This is an unusual testing laboratory. A rectangular room with a slope in the middle in which treats are given to the dogs. Surrounding it is an oval green ramp with several small stations containing human breath samples. In one corner, a dog trainer holds an EEG helmet and harness that measure neurobehavioral signatures on 2-year-old beagle Chloe, the star of this session.
Once released, Chloe heads down the slope to check the dishes, but finding nothing, she wanders around the oval walkway, stopping at 12 different stations to smell samples of human breath. Cameras on the ceiling and at each station record his every movement, velocity and sniff pattern. The harness and helmet she wears record her respiration, biomarkers and EEG.
Finally, she smells a sample that contains cancer, and heads to the slopes again. This time, there is a treat waiting for him. The entire exercise takes less than a minute. In that one minute, Chloe smelled about 72 samples and identified one of them as cancerous.
We’re at Dognose’s farm in Neelamangala, outside Bengaluru, where fifteen dogs are being trained to smell and detect cancer from human breath. “It’s a weird technology,” admits co-founder and product head Itamar Bitan, who also trains dogs.
Most technicians and hospitals are used to sending blood, urine and breath samples to laboratories where they are analyzed by machines. In the case of a startup, because it involves a biological mammal with individual temperaments and nuances, the efficacy of the test is questionable. “Because it is unconventional, it has to be 10 times better than other tests for it to be adopted,” admits the 27-year-old. Bitan, who is from Israel, served in the Israeli Army for four years until 2020 in a canine unit for bomb detection. Before starting Dignosis in India, he did a similar startup in canine scent to sniff out Covid-19.
Dogs Can Smell Diseases, But How Do They Tell You?
We’ve known for nearly two decades that dogs can smell diseases in humans. Dog smells amazing. They have up to 300 million olfactory receptors. Humans number 5-6 million. Studies around the world have proven that dogs can smell volatile organic compounds (VOCs) linked to a variety of diseases, from cancer to COVID-19, malaria to Parkinson’s disease.
The only obstacle is the human interpretation of what the dogs are smelling. Currently canine olfactory technology requires a dog that can express what it is smelling and a trained human to analyze what the dog is communicating. Dognosis wants to standardize this process using video data, sensors, AI, and machine learning. “Can we create data from sensors and put machine learning models on it to predict and analyze the mental state of dogs through their behavior and their facial features?” asks Aakash Kulgod, the startup’s other co-founder.
In 2023, Kulgod, then 26, contacted Bitton on LinkedIn with the idea and the two started Dognosis. The startup raised $1.6 million from two venture capitalists who loved doing “crazy sci-fi ideas,” giving them three years of runway to build a lab, ML platform, and a team of 50 employees aimed at validating their idea.
Discovery of olfactory standardization
The big gap in technology – which is why only three startups worldwide are attempting this – is not only biological, but also because unlike bombs where dogs are trained on specific chemicals, people have not been able to define which chemicals the dogs are sniffing out as indicating cancer. “These chemicals are too low resolution and variable across individuals and environments to quantify,” explains Bitton.
Instead of decoding and annotating scents, which has been an impossible problem until now, Dognose maps individual dogs using multiple sensors. Each dog has a digital twin, called a sniff diary, which contains historical data, preferences, tendencies, behavior and sensor data, respiration, brain data, movement as well as trainer annotations of each session. This data is fed into their ML system and analyzed to find similarities. The model creates customized signatures and stories for each individual dog. “Our aim is to create an impression of the individual dog so that we know that if this dog behaves in a particular way in front of the sample, it means that the sample is cancer positive,” Bitton says.
To be careful, each sample is sniffed by several dogs. The AI model combines the outputs of multiple dogs and gives results accordingly, giving less importance if the dog is not in a good mood behaviorally. Kulgod hopes the AI trainer will get good enough to spot something the human trainer couldn’t catch.
Over the last 18 months or so, the startup has partnered with six Karnataka-based hospitals and conducted clinical trials on over 3000 people. Earlier in April, they published a first-of-its-kind report of a successful trial in the Journal of Clinical Oncology, which concluded that canine olfaction demonstrated high analytical accuracy in detecting multi-cancers from breath. Thanks to eleven family doctors, Kulgod was able to expedite the testing. The oncologist leading the study, Dr. Sanjeev Kulgod of RedOne Cancer Center in Hubli, is his uncle.
scaling canine olfaction
Currently, the startup has 15 dogs and can conduct five lakh tests per year. “You can go from an unknown dog adopted from a shelter to smelling cancer in eight weeks,” says Kulgod, “and there is no shortage of dogs.” The startup has a manual for finding motivated dogs who can do this repeatedly without getting bored. “They have to love food a lot,” Bitton says. The dogs don’t receive a salary yet, but they get long walks, plenty of play time and regular vet visits, as well as dry meat, fruits and vegetables to keep their tummies healthy. They work for 30 minutes a day.
Eight months ago, after feedback from its dogs, Dognose designed a new lab to make it dog-friendly. The oval ramp we saw enables dogs to run around, reduce biases, and sniff out faster. Automating sample swapping and presentation of 72 samples makes it faster so that dogs don’t get bored or frustrated with the manual swapping or cleaning of stations that used to happen. Bitton explains that the higher number of samples also ensures that the dog catches at least one cancer sample and is rewarded.
Their collection process has also been streamlined. Participants breathe into a mask for ten minutes, which is then sealed in a container, cooled in a vaccine box, and brought to the laboratory to be stored at approximately 4–5 °C before being introduced to the dogs.
The holy grail of cancer screening
As a startup, Dognosys is aiming for the holy grail of screening tests – a multicancer screening test. “Cancer is the largest, most mature market for preventive care, so that’s where we’re focused,” Bitton says. Their aim is to create a home testing kit for the healthy population to complement the approved screening tests. “Before a mammogram or Pap smear, you can use our test to see if you’re likely to have cancer,” says Kulgod.
Although one would think it would be easy, it turned out to be difficult to raise a new round of funding even after studies that proved their idea. “We talked to 80-100 VCs and got a lot of rejections, saying ‘dogs can’t be the future of medicine; dogs can’t scale; you can’t sell this product’,” says Kulgod, adding that the human bias that dogs can’t be standardized for the lab will continue, even if they show that dog scent can be scaled and standardized.
Despite critics, the startup is working towards launching a product next year and plans to launch its own home testing kit for cancer screening. “That’s the level of evidence we have,” Bitton says. Concurrently, the startup is collaborating with hospitals in different states across the country to run more clinical trials. His laboratory is also being replicated in the UK in Medical Detection Dogs which has been testing dogs’ sense of smell since 2004. They have eventually secured an undisclosed tranche of funding from a global venture capital to keep them running for the next three years.
In the laboratory, a cat roams the area eyeing everyone with democratic suspicion. No one has created a predictive, multimodal algorithm for canine olfaction before. “It’s pretty crazy,” Kulgod reflects, but so are other trials in the world. To detect cancer, a PET CT scan involves fasting, inserting a radiotracer into the patient’s vein, putting the person in a very expensive magnet, and analyzing multiple images made by the machine. On the other hand, the diagnosis test requires up to ten minutes of breathing time in a mask to produce a sample.
“A dog can sniff out 250 samples in 30 minutes, each sample takes a few seconds and these dogs can be adopted from shelters and trained within eight weeks and all they ask for in return is dog food,” says Kulagod.
He asks which technology is more sustainable.
(Shweta Taneja tracks the evolving relationship between science, technology, and modern society. She also works as a philanthropy researcher and consultant.)
