5.2.1
Introduction and rationale
Image reconstruction using algebraic reconstruction technique provides a method of analysis to solve a large matrix equation of image data for the most likely solution. It allows the specific response of a collimated detector to be taken into account and can rapidly find a solution of source distribution(s), given many thousands of data points. This allows the solution of data from many slot angles (in contrast to two slot angles in experiment 1) and many pan angles, to solve for up to full 4π images. Using this method, high resolution
images can be produced with a lower data acquisition time than “pixel by pixel” imagers. Spectroscopy is commonly used with gamma-ray imaging to identify the distribution of specific radioisotopes. When collecting spectroscopic data, specific photopeaks can be selected and analysed allowing the corresponding radioisotope distributions to be imaged. For example, isolating the 662 keV peak would allow imaging of the137Cs distribution with minimal interference from other radioisotopes (in most cases). The general distribution of gamma-ray emitters can therefore be further interrogated.
This capability may also be advantageous when imaging neutron fields; for example in nuclear decommissioning applications, the ability to discern neutron-emitting materials (e.g. a shielded californium source from plutonium residues) might greatly impact the cost and strategy of a decommissioning operation. Knowledge of the distribution of radionuclides and/or their emitted neutron energies may lead to improved accuracy beyond imaging, for example dose maps of a given region which are dependent on the isotopic distribution.
This experiment sought to test the feasibility of source recognition using neutron pulse- height spectroscopy. First the sources were imaged to identify their locations, as would be the case without prior knowledge. The collimator slot was then oriented to each hotspot (corresponding to each source) in turn, and neutron pulse-height spectra were collected. These spectra were then compared with a library of spectra, previously recorded, used to identify the source type.
The C0 collimator was used for these experiments due to the minimal impact of the tungsten on the measured neutron pulse-height spectrum, due to the low energy loss per collision in high-Anuclides. In contrast, a hydrogenous collimator such as C1 would moderate the neutron energy spectrum, in turn shifting the pulse-height spectrum and making the spectra obtained from the imager more difficult to compare with those previously recorded.
5.2.2
Experimental set-up and apparatus
The apparatus used in this experiment are summarised in Table 5.3. This experiment was performed at the Schuster Laboratory, University of Manchester, UK.
The imaging system was deployed on a trolley raised approximately 50 cm from the ground. The252Cf source was placed in the horizontal plane containing the detector at 10 cm from the front of the face of the collimated detector. Data were recorded for one hour to allow set-up of the neutron-gamma discrimination parameters in post-processing.
The radiation sources were placed in the horizontal plane containing the detector, 50 cm from the detector front face. The sources were located in the geometric centres of the cannisters which were separated by 20° in azimuth. A photograph of this set-up is provided in Fig. 5.4 showing the probe,241Am/Be source (left) and252Cf source (right). The imaging routine was initiated and the data set was collected, the total data collection time for this image being 15 hours. Following these measurements, the probe was oriented to ˆα = 0°,
i.e. with the slot in the vertical position. The probe was then rotated through angle ˆβ such
that the detector was directly facing the first radiation source, aligning the sensitivity region (the minimally shielded slot void). Spectroscopic data were collected for 30 minutes. This process was then performed for the second radiation source.
The image data were analysed using the ART algorithm outlined in section 3.3.3. The collected neutron spectra were analysed using the method outlined in section 3.3.6 using pre-recorded spectroscopic data shown in Fig. 4.32.
Table 5.3 Summary of materials and methods used in experiment 2. MFA Single channel MFAX1 [section 3.2.2]
Data collection Ethernet Discrimination Post-processed
Detector(s) Miniature EJ-301 [section 3.2.1] Collimator Tungsten C0 [section 3.2.3] Imaging method ART reconstruction [section 3.3.3] Imaging parameters 35 slot×49 pan,td = 30s
Radiation sources 241Am/Be (targetTa) with 1 cm lead shield (2.2×105neutrons s−1),
252Cf (targetT
b) (1.5×105neutrons s−1)
5.2.3
Results
The discrimination set-up used in this experiment is illustrated in Fig. 5.5. GARR was measured with a 137Cs source at 1.21%. The unprocessed image data recorded in this
Figure 5.4 Photograph of the probe and radiation sources in experiment 2 during data acquisition. The two sources are (left to right) 241Am/Be and 252Cf placed on hollow cardboard and stools level with imager with a separation of 20° in azimuth.
experiment are shown in Fig. 5.6. The reconstructed neutron images are shown in Fig. 5.7. Fig. 5.7a shows the full neutron image solution; the same data have been replotted with a raised lower threshold on the colour scale shown in Fig. 5.7b. The associated gamma-ray image is also shown in Fig. 5.7c.
The results from the spectral analysis are shown in Fig. 5.8. At this stage it was assumed that two clusters of neutron-emitting materials had been identified from the image, their constituents were known to be of two types and the collected spectroscopy in each case was compared with that of252Cf and241Am/Be sources using the method outlined in section 3.3.6. Each experimentally recorded “target” spectrum from targetsTa(241Am/Be) andTb(252Cf)
was compared against a hypothetical spectrum comprising contributions from241Am/Be and
252Cf. The contributions of each hypothetical spectrum were investigated over the weighting
range 0% to 100% to find the lowestχ2value, and therefore closest match to the recorded
“target” spectrum, e.g. the most likely ratios of241Am/Be to 252Cf making up a “target” spectrum. Weightingw corresponds to the weighting of241Am/Be in the spectrum. The lowestχ2value for each “target” of spectroscopy indicates the scenario of closest match, and
the relative weightings of the241Am/Be (w) and252Cf (1-w) neutron pulse-height spectrum in each targetTa(241Am/Be) andTb(252Cf).
Figure 5.5 Plot of events as a function of the discrimination parameters in experiment 2 showing discriminated gamma rays in red and neutrons in blue.
(a) Neutron image data (b) Gamma-ray image data
Figure 5.6 Unprocessed image data obtained in experiment 2: discriminated events as a function of slot and pan position.
5.2.4
Discussion
The discrimination parameters shown in Fig. 5.5 demonstrate a clear separation of neutrons and gamma rays in the plot with some overlapping at lower energies. A modified polyline was applied in post-processing for neutron-gamma discrimination, allowing a lower GARR than in Experiment 1. This indicated better isolation of the neutron field, though a small
(a) Neutron image solution (b) Neutron image solution with applied threshold
(c) Gamma-ray image solution
Figure 5.7 Radiation images produced in experiment 2 as a function of elevation and azimuth angle. Two sources are (left to right)241Am/Be and252Cf placed level with the imager at a separation of 20° in azimuth.
percentage of neutrons, estimated by GARR, will have been misclassified as gamma rays. The neutron image solution in Fig. 5.7a shows two hotspot regions, one centrally and one to the right-hand side of the plot. The central hotspot corresponds to the location of the two neutron sources which have not been individually resolved in this image solution. The hotspot to the right-hand side is an image artefact and does not correspond to the location of a neutron source; other smaller image artefacts also appear in other regions. These artefacts are a result of a mismatch between the sensitivity map and the physical collimated detector and manifest in regions where there is the least data available to define the solution, in these cases on the outsides of the image space. Applying a low-flux threshold to these images (Fig. 5.7b) removes these image artefacts and allows the two sources to be resolved. The
Figure 5.8 Plot of reduced chi squared values as a function of weighting,w, when comparing target spectra from241Am/Be (wa) and252Cf (wb) with hypothetical spectra comprising the fractionwof241Am/Be and (1-w)252Cf measured in isolation.
gamma-ray image in Fig. 5.7c shows a single hotspot corresponding to the location of the
252Cf source. The241Am/Be source was not visible in the gamma-ray image which was
thought to be due to the 1 cm lead shield reducing a significant proportion of the emitted gamma rays. These result therefore demonstrates the benefits of neutron imaging in addition to gamma-ray imaging allowing neutron sources to be identified in the presence of high-Z shielding.
The pulse-height spectrum fitting results in Fig. 5.8 show two minima at distinct weight- ings ofw for each target. This demonstrates that the pulse-height spectra observed were significantly different in terms of containing unequal weightings of the241Am/Be and252Cf spectra. It can also be seen that when 241Am/Be was the target, the spectrum could be identified to have a larger component of the241Am/Be spectrum when compared with the case when 252Cf was the target and vice versa (wa > wb). These results indicate that a method of spectral analysis, such as this simplistic approach, performed alongside this method of radiation imaging, can be used to discern two sources of neutron radiation which emit different energy spectra.
The ability to discern each source in the image with this method demonstrates the successful application of the slot-modulated approach with ART reconstruction to combined fast-neutron and gamma-ray imaging. These images contained significant image artefacts which was a focus of further research. The limitations of this imaging approach were thought to be associated with the collimator C0 which only provided a limited spatial biasing
compared with the further developments (see section 4.4), although the success of source identification was likely due in part to the preservation of the neutron energy spectra by limited energy loss in elastic scatters with tungsten. Spectral source identification with hydrogenous collimators would have to be investigated separately due to the moderating effects of such collimators. These results indicate that an integrated imaging system with combined source identification could be achievable. The ability to identify the radionuclide constituents in neutron emitters may also lead to higher accuracy in image reconstruction. This is because the collimator sensitivity map is a function of neutron energy, knowledge of the radionuclide type and emitted energy allows the sensitivity map to be calculated more accurately leading to more accurate image solutions. A possible research direction from this point was to expand upon this method by applying a more rigorous spectral analysis tool directly to image data. This was however discontinued because of the need to move to higher data collection speeds where spectral data was no longer supported by the MFAs.