Artificial Neural Networks Could Learn from Cancer-Detecting Dogs

A new study has demonstrated the ability of dogs to detect aggressive prostate cancer from urine samples and suggests that an artificial neural network could learn from this olfactory ability, with an eye toward replicating it in novel detection tools. The research team from the Claire Guest of Medical Detection Dogs published their findings in PLOS ONE.

The researchers set out to combine the strengths of canine olfaction with other promising detection methods in order to surface new insights that could aid development of better prostate cancer tests. They trained two dogs to detect aggressive prostate cancer from urine samples. These dogs showed 71 percent sensitivity and 70 to 76 percent specificity in detecting prostate cancer with a Gleason score of 9, indicating highly aggressive disease.

The team also applied two laboratory detection methods to the urine samples: Gas chromatography-mass spectroscopy analysis of volatile compounds and analysis of microbial species found naturally in urine. Both methods surfaced key differences between cancer-positive and cancer-negative samples.

Finally, the researchers used the dogs' data to train an artificial neural network to identify specific portions of the spectroscopy data that contributed significantly to the dogs' diagnoses. This also revealed specific differences between positive and negative samples. The findings suggest that larger studies could further integrate these disparate methodologies in order to improve detection of advanced prostate cancer and aid development of new diagnostic tools that replicate dogs' olfactory capabilities. The authors add, "We've shown it is possible to replicate the dog's performance as sensors and brains, it is now time to put this technology in every smartphone."