Reconstruction Algorithms for CT-Based Explosive Detection Equipment
The final report for this workshop is available at:
The purpose of ADSA04 was to discuss how third parties could participate in the development of reconstruction algorithms for explosive detection equipment based on CT scanning.
The workshop addressed advanced reconstruction algorithms for CT scanners. Most of the presentations dealt with medical imaging. However, a few presented results for for explosives detection. An advanced reconstruction algorithm for security should lead to improved performance of explosive detection equipment. Performance is defined as increase PD, decreased PFA, larger population of explosives being detected and lower minimum mass. Reconstruction algorithms were discussed for the following types of scanners:
- Few-view (<50) and multi-view (>150) scanners
- Photon integration and photon counting detectors
- Single energy, dual energy and spectral CT
- Rotate-rotate and non-mechanical rotation
- Full and truncated projections
- Full and partial rotation
- Limited angle
The types of reconstruction algorithms considered are as follows:
- Iterative reconstruction techniques
- Sinogram processing, also known as pre-processing
- Targeted artifact removal
- Hybrid algorithms including improvements to filtered back projection
- Image post-processing
The artifacts considered are as follows:
- Partial volume
- Dishing and cupping (low-frequency shading)
The following operational topics were also considered:
- Acceptance criteria for advanced reconstruction algorithms
- Availability of simulation tools and standard phantoms
- Requirements for projection data
- Funding requirements
- Legal/contractual requirements and limitations
- Timeline for developing new algorithms
- Seeding development of algorithms by third parties
- Working with SSI/OUO and classified data
- Need to publish
- Long termn versus short term R&D
Note that implementation and deployment considerations were not part of the scope of the workshop. The implementation considerations included the following topics.
- Reconstruction speed
- Hardware or software implementations
- Royalty payments
- Iterative reconstruction techniques (IRT) are well-suited for security applications because the threats (explosives) consist of many voxels. Image constraints, such as a smoothness function, are particularly applicable to homogeneous threats. Threats with texture may pose a problem for IR. IR may have to be applied to the raw data instead of the corrected data. Compressed sensing, model based, total variation and Simultaneous image reconstruction (SIRT) all looked promising. IRT will have to be optimized differently for security than it was for medical. High-resolution targeted reconstructions may be necessary if computational times are too long for the complete scan field-of-view.
- IRT may also be useful to improve the performance of dual-energy and spectral imaging.
- Sinogram processing may be useful to reduce or eliminate streak, cupping and dishing artifacts. These artifacts are caused in part due a polychromatic x-ray source, scatter and electronic noise. However, not enough time was devoted at the workshop on this topic.
- The method of Kachelrieß of using localized beam hardening correction may be used because the diverse nature of objects (threats and non-threats) present in luggage.
- Image processing may yield the same or similar benefits as IRT. The equivalent methods used in medical and nondestructive evaluation imaging should be evaluated on scans of luggage.
- Simulations could be preformed to assess the impact of different x-ray spectra.
- Hybrid reconstructions methods (1-2 iterations plus FBP) lookpromising as a compromise to IRT.
- Reconstruction using truncated projections should be considered, especially if leading to high-resolution scanning of targeted regions of interest.
- Combining research into reconstruction and ATR should and could lead to better solutions for PD/PFA.