Application to the Integrated Checkpoint
ADSA01

The final report for this workshop is available at: https://myfiles.neu.edu/groups/ALERT/strategic_studies/ADSA01_final_report.pdf.

This workshop was conducted to discuss the generation of advanced algorithms for security applications. The focus of the workshop was to spark the development of new algorithms for detecting explosives at an integrated checkpoint. The objectives of the workshop were to:

  • Provide an analysis of the opportunities and research barriers associated with next-generation algorithms for Homeland Security applications, using the integrated checkpoint as a basis of discussion.
  • Consider the following questions:
    • What will be the consequences of maintaining the current trajectory using existing technologies and strategy
    • How can we foster out of the box solutions using new technologies and strategies?
  • Facilitate 3rd party involvement, especially academia and the medical imaging community, in an algorithm development strategy that would be effective for DHS.
  • Identify 3rd parties who can respond to RFIs and BAAs related to algorithm development.

Workshop Outcomes

  • Grand challenges (now known as initiatives) should be established for different aspects of threat detection and different modalities. The aspects include reconstruction and processing of sensor data, image segmentation, automated threat detection and improved operator performance. The modalities include X-ray CT for checked and carry-on baggage, whole body imaging, cargo inspection and stand-off detection. Implementing grand challenges will entail putting the following information and materials into the public domain: data sets, sensor descriptions and acceptance criteria. People working on grand challenges should be provided financial incentives to advance the state of the art.
  • Develop accurate scanner simulators to predict the performance of future systems. The simulators can also be used to provide data for the participants in grand challenges. The scanner simulators may also be considered to be part of sensor or system modeling. We do not mean verbal/written descriptions of sensors.  We mean analytical and computation models for mapping parameters of interest (e.g., spatial distributions of Compton scattering and photoelectric parameters) to observed data either for fielded systems or for model systems that approximate those in the field well enough to allow for meaningful evaluation of the processing results.  All sensor related effects seen in the field which have an impact on the data (scatter, beam hardening etc), should be included in the model.  Preferably, both analytical expressions for the models as well as computational realizations in Matlab or C should be provided or developed.
  • Studies should be performed on systems that include a human in the decision making process. Methods should be established to assess and improve the performance of the human while taking into consideration that the prevalence of threats is low at the present time. In that context, the potential reduction of the probability of detection in automated threat detection algorithms in exchange for lower false alarms should be studied to assess the impact of lower probability of false alarms (PFA) on human performance. Algorithms that estimate the amount of clutter in images could be used to send cluttered bags directly to secondary inspection, bypassing the operator, thus allowing the operator more time to view uncluttered bags. Algorithms could be developed to produce threat image projections (and equivalents for other modalities) independent of vendor.
  • Advanced reconstruction algorithm approaches should be evaluated for their use on existing X-ray CT equipment to see how they might lead to the design of better CT scanners. An example of this methodology is denoted interior tomography, where the line-integrals are only collected within a region-of-interest that is smaller than the complete object being scanned. Interior tomography may allow for higher quality scans of threats during secondary screening.
  • Communication – both formal and informal – between the government, vendors, academia and national laboratories was quite useful and additional communication should be fostered.
  • All participants agreed that the involvement of 3rd parties would benefit both the vendors and the government. However, the various stakeholders had different expectations in the time required to make contributions. The government and vendors wanted advanced algorithms developed in the six- to twenty-four-month time-frame. Academia felt that game-changing algorithms would take five to ten years to develop.
  • The use of orthogonal technologies (also denoted fused systems or systems of systems) should be explored.
  • An open source model should be employed for the distribution of algorithms, code, specification and databases. Standardized image and data formats, such as DICOS, would allow 3rd parties to develop algorithms more quickly.
  • Before developing new algorithms based on a specific sensing modality, one should predict the best possible performance (PD/PFA) for that modality. If a modality is currently operating close to its best possible performance, then do not fund additional advanced algorithm research in that arena.
  • CT-based explosives detection systems (EDS) were derived from the medical imaging application and as a result have perhaps not been sufficiently optimized for the security application. Numerous opportunities were identified to springboard from medical imaging approaches and develop algorithms targeted specifically for the security application. Examples include:
    • Reconstruction optimized for security scenario
    • Targeted reconstruction to specific threat
    • Segmentation-oriented reconstruction method
    • Local reconstructions optimized for a threat found during segmentatio
    • Targeted reconstruction for detection versus display
    • Reconstruction algorithms for multi-view line scanner
    • Parametric reconstruction
    • Iterative/statistical reconstructio
    • Artifact reduction such as from scatter and meta
    • Improved dual-energy decomposition
  • Use video surveillance to identify passengers that should be subjected to increased scrutiny at the check-point. Also use this method to associate divested items with the passenger.
  • Vendors should be incentivized to deploy scanners with improved performance.

Recommendations are included in the final report for the workshop on how to continue to get 3rd parties involved with advanced algorithm development. In particular, it has been recommended that an initial grand challenge be conducted for image segmentation for CT-based EDS equipment and a second workshop be held on implementing this specific grand challenge. Additional grand challenges can be held for other modalities and applications such as whole body imaging (WBI) and cargo screening.