Decoding Odor Mixtures in the Dog Brain: An Awake fMRI Study

• Published in: Chemical senses 2020-12, Vol.45 (9), p.833-844
• Main Authors: Prichard, Ashley, Chhibber, Raveena, King, Jon, Athanassiades, Kate, Spivak, Mark, Berns, Gregory S
• Format: Article
• Language: English
• ISSN: 0379-864X; 1464-3553
• DOI: 10.1093/chemse/bjaa068

Abstract In working and practical contexts, dogs rely upon their ability to discriminate a target odor from distracting odors and other sensory stimuli. Using awake functional magnetic resonance imaging (fMRI) in 18 dogs, we examined the neural mechanisms underlying odor discrimination between 2 odors and a mixture of the odors. Neural activation was measured during the presentation of a target odor (A) associated with a food reward, a distractor odor (B) associated with nothing, and a mixture of the two odors (A+B). Changes in neural activation during the presentations of the odor stimuli in individual dogs were measured over time within three regions known to be involved with odor processing: the caudate nucleus, the amygdala, and the olfactory bulbs. Average activation within the amygdala showed that dogs maximally differentiated between odor stimuli based on the stimulus-reward associations by the first run, while activation to the mixture (A+B) was most similar to the no-reward (B) stimulus. To clarify the neural representation of odor mixtures in the dog brain, we used a random forest classifier to compare multilabel (elemental) versus multiclass (configural) models. The multiclass model performed much better than the multilabel (weighted-F1 0.44 vs. 0.14), suggesting the odor mixture was processed configurally. Analysis of the subset of high-performing dogs’ brain classification metrics revealed a network of olfactory information-carrying brain regions that included the amygdala, piriform cortex, and posterior cingulate. These results add further evidence for the configural processing of odor mixtures in dogs and suggest a novel way to identify high-performers based on brain classification metrics.