://r0k.us/graphics/pnglibrary.html