Dogs are incredible at sniffing out cancer. They’re also capable of detecting malaria, Parkinson’s, and maybe even COVID-19 from a variety of different sources such as blood, saliva, and urine. Now, scientists are trying to replicate what all types of dogs do naturally to develop a better cancer detector.
In a new small study, scientists tested the feasibility of using sensors and artificial intelligence to replicate the way a dog detects prostate cancer. The ambitious goal of this study is to find the right components needed to build a tiny smell-based cancer diagnostic that can be embedded into a smartphone.
In this latest paper, a cross-disciplinary group of scientists from Massachusetts Institute of Technology, the Prostate Cancer Foundation, Johns Hopkins University School of Medicine, Harvard Medical School, Cambridge Polymer Group, and the research nonprofit Medical Detection Dogs in the U.K first attempted to disprove that dogs can sniff out prostate cancer in urine. While there are several studies that show how good dogs are at detecting illness, critics say they’re not rigorous enough. However, in a double-blind randomized clinical trial, the scientists showed that dogs were able to correctly identify when a patient had (and did not have) cancer from their urine sample roughly 70% of the time. This was an important first step, because they needed to prove that the dog’s senses were worth replicating.
At the same time, the study attempted to understand what exactly dogs smell that indicates the presence of prostate cancer. The group used a technique called gas chromatography-mass spectroscopy to identify the individual molecules in a urine sample’s vapors and used microbial profiling to understand the genetic composition of the urine. Finally, they trained a neural network with both the dogs’ findings and the chromatography-mass spectroscopy readings to see if the algorithm could make connections between the two. In the paper, they discuss a few molecules of interest that may ultimately tell scientists what exactly dogs are detecting in urine that points to cancer.
“This paper was about integrating all the techniques that we know can work independently and finding out what of all this can go and become [part of] an integrated smartphone-based diagnostic,” says Andreas Mershin, an MIT researcher and one of the authors of the paper.
Scientists have long sought to make robots that smell. Inspired by bomb-sniffing dogs, the Defense Advanced Research Projects Agency has spent at least two decades trying to get scientists and private companies to develop a device that can detect bombs through smell. Its program has produced several innovations, but none that are quite as reliable as using dogs.
Mershin participated in DARPA’s 2007 Real Nose project. During a 15-month sprint, he and his team were able to develop a device that could inhale and detect individual scents in a controlled setting. They did this through training a robot just like dogs are trained to detect a given scent—by showing it different smells. It worked even better than dogs’ noses at detecting super-low concentrations of bomb-related smells. While the robot nose was less effective in uncontrolled environments, it was a big breakthrough. But Mershin and his colleagues still didn’t understand how the dogs were detecting the bombs’ scents. The reason it’s so difficult, says Mershin, is that there are no obvious biomarkers that are linked to scents.
Dr. Jonathan Simons, the president of the Prostate Cancer Foundation, who also worked on the prostate cancer study, says dogs are like “cancer sommeliers.” They’re not picking out single scents or biomarkers. “It’s more like a chord of notes, not a single note,” he says. That is why the researchers had to take a cross-disciplinary approach to untangle how dogs understand cancer—they needed to find all the elements of that chord.
It took two years and a team of artificial intelligence experts, animal behaviorists, urologists, microbiome experts, and gas chromatography-mass spectroscopy specialists. Using this multidisciplinary approach and gathering lots of different data about the urine samples, the researchers were able to produce artificial intelligence that seems to be able to generalize what cancer smells like. The initial findings are exciting because they put scientists closer to creating a very powerful diagnostic tool.
An artificial dog nose stands to be much cheaper than the most common prostate cancer diagnostic, a prostate-specific-antigen blood test. This test, which Simons says detects prostate disease broadly, not prostate cancer specifically, is very expensive. A study from 2018 published in JAMA estimates that Medicare spent $1.2 billion on prostate screening for men over 70 in a three-year period.
The next study will need to be much larger, says Simons. Where there were only 50 samples in this study, there will have to be 800 in the next one to prove efficacy. He and the other researchers have plans to create an open-source database of this work so that others can access it and built on it.
Meanwhile, Mershin is hoping that smartphone makers will start testing out this technology in their phones. “It already has an eyeball—the camera. It already has an ear, which is a microphone. It’s almost becoming like the human face! This extra sense on your smartphone is essential,” he says.