A new X-ray technique that works alongside a deep learning algorithm to detect explosives in luggage could eventually catch potentially deadly tumors in humans.
Concealing explosives inside electronics and other objects can make it difficult to detect them using conventional X-ray techniques. A new X-ray method, in conjunction with an AI identification algorithm, was able to detect explosives with 100% accuracy under test conditions, according to new research.
While the most obvious application would be to scan for bombs and other dangerous items and substances at airports, the findings, described in Nature Communications today, could also help to detect cracks and rust in buildings, and eventually identify early-stage tumors.
A team of researchers from UCL hid small quantities of explosives, including Semtex and C4, inside electrical items, such as laptops, hairdryers and mobile phones. The items were placed inside bags with toothbrushes, chargers, paracetamol and other everyday objects to closely replicate a traveler’s bag.
While standard X-ray machines hit objects with a uniform field of X-rays, the team scanned the bags using a custom-built X-ray security scanner containing masks—sheets of metal with holes punched into them, which separate the beams into an array of smaller beamlets.
As the beamlets passed through the bag and its contents, they were scattered at angles as small as a microradian (around 20,000 times smaller than a degree).The scattering was analyzed by AI trained to recognize the texture of specific materials based on a particular pattern of angle changes.
The AI is exceptionally good at picking up the textures of these materials, even when they’re hidden inside other objects, says lead author Sandro Olivo, from UCL Medical Physics & 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 algorithm was able to correctly identify explosives in every experiment carried out under test conditions, although the team acknowledged it would be unrealistic to expect such a high level of accuracy in future larger studies that resembled real-world conditions more closely.
The technique could be used in medical applications too, particularly cancer-screening, the team believes. Although Olivo and his team are yet to test whether the technique could successfully differentiate the texture of a tumor from the surrounding healthy breast tissue, for example, he’s excited by the possibility of detecting very small tumors that could have gone previously undetected behind a patient’s ribcage.
“I’d love to do it one day,” he adds. “If we get a similar hit-rate in detecting texture in tumors, the potential for early diagnosis is huge.”
But the human body is a significantly more challenging environment to scan than static, air-filled objects like bags, points out Kevin Wells, associate professor at the University of Surrey, who was not involved in the study. Additionally, the researchers would need to downsize the bulky equipment and ensure the cost of screening was equivalent to existing techniques before it could be considered as a potential screening method for humans.
“What’s presented here looks extremely promising. I think it has great potential for certain types of threat detection, and for detecting cracks,” he says.
“For the medical, cancer-type application, it’s a possibility, but there are a few steps to go before you could demonstrate efficacy in a clinical context.”
