Previous experiments have found that dogs seem to have an uncanny ability to "sniff out" individuals with cancer. Could that olfactory ability be leveraged for actual diagnostic use? U.K. researchers put that theory to the test in a study published February 17 in PLOS One.
Researchers with Medical Detection Dogs of the U.K. studied the olfactory ability of dogs to detect aggressive prostate cancer from urine samples and used it to train an artificial neural network (ANN) to better detect cancer in diagnostics. They found that the cancer-sniffing ability of the canines was just one of several factors that could be used to optimize their neural network.
Diagnostic technologies that are noninvasive and more sensitive and specific are needed to detect prostate cancer. What's specifically needed is a prostate cancer diagnostic tool that differentiates between cancers that are potentially lethal -- with high Gleason grades -- and that have metastatic potential from indolent, low-grade cancers that don't pose a threat to patients.
The limitations of current diagnostic methods, such as the widely used prostate-specific antigen (PSA) screening test, and molecular volatile organic compounds (VOC) analysis by gas chromatography-mass spectrometry (GC-MS), led the scientists to consider taking a multiparametric approach to optimize diagnosis to better understand the underlying disease pathology, and to illuminate the way toward machine olfaction-based urinary screening and diagnosis.
Over the years, a number of scientific experiments have been performed demonstrating that dogs -- with their highly refined olfactory abilities -- seem to be able to detect the presence of cancer in humans. Researchers speculate that dogs are able to detect trace elements of cancer from human waste material like breath, plasma, urine, and saliva.
In the current study, the U.K. researchers trained their ANN to detect prostate cancer by combining streams of data obtained from a variety of methodologies, each of which is capable to various degrees of making the diagnosis on its own:
- Canine olfaction
- Conventional GC-MS analysis of urine headspace VOCs
- Urinary microbiota profiling of the same samples
- A novel, purpose-developed ANN
Rather than rely solely on one method -- canine detection, volatilomics, or urinary microbiota profiling -- to improve diagnostic efficacy, in this double-blinded pilot study the researchers combined the strengths of each to create new insights into how further integrative diagnostic advances might be achieved.
Two dogs were trained to detect Gleason 9 prostate cancer in urine collected from patients with biopsy-confirmed cancers. The investigators indicated they used biopsy-negative controls to assess canine specificity as prostate cancer biodetectors.
The canines were able to discriminate between Gleason 9 prostate cancer and biopsy-negative controls at a high sensitivity and specificity, the researchers wrote in their study, representing an olfactory approach that depends on the emergent property of scent character.
At the same time, urine samples were simultaneously analyzed for their VOC content in headspace using GC-MS. The scientists used 16S rDNA Illumina sequencing to analyze urinary microbiota content. Then, they used the canine's diagnoses to train an ANN to detect significant peaks in the GC-MS data.
The canine olfaction system was 71% sensitive and between 70% to 76% specific at detecting Gleason 9 prostate cancer, wrote lead author Claire Guest of Medical Detection Dogs. In addition, the trained ANN identified regions of interest in the GC-MS data, which had been pointed out by the canine diagnoses.
On the basis of VOCs collected by headspace solid-phase microextraction (SPME) and analyzed by GC-MS, the researchers used raw ion chromatographs to train the ANN to emulate canine cancer diagnoses of urine samples. Both network skeletonization and auto-associative filtering techniques revealed the most important chromatograph peaks contributing to the canine diagnosis.
The scientists believe they have established basic methodology and feasibility that could be used to integrate canine olfaction, urinary VOCs, and urinary microbiota profiling into larger-scale studies that would point the way to machine olfaction diagnostic tools.
Scalable multidisciplinary tools may then be compared to PSA screening for earlier, noninvasive, more specific, and sensitive detection of clinically aggressive prostate cancers in urine samples, according to the researchers.
"Our results indicate that there may be information synthesized by the dogs regarding the nature of cancer that may not be readily identified by traditional single channel molecular biomarker analysis, and may instead be an emergent property," wrote the authors.
The next phase of research will involve comparing canine olfaction and chemical and microbial profiling with different diagnostic tests that are specific molecular ID biomarker-driven diagnostic metrics, unlike the olfactory approach which depends on the emergent property of scent character.