We’ve heard of dogs sniffing out cancer—an unsurprising skill given that they have over 200 million scent receptors (40 times more than humans), each of which is 10,000 times more powerful than the ones in our noses. Dogs’ noses are so powerful that they can detect a single drop of liquid mixed in enough water to fill 20 Olympic-size swimming pools and trace amounts of cancer biomarkers in biological samples. A growing number of studies has demonstrated that trained dogs can detect cancers with remarkable accuracy: one study showing that dogs can pick out the urine samples from prostate cancer patients’ with 99 percent accuracy.
“Dogs, for now 15 years or so, have been shown to be the earliest, most accurate disease detectors for anything that we’ve ever tried,” said MIT researcher Andreas Mershin.
“So far, many different types of cancer have been detected earlier by dogs than any other technology.”
While many of us would welcome the idea of a canine presence at the doctor’s office, diagnostic dogs are unlikely to become standard features at healthcare institutions anytime soon. For one, researchers are still unsure of the specific biochemical compounds that dogs are detecting when they alert to the presence of disease. This makes the design of custom training regimens for specific conditions particularly challenging and difficult to test and control.
To overcome this, could next-generation sensors and AI-driven technologies capture dogs’ powerful diagnostic capabilities? Yes, says a team of MIT researchers led by Mershin who have built a miniaturized device that can detect and process chemical signals in a sample of air with sensitivities that surpass dogs’ abilities. By pairing this odor-detection system with machine-learning algorithms, the platform can analyze and identify the presence of “disease smells.” The study was published in PLOS One.
Mershin and colleagues have been creating detector systems that use stabilized mammalian olfactory receptors as sensors, with the latest iteration being the product of many years of work. These in-built receptors can pick up volatile organic compounds and the presence of microbes in biological samples. No sophisticated computer systems are required to analyze the data collected—even a smartphone’s processing power could handle it.
The team put their device to the test with 50 urine samples from healthy volunteers and patients with prostate cancer diagnoses. This side-by-side comparison using both the new platform and trained medical detection dogs brought surprising results: it was a tie. Both the canines and the artificial system red-flagged the cancer samples with over 70 percent accuracy.
The diagnostic device’s performance has significant room for improvement. After all, its sensors are around 200 times more sensitive than those in a dog’s nose. Tweaks and improvements to the algorithm could assist the system make sense of the complex odors it captures more effectively, linking them to the presence of disease.
“We knew that the sensors are already better than what the dogs can do in terms of the limit of detection, but what we haven’t shown before is that we can train an artificial intelligence to mimic the dogs,” said Mershin.
“And now we’ve shown that we can do this. We’ve shown that what the dog does can be replicated to a certain extent.”
This study points towards the potential and feasibility of an eventual clinical implementation of such a technology. Follow-up studies will expand the testing and validating scope, with Mershin hoping to study how the set-up performs in much larger panels of samples.
