Cancer-sniffing Dogs May Help Train Robot Noses to Detect Tumors Early

Cancer-sniffing Dogs May Help Train Robot Noses to Detect Tumors Early

Understanding the molecular components of odors used by trained dogs to detect prostate cancer could help researchers to develop computers with similar diagnostic accuracy, a new study shows.

The study, “Feasibility of integrating canine olfaction with chemical and microbial profiling of urine to detect lethal prostate cancer,” was published in PLOS One.

An ongoing need in prostate cancer care is for more accurate non-invasive tests to detect the cancer, and to distinguish between potentially deadly tumors and benign growths.

Currently, the most common approach is to measure levels of the prostate specific antigen (PSA) protein in the blood. However, PSA testing is liable to miss some tumors, while flagging those that are unlikely to be harmful.

In recent decades, various studies have shown that dogs can be trained to detect some human diseases by smell, including prostate cancer (by smelling urine). However, using trained canines as a diagnostic test isn’t feasible on a large scale, leading to the question of whether cancer-smelling technology could be used instead.

In the new study, an international team of researchers performed a series of proof-of-concept experiments to show how this might be done. They used urine samples collected from men with biopsy-proven prostate cancer, or no cancer (controls).

Of note, all of the cancer samples used were from individuals with Gleason 9 prostate cancer (the most lethal grade of prostate tumors), while the controls were by definition taken from those who had undergone prostate biopsy, typically due to PSA or other testing that indicated potential prostate disease.

First, two dogs — named Midas and Florin — were trained to detect prostate cancer from urine samples, then were tested on their ability to do so. Both canines correctly identified five out of seven prostate cancer samples. Florin correctly identified 16 of 21 control samples as negative for prostate cancer, while Midas correctly identified 14 out of 20 negative samples.

“Specialist-trained cancer detection dogs, Florin and Midas, detected extremely aggressive prostate cancers quickly and accurately from urine samples, even discriminating these against urine from patients that had other diseases of the prostate,” study co-author Claire Guest, co-founder and chief scientific officer of Medical Detection Dogs, said in a press release.

The researchers then used a technique called gas chromatography-mass spectrometry to analyze volatile organic compounds (VOCs) in some of the urine samples. Conceptually, VOCs are “smelly” molecules that can easily move through the air to be picked up by a sensor, whether that sensor is a machine or a dog’s nose.

The researchers also used a genetic technique called 16S rDNA sequencing to analyze the bacteria found in the different urine samples.

In preliminary analyses, the researchers noted several general differences between the VOC and bacterial profiles of cancer or control urine samples.

Then, the researchers used the VOC data, combined with the dogs’ identification of the various samples, to train an artificial neural network (ANN). As the name suggests, an AAN is a type of machine learning algorithm that is designed to somewhat mimic an organic brain.

In essence, the researchers fed into a computer the VOC data for urine samples, as well as whether the sample was detected as cancer by Florin and/or Midas. Then, the AAN looked for patterns in the VOC data that would allow it to make the same calls as the dogs.

In other words, the computer program looked for the particular components of the urine samples that the dogs were smelling to make their diagnosis. Two different computational techniques (skeletonization and auto-associative filtering) yielded generally consistent results in this analysis.

“Although tested on a small sample set which does not enable us to make definitive conclusions about accuracy, the results achieved in this pilot support the potential of specialist trained detection dogs directly assisting in the development of an ANN to run on a bio-electronic machine olfaction [smell-based] diagnostic device,” the researchers concluded.

They added that their results “pave the way towards development of machine-based olfactory diagnostic tools that define and recapitulate what can be detected and accomplished now via canine olfaction.”

Study co-author Jonathan W. Simons, MD, president and CEO at the Prostate Cancer Foundation, said: “With compelling evidence of this approach, we are planning larger-scale studies using canine olfaction, urinary VOCs and urinary microbiota profiling to develop a machine olfaction diagnostic tool, a ‘robotic nose’ if you will, that may ultimately take the form of a smartphone app of the future.”