Given the high prevalence of prostate cancer, and the poor state of current technologies used to screen for the disease, it should come as no surprise that many groups are working on better solutions (such as this one we recently reported on).
The best diagnostic tool available for prostate cancer detection at present is decidedly low-tech — trained dogs. While these cancer-sniffing canines are more than happy to do the job for a pat on the back and a few treats, the training needed to make them expert diagnosticians takes many years and is quite expensive. To be widely used, a more scalable, technological replacement would be needed, or so a team led by researchers at Medical Detection Dogs and MIT makes the case for in a paper just published in PLOS One.
As sensitive as a dog’s nose may be at detecting trace molecules, electronic sensors exist that are actually far more sensitive — up to two hundred times more sensitive in some scenarios. But the problem is not detection; the problem is interpretation. The complex chemical signatures emitted by cancerous cells have eluded our best efforts to understand them, but for a trained dog, it is as simple as a quick sniff.
In an unexpected reversal that will have science fiction writers unsure whether it will be intelligent machines — or now instead dogs — that will lead to the demise of humanity, the trainees have become the trainers. The expert pattern detecting abilities of canines were used by the researchers to label training data for an artificial neural network.
Analytical data was collected from gas chromatography mass spectrometry and microbial profiling. This data was fed into the neural network, and with the assistance of canine classifiers, the patterns were mapped to diagnostic decisions.
In a small, fifty sample test of the new method, the artificial cancer detector was found to perform as well as trained dogs, with better than 70% accuracy observed for both methods. In the future, the team would like to perform a larger scale test of their method as a step towards better assessing clinical utility of the diagnostic tool.