New X-Ray Approach Offers Near-Perfect Threat Detection

In a recent study published in the journal Optica, researchers from the University College London produced multi-contrast images that can be utilized in thousands of complex scenarios to identify potentially dangerous objects (like explosives). This was achieved by combining different X-Ray imaging technologies.

The new method could be helpful for security screening and applications in the physical and biological sciences. It also uses easily accessible machine learning processes for materials classification.

This method is particularly well suited to discriminating objects with very similar elemental composition. It could be used in airport security or any inline scanning operation to examine materials flagged as suspicious by an initial fast scan, such as a traditional X-Ray system.

Thomas Partridge, Study Team Leader, University College London

The new method successfully identified and detected explosives in nearly 4,000 scans of threatening and non-threatening materials concealed inside bags or obscured by various objects. With only one false negative, they attained an almost flawless 99.68 % recall rate from the instances that posed a threat.

Although more work is needed, this approach could also prove useful for medical imaging. While traditional X-Ray imaging struggles to separate healthy from diseased tissue, other studies have suggested that phase contrast imaging might be able to capture textures that could be used to distinguish healthy and benign tissues.

Thomas Partridge, Study Team Leader, University College London

Unlocking Material Secrets

X-Ray attenuation, which depicts the decrease in X-Ray intensity as the radiation passes through a material, is the basis for the X-Ray machines found in airports and medical facilities. The new method produces multi-contrast images by merging traditional X-Ray attenuation data at different X-Ray energies with X-Ray phase information, which comprises refraction and dark-field channels.

Many explosives and common everyday items are composed of primarily carbon, hydrogen, nitrogen, and oxygen, a similarity that makes them difficult to separate with X-Ray attenuation alone. The additional channels offer significantly better enhancement of edges as well as textures and grains of materials, allowing the discrimination of objects with very similar elemental compositions.

Thomas Partridge, Study Team Leader, University College London

This work expands upon the researchers' prior attempts to combine machine learning techniques with multi-contrast X-Ray phase enhanced imaging for threat identification with fewer explosives and innocuous items.

To more closely resemble real-world circumstances, they significantly increased the number of materials studied and imaging scenarios in the current experiment. They also developed a more efficient scanning system whose resolution could be adjusted by modifying the scanning speed and applying phase contrast for edge illumination.

Masks are positioned before and after the sample in edge illumination to provide the sub-pixel X-R-ray “beamlets” required to raise the system's sensitivity to phase signals. This illumination approach's ability to function with incoherent X-Ray sources expands its usefulness, which is one of its main advantages.

The scientists used machine learning with a hierarchical design that segregated the cluttering objects before differentiating between material kinds. This is because the increasingly complex imaging settings demanded more complex protocols. This allowed for the quick identification of materials based on key distinguishing characteristics by detecting minute variations in forms and textures.

Detecting Threats

To test the novel method, the scientists employed 56 non-threat materials and 19 threat materials, all three thicknesses and hidden by various items that passengers typically carry on in their carry-on bags, like socks, brushes, and face wipes.

In several instances, the researchers were able to show not just material discrimination but also identification by utilizing all of the obtained contrast channels. With only one mistake out of 313 danger situations, the signals from the combination of X-R-ray contrasts could be analyzed using deep learning, yielding highly encouraging results.

According to the researchers, implementing this strategy commercially would necessitate increasing scanning speed through additional system optimization. Testing the material discrimination's robustness using a more extensive dataset is also necessary.

The team's current research topic involves combining the technique with 3D computed tomography scanning. This technology is being investigated for security applications because it can produce precise, three-dimensional images of items.

Journal Reference:

‌ Partridge, T., et al. (2024) Multi-contrast X-Ray identification of inhomogeneous materials and their discrimination through deep learning approaches. Optica. doi.org/10.1364/optica.507049.