A new AI-powered x-ray technique for detecting explosives could identify cancer

Researchers at University College London (UCL) in the UK have created a novel x-ray technique for detecting explosives that might also be used to identify tumors.

The potentially early-stage fatal tumors in humans could be noticed by the new x-ray method that collaborates with a deep-learning Artificial Intelligence (AI) algorithm to detect explosives in luggages, according to a report published by MIT Technology Review on Friday.

"The AI is exceptionally good at picking up these materials even when they're hidden inside other objects," said Sandro Olivo, study lead author from the UCL's Department of Medical Physics and Biomedical Engineering.

"Even if we hide a small quantity of explosive somewhere because there will be a little bit of texture in the middle of many other things, the algorithm will find it."

The technique could be used in medical applications, particularly cancer screening, according to the research team.

Although the researchers have not yet tested whether the technique can successfully distinguish the texture of a tumor from surrounding healthy breast tissue, they are excited about the possibility of detecting very small tumors that would have previously gone undetected behind a patient's rib cage.

"I'd love to do it one day," said Olivo. "If we get a similar hit rate in detecting texture in tumors, the potential for early diagnosis is huge."

Explosives inside electronics

It can be challenging to find explosives using traditional x-ray techniques when they are hidden inside electronics and other things. However, researchers found that under test settings, the new approach had a 100 percent accuracy rate for detecting explosives.

Small amounts of explosives, such as Semtex and C4, were concealed by the UCL team inside electrical appliances like computers, hair dryers, and cell phones to closely resemble a traveler's bag. The products were put inside bags together with toothbrushes, chargers, and other everyday items.

The researchers used a specially constructed machine with masks—sheets of metal with holes drilled into them, which split the beams into an array of smaller beamlets—to scan the bags instead of using ordinary x-ray machines, which hit objects with a uniform field of x-rays.

The beamlets were scattered at angles as small as a microradian (roughly one 20,000th the size of a degree) as they moved through the bag and its contents. The scattering was examined by AI, which had been trained to identify the texture of specific materials from a particular pattern of angle changes.

Although the scientists admitted that it would be unrealistic to expect such a high degree of accuracy in larger studies that more closely reflect real-world situations, the algorithm was able to properly identify explosives in every experiment conducted under test conditions.

"This latest work from the UCL teams presented here looks extremely promising. It combines novel X-ray imaging with AI and has major potential for the extremely challenging tasks of threat detection in hand baggage and NDT applications such as crack detection," said Kevin Wells, associate professor at the University of Surrey, UK.

Speaking about the use of this technique for detecting early-stage tumors, Kevin added, "Cancer detection involves its own set of challenges and we look forward to seeing the work progress in this area in due course."

The research was first published in the scientific journal Nature Communications on Friday.

Study abstract:

X-ray imaging has been boosted by the introduction of phase-based methods. Detail visibility is enhanced in phase contrast images, and dark-field images are sensitive to inhomogeneities on a length scale below the system’s spatial resolution. Here we show that dark-field creates a texture which is characteristic of the imaged material, and that its combination with conventional attenuation leads to an improved discrimination of threat materials. We show that remaining ambiguities can be resolved by exploiting the different energy dependence of the dark-field and attenuation signals. Furthermore, we demonstrate that the dark-field texture is well-suited for identification through machine learning approaches through two proof-of-concept studies. In both cases, application of the same approaches to datasets from which the dark-field images were removed led to a clear degradation in performance. While the small scale of these studies means further research is required, results indicate potential for a combined use of dark-field and deep neural networks in security applications and beyond.