The semiotic canine: scent processing dogs as research assistants in biomedical and environmental research
The use of dogs in biomedical diagnosis, detection and alert as well as for the search and monitoring of species-at-risk is an emerging field of research. Standard practices are converging towards models that are not necessarily consistent with the well established field of (animal) psychophysics. We briefly discuss the different challenges of applied canine olfactory processing and discuss the adoption of more valid and reliable methods. For mostly historical reasons it seems, scent processing dogs are trained and tested using multiple alternative stimuli in choice tasks (e.g., line-ups including 6 alternative choices, or 6AFC). Data from psychophysics suggest that those methods will reduce or at the very least misrepresent the accuracy of canines. Unless canines are an exception to the rule, sensory, perceptual and cognitive arguments (e.g., Gadbois & Reeve, 2014) can be made against most multiple alternative forced choice tasks (mAFC’s) in favor of detection tasks (yes/no and go/no-go procedures) or, as a compromise, simpler discrimination tasks (2AFC or 3AFC at most). We encourage the use of Signal Detection Theory as it focusses on two important factors in defining the validity and reliability of scent processing dogs: 1) It is a robust measure of sensitivity, an important factor in both diagnosis and sensory detection, and, 2) It describes the type of errors (false alarms vs. misses) that a given dog is most likely to commit, allowing for a solid assessment of performance and potentially a readjustment in training. We give an example with Diabetes Alert Dogs (DAD’s) specialized in Hypoglycemia Detection in vitro and discuss the potential advantages of keeping a low number of alternatives during training and testing, the importance of low saliency training (LST), as well as adopting pure detection tasks requiring a response commitment from the dogs for both “yes” and “no” responses. The value of d’ (a detectability or discriminability measure) and bias measures (criterion) are discussed in the context of canine selection, performance assessment and diagnostic accuracy across applications.