Your terrible breath is trying to tell you something—and not just that it’s time to crack open a bottle of Listerine. Within that cloud of onion and stale tuna fish odors are hundreds of chemical compounds, which combine in your mouth to create a ratio as unique as a fingerprint. By analyzing that ratio, researchers have come up with a powerful new way to detect the signatures of various diseases, from prostate cancer to Parkinson's.
Today in the journal American Chemical Society Nano, researchers unveil a sensor array that identifies and captures the unique “breathprint” of 17 different diseases. The researchers hope that their array, which uses artificial intelligence to match up the varying levels and ratios of 13 key chemical compounds found in human breath to different diseases, will pave the way for a versatile medical diagnostic tool. After sampling the breath of more than 1,400 people, they found that their technique was able to discriminate among diseases with 86 percent accuracy.
The science behind the scent of a person's breath lies within the suite of organic chemical compounds that we routinely expel into the air with every laugh, yell or sigh. These compounds often come marked with the signs of biochemical changes wrought by specific diseases—a phenomenon that forms the basis of modern breath diagnostics. The problem is, there’s a lot of background noise to sift through: In a cloud of exhaled breath, you’ll typically see hundreds of these compounds.
Ancient physicians dating back to 400 BC knew there was something to be gleaned from sniffing a sick person’s breath. The famed Greek physician Hippocrates, among others, used to smell his patients' breath to find out what ailed them. (Even worse, some physicians used to smell their patients’ urine or stool.) We’ve gotten slightly more sophisticated since then; breath analysis has been successfully employed to diagnose cirrhosis of the liver, diabetes and colorectal cancer. There's even a dedicated Journal of Breath Research.
But previously, such efforts have mainly been used to detect a single disease. In the new study, Hossam Haick, a nanotech expert at Technion—Israel Institute of Technology, and several dozen international collaborators aimed to lay the groundwork for a general diagnostic tool to identify the breath signatures of many diseases, including kidney failure, lung cancer, Crohn's disease, MS, prostate and ovarian cancer, and more. Their array first assesses each compound's relative abundance within a person's breath, and then compares disease signatures against healthy individuals.
“We have a mixture of compounds which characterize a given disease, and this picture is different from one disease to another,” explains Haick. Using mass spectrometry analysis, the group first identified the specific compound signatures for 17 different diseases. They then sampled the breath of more than 1,400 people, using a sensory array of carbon nanotubes and gold particles to register which mix of compounds they exhaled. A suite of computer algorithms deciphered what the data told them about the presence or absence of each disease.
That’s when the artificial intelligence comes in. “We can teach the system that a breathprint could be associated with a particular disease,” says Haick, who co-led the study. “It works in the same way we'd use dogs in order to detect specific compounds. We bring something to the nose of a dog, and the dog will transfer that chemical mixture to an electrical signature and provide it to the brain, and then memorize it in specific regions of the brain … This is exactly what we do. We let it smell a given disease but instead of a nose we use chemical sensors, and instead of the brain we use the algorithms. Then in the future, it can recognize the disease as a dog might recognize a scent.”
Jonathan Beauchamp, an environmental physicist at the Fraunhofer-Institute for Process Engineering and Packaging in Germany, said the technology presents a promising way to surpass a major hurdle in breath analysis. “The same VOCs (volatile organic compounds) often light-up as markers for many different diseases,” he says. “Indeed, it is now widely accepted within the breath research community that unique VOCs for specific diseases are unlikely to exist.”
Therefore, searching for concentrations of various VOCs in relation to one another, as Haick and colleagues did, may prove the more accurate diagnostic method, he adds. “These results demonstrate high accuracy in discriminating one specific disease against another ... The current study clearly demonstrates the power and promise of the gold nanoparticle array technique," he says.
The study involved dozens of scholars based at 14 research institutions across five different countries. Its participants were equally diverse: The mean age was 55; about half were male and half were female; and about one-third were active smokers. Participants were recruited around the world in the United States, Israel, France, Latvia and China. “The large number of subjects over varied geographical areas is really a key strength of this study,” says Cristina Davis, a biomedical engineer who heads the bioinstrumentation lab at the University of California at Davis.
“Larger clinical trials like this will help push the boundaries of breath analysis forward, and should help lead to promising medical tools for clinical practice,” adds Davis, who wasn't involved in the study. “They have taken new mass spectrometry knowledge and coupled it to their novel sensor output.”
Haick hopes that his team's widespread testing will lead to widespread use of the nanosystem. He says that because it's affordable, non-invasive and portable, it could be used to screen widely for disease. By screening even those with no symptoms, such a tool could enable the types of early interventions that lead to better outcomes.
But this AI-fueled “nose” might also have applications far beyond medical diagnostics. Several companies have already licensed it for other applications, says Haick. Among the many potential uses, he noes that the array could be used for quality control by detecting food spoilage. It could also be used for security at airports, by detecting the chemical signatures of explosive devices.
“The system is highly sensitive, and you just need to train it to different types of applications,” he says.