My biomedical scent detection project

Guest blog: Mirkka Koivusalo, MSc, PhD, KPA CTP

For the past year and a half I have been doing a project at a hospital in Toronto (a collaborative project with researchers at Dalhousie University in Halifax) which has been a dream come true in a sense: I have been able to combine my PhD education and 15+ years of experience as a researcher in cell biology and my love of training dogs. The project is in a relatively new and emerging field, biomedical scent detection research, the purpose of which is to investigate whether dogs’ sense of smell can be used to diagnose diseases. In my case the focus is on training dogs to detect a pathogenic micro-organism which is a major concern in healthcare facilities and certain communities. I still need to be cryptic here as the study is not published as we speak, but I want to share some observations and personal thoughts that have been on my mind regarding the training aspect of it.

What did surprise me is that the nature of the research is more or less the same whether you are looking at one cell under a fluorescence microscope or training a whole organism. In both cases you are training the system to obey your will in a way so that you can hopefully prove your hypothesis. On the way there you hit a lot of roadblocks, you take wrong turns and sometimes travel on the wrong path a long way until you realize that you need to backtrack to the previous intersection. It is a lot of fumbling in the dark until you see a glimpse a light.

My own Australian Shepherds Forbes and Flirt were my first prototypes that I tested everything on. Because we are dealing with dogs and not a microscope, the challenge is that when I make a mistake, my ‘instrument’ can get very frustrated. In training the trainer should raise criteria incrementally to help the learner get the concept. In my case the challenge is that the dogs should consistently distinguish samples from each other that are extremely similar in scent, and do this even when the scent changes within a sample. I like to think of it as someone presenting me with two extremely similar shades of the same colour, they are different from each but they each have slightly different hues within them as well. Would I be able to tell the difference between these shades of the same colour consistently every day, if I haven’t seen the colours that day yet.  Not surprisingly, it has turned out that I am very blind to the criteria change when presenting the samples to the dogs: what I think is easy or difficult for the dogs to distinguish clearly has been proven wrong by them, after all I cannot smell anything at all. This has lead to a lot of frustration, and as I made most initial tests of some crazy criteria changes with Forbes and Flirt, I have actually let them retire since, and moved onto fresh dogs with whom the aim is to do experiments more systematically. My dogs contributed to the learning curve.

What surprises a lot of people is that the research job for the dogs is actually not particularly glamorous. When you see scent detection dogs at the airport  looking for drugs or money on belts of suitcases, it looks exciting. But in biomedical scent research the dog is a lab assistant, who performs the same task over and over again so that we get statistical significance, and to the untrained eye it looks plain boring after a while. Which brings me the question of what kind of dog is best suited for the purpose? It would be great if we could train more or less any dog for the job, and with enough time and resources that might be doable, but dogs certainly have a lot of individual differences, and I think it comes down to a combination of trainability, sense of smell and temperament to be able to handle the repetitive nature of the task.

The absolute star of the study has been a female Golden Retriever Nimbus, we have collected nice data sets with her, and her owner’s dedication in bringing the dog to me -be it rain, sleet, snow or shine- has been phenomenal. I am not entirely sure what it is about Nimbus, but she is able to overcome her frustration during training and seems to solve roadblocks by using her nose. Here is Nimbus showing what our read-out is: a line-up of three sample stations in a double-blind trial, she indicates the positive sample with her lovely long-duration nose targeting behaviour and then gets rewarded also with extra bonus freeze-dried lamb treats:

On the other hand, my Australian Shepherds have been a mind boggle: they get the task but at some point they seem to start using anything but their noses to get to the reward, and then the frustration of them not getting their way may throw them off for the rest of the day. You also need a dog, whose temperament is very determined when they try and get the reward, ultimately only the dog knows which is the right answer, and they need to be able to work independently. I had a lovely young water dog come in, she was an absolute delight to train and picked up the idea of discrimination fast, but probably due to her young age, she was hesitant in being able to trust her own choices. I have given a relatively fair try to the above types, all of them are able to solve the task by using their nose, but when the discrimination gets tough, it is easy for the experiment to fall apart.

So where are we now with the research? We are training and recruiting new dogs to replicate experiments similar to what Nimbus has already done so that we can publish our first study. Publication of well controlled studies showing that dogs can be used as a diagnostic tool is the key in getting funding agencies convinced that these are studies worth funding (a paradigm in itself, how can we do it in the first place without funding?!). The topic of biomedical scent detection is very media sexy, dogs detecting cancers and other diseases easily make the news, BUT is the published evidence out there strong enough to show this is a viable diagnostic method? All medical diagnostics, including the use of dogs, should be backed up by research studies and the ability to replicate the results. The number of published studies out there is very small, the methods and controls are variable and reproducibility is still lacking.  Biomedical scent detection research needs a more solid publication record to convince everyone! I am in it to objectively show whether dogs are reliable enough for the task we have chosen, even if it is a ‘no’ answer.  Such is science!

If you are interested in having your dog participate in the study and you are committed to science, we are looking for back-up dogs. You can e-mail me at, here are the criteria:

– The dog comes into training 2-3 times a week for a couple of hours between 10 am-3 pm to Finch and Bathurst in Toronto. The owner would need to either bring the dog, or if the dog is in North York area and close to it a pick-up arrangement is possible.
– The time commitment is first about 2 months to see if the dog picks up on the signal, then another 2-3 months to collect data.
– PLEASE NOTE: this requires owner commitment, as it is not worthwhile for us otherwise to invest in training the dog. Be sure that you are on board with this!
– The compensation is that the dogs who are able to do it will proudly be part of a published scientific study. Also as a perk I will offer a private dog training session of 1 h (value $125 + HST). Plus it is free daycare and lots of mental stimulation and fun for the dog on training days.
– Requirements for the dog: no prior scent detection training is required at all. The dog has to be high energy, VERY willing and motivated to work for food and able to handle a lot of repetition. If the dog is crate trained that is a definite plus, so that they can wait in a crate during breaks. We will favour certain breeds.
– We adhere to current hospital standards and good laboratory practice in handling of the micro-organisms, the risk of colonization is there but it is a very small one. The micro-organisms are secured inside metal baskets, so there is no contact with the dogs’ noses. Something to bear in mind is that most of the bacteria are all around anyways, it is usually only the immunocompromised, who get infected. The owners will be asked to sign a waiver saying that they understand the (very minor) risk involved.