Research Beyond the Scope of this Thesis

The work completed for this project so far has been undertaken exclusively in a laboratory environment using experimental detectors. The resulting fused images, while accurate, can only be developed individually and require a number of different steps to generate. The next stage of this project will likely have two main areas of focus: onsite testing of the current system and development of real-time fusion algorithms.

• Now that fused images have been collected in the laboratory, the same tests must be proven to work in a nuclear decommissioning environment. There will be challenges for both imaging modalities; stereoscopic imaging may struggle if the environments are bland or repetitive, while the radiometric imaging will have to be able to cope with stronger sources, increased standoff distances and increased levels of background radiation. Whilst these problems may make imaging difficult, this environment will be hugely beneficial in understanding how the system copes in practical situations as opposed to carefully controlled laboratory experiments. The data gathered from onsite tests should help inform choices on the parameters of the final fused imaging system and potentially drive research into novel techniques to overcome the challenges of the nuclear decommissioning environment. These tests will likely be of use in other Compton projects, particularly homeland security applications.

• The GUI presented in this thesis proved the concept of fused imaging but has little practical use for real time identification and quantification of radioactive sources. The algorithms developed can be used as a basis for the software however, as the image generation and fusion codes work correctly and efficiently. The most important step will be to use a well supported data acquisition environment that can process code from a number of different sources, notably C++ and Matlab in this case. To this end the LabVIEW environment [94] should be given serious consideration for use as the DAQ software for this project. It is a widely used piece of system design software with a large user base in industry and well supported integration of C++

and Matlab codes. Furthermore there are official drivers for communication between LabVIEW and the CAEN V1724 digitizer cards used in this project, making it easier to output the raw data from the electronics into LabVIEW. With this setup imaging events could be passed to the back projection imaging algorithm as soon as they are processed and using the C++ camera control libraries images could be acquired from the stereoscopic camera automatically or as the user desires. The Matlab GUI could be built into this LabVIEW environment allowing real time fused images to be produced and the system at this stage would be much more representative of a commercial system than the current set up.

• It would be beneficial to develop the simulation of the experimental data using more comprehensive tools that account for the digital processing of the collected charge. This would allow PSA to be applied to simulated data, providing a comparison for the optimised Compton camera in its current state. In addition, the rise time and multiple interaction gates could be understood in greater depth, optimising the acceptance of events that PSA can be applied to.

• Something that has not been touched upon in this thesis but may prove to be critical is the development/use of alternative radiometric imaging algorithms. The filtered back projection algorithm employed in this work is used as it is relatively quick to process events, leading to efficient image generation. It has drawbacks however, not least the difficulty in ascertaining image depth correctly. There are other algorithms that, while generally more computationally expensive, can lead to accurate image resolution in x, y and z. An approach worth considering is the stochastic algorithm [95], an iterative method. Iterative reconstruction considers the problem of image generation from a different perspective compared to analytical methods; while filtered-back projection draws a cone for each event that covers every possible source location, an iterative code will choose a point on the surface of the cone thereby making a guess at the source location. Every event is processed in this fashion and an image is reconstructed from these guesses where a large amount of them will be incorrect, giving a poor image. From this point the initial guesses influence the next set of points. For each point if the image is deemed to improve this new guess will be kept as the true event, if not the old guess will be kept. By continually guessing the path of each gamma ray the

image should converge to an optimum quality which should have a significant noise reduction compared to filtered back projection images.

There are similarities between this method and the iterative method used to correct for image disparity in Chapter 4.1.2. However radial distortion had only two unknowns (the distortion parameters), whereas the scattering angle of each event is an unknown in Compton camera iterative algorithms (in one spherical dimensionθ, it is known in the other dimension φ). This means that the processing time required to converge on an optimal image is huge; as shown in [95] 500 iterations of 420000 counts took 2442 s whereas image processing of 52469 counts in the Matlab GUI takes 1.4 s. This equates to a factor of 427 increase in processing time by switching to iterative reconstruction (though this only a rough estimate as the processing was carried out on two different computers). However the potential image improvements mean that it is worth developing an iterative algorithm and as computing power improves the increased processing time may be significantly reduced in the future.

Experimental Measurement Log

This appendix will act as a reference for detailed data about each dataset collected in experimental measurements for this project.

A.1

Compton Camera Experiment: February 2012

This experiment used the 5 mm + 20 mm HPGe Compton camera system outlined in Chapter 3 and the procedure was explained in Chapter 4:

• The DC side of scatterer and the AC side of the absorber was facing the source.

• The energy thresholds of both detectors were set to be around 7 keV.

• The time coincidence window between both detectors was set to be 75 ns.

• A square scan grid of 6 x 6 positions separated by 20 mm in x or y was measured for each dataset.

A.1.1 Radioactive Sources Used