The Department of Homeland Security's Science and Technology Directorate (S&T) has released a new dataset of 250 synthetic computed tomography (CT) X-ray images of carry-on baggage — a first-of-its-kind resource designed to solve a persistent bottleneck in how airport screening algorithms get built and tested.
The Vetting Bottleneck
Every checkpoint CT scanner runs a detection algorithm trained to recognize weapons, explosives, and other prohibited items inside a 3D X-ray image in real time. Building and improving those algorithms requires large volumes of realistic bag imagery to train on — but real TSA scan data is sensitive, and access has historically been tightly restricted. Developers seeking to work with authentic bag images have had to undergo a vetting process to view a limited library, a barrier S&T says has discouraged some innovators from entering the field entirely.
The new dataset sidesteps that problem by using synthetic imagery — computer-generated scans engineered to mirror the resolution and characteristics of TSA's latest CT systems without exposing real passenger data. Cignal LLC, a small technology firm based in Reedsville, Pennsylvania, developed the images under S&T's Screening at Speed Program, with funding routed through the Silicon Valley Innovation Program (SVIP) — the S&T office that finances early-stage startups working on homeland security challenges.
"Detection algorithms are key to reducing false alarms that cause hands-on bag inspections and longer checkpoint wait times. Enabling rapid algorithm development from the best innovators not only will increase detection accuracy at the checkpoint but also improve the traveler experience." — Pedro Allende, DHS Under Secretary for Science and Technology
Why False Alarms Matter
Allende's comment points to the real operational cost this initiative is aimed at: every false alarm at a CT lane means a bag gets pulled for manual inspection, adding minutes to a single passenger's screening and compounding delays across the checkpoint during peak travel periods. Detection algorithms trained on thin or narrow datasets tend to either miss genuine threats or over-flag benign items with unusual shapes or densities — laptops, dense electronics, liquids, tightly packed clothing. More varied, realistic training data is one of the more direct levers regulators have to bring both error types down simultaneously, rather than trading one for the other.
This mirrors a pattern BorderTrend has tracked closely on the cargo side: NII X-ray screening at ports faces the same fundamental constraint — detection systems are only as good as the imagery used to train the humans and algorithms reading them, and access to realistic training data has long lagged behind the sophistication of the scanners themselves.
What Comes Next
S&T has indicated the initial 250-image release is a starting point rather than the full effort. Future releases are expected to include larger and more structurally complex bag sets, along with synthetic millimeter-wave imagery — the technology used in passenger body scanners — extending the same synthetic-data approach beyond baggage to checkpoint screening more broadly.
The agency frames the initiative as part of a broader push to speed up security technology development while supporting trade and travel efficiency goals. Whether synthetic data can fully substitute for the messiness of real-world bag contents — damaged items, unusual packing, worn luggage — remains an open question that will only be answered once algorithms trained on it are tested against live checkpoint conditions. BorderTrend will continue monitoring S&T's Screening at Speed Program and follow-on releases as they're announced.