Dogs can detect the unique body odor of infected individuals, which is 10,000 times more sensitive than that of humans. Medical detection dogs have already identified patients with cancer, malaria, and even coronavirus infections. Scientists have combined artificial intelligence (AI) with medical detection dogs to increase the accuracy of cancer tests to 94%.
The so-called hybrid diagnosis, combining dogs and AI, allows the dog to detect human scent while AI analyzes the dog's behavior. With the integration of AI, even types of cancer not previously trained on or early-stage cancer patients have been detected. This represents a new pathway for cancer diagnosis and treatment through the combination of nature's wisdom and human knowledge.
◇ Identified four types of cancer in experiments
Israeli biotech startup SpotitEarly announced earlier this month that it has developed a bio-AI hybrid cancer screening method that successfully detects cancer early. The company revealed that medical detection dogs, combined with AI, identified breast cancer, lung cancer, prostate cancer, and colorectal cancer with over 94% accuracy. The results of this research were published last month in the scientific journal Scientific Reports, a sister journal of Nature.
SpotitEarly collected breath samples from 1,386 individuals suspected of malignant tumors in four Israeli medical institutions. Masks worn by patients for 5 minutes were collected and sealed. Individuals who smoked or drank alcohol, which could affect body odor, were excluded. Among the study subjects, 6 out of 10 were male, with an average age of 56.
The results of biopsies conducted at hospitals revealed that 1,048 individuals (75.6%) tested negative for cancer, while 338 individuals (24.4%) were positively diagnosed with cancer. With the aid of AI, the sensitivity in identifying patients with four types of cancer through mask odor was 93.9%, and its specificity in distinguishing healthy individuals was 94.3%. This was nearly consistent with the biopsy results, particularly identifying early-stage cancer patients with 94.8% accuracy.
Medical detection dogs have been trained to sit when they detect cancer in human scent, and if they do not, they move on to the next sample. The dogs can sit or move to the side in less than one second. The challenge is that it is not easy to consistently read the dogs' body language. It is difficult to determine whether they correctly sat down or hesitated if they moved slowly. This is why AI is needed.
The research team at SpotitEarly trained an AI model to interpret the signals from the dogs. Experiment results showed that when the dogs conducted tests alone, they identified cancer patients with 64.4% accuracy, but when paired with AI, there was an improvement of over 30% in accuracy. The bio-AI hybrid system proved to be superior to the AI analyzing scent samples alone.
Dr. Len Lichtenfeld, chief medical officer at SpotitEarly, said, "We developed a technology that can detect various common cancers early with a single test by combining the olfactory abilities of medical detection dogs with the power of AI," and he noted, "I expect the results of this research will form the foundation for next-generation cancer screenings."
The company aims to conduct large-scale clinical trials in the United States, targeting to report early results by 2026. The experiment has trained the Labrador Retriever breed, which has been extensively involved in medical detection, and plans to prepare Beagles next, as they are smaller and easier to train.
◇ Detecting everything from cancer to malaria and COVID infections
Dogs can find cancer patients by scent because their sense of smell is 1,000 to 10,000 times more powerful than that of humans. Dogs possess 300 million olfactory receptor proteins, vastly outnumbering humans' 5 to 6 million. These receptor proteins are attached to the surface of thin bone structures in a maze-like arrangement, with a surface area in dogs being 150 to 170㎠, much larger than humans' 5 to 10㎠. This allows for a greater reaction between odor molecules and olfactory receptors. Additionally, although their brain size is only one-tenth that of humans, their smell processing area is three times larger.
Medical detection dogs have gained attention since 16 years ago. Dr. Claire Guest, a psychologist from the UK, established the Medical Detection Dogs Foundation in 2008. He felt an unusual sense of alarm when his Labrador companion Daisy pressed her nose against his chest and tapped with her paw, prompting him to visit the hospital. Thanks to Daisy, Dr. Guest detected breast cancer early and saved his life.
Medical detection dogs have diagnosed various diseases over the years. In 2015, Italian researchers announced that two German Shepherds identified biopsy samples from prostate cancer patients with 98% accuracy. Researchers from Durham University in the UK and the Medical Detection Dogs Foundation reported in 2019 in the medical journal The Lancet Infectious Diseases that detection dogs detected malaria-infected children through the scent of socks worn by them with 70% accuracy.
They have also identified COVID patients. A research team led by Professor Cynthia Otto at the University of Pennsylvania's School of Veterinary Medicine reported in April 2021 in PLoS ONE that detection dogs identified COVID infections from urine and saliva samples with 96% accuracy. That same year, the Bristol County sheriff in Massachusetts became the first in the U.S. to introduce COVID detection dogs.
Recently, medical detection dogs have even been used to detect mental health issues. They can live alongside patients with unstable psychological conditions in hospitals and recognize unusual signs through scent. In 2022, researchers from Queen's University in Belfast, UK, confirmed in PLoS ONE that dogs can distinguish between stressed individuals and those who are not through sweat and breath. With the addition of AI allies, it is expected that medical detection dogs will see even more active roles.
References
Scientific Reports(2024), DOI: https://doi.org/10.1038/s41598-024-79383-2
PLoS ONE(2022), DOI: https://doi.org/10.1371/journal.pone.0274143
PLoS ONE(2021), DOI: https://doi.org/10.1371/journal.pone.0250158
The Lancet Infectious Disease(2019), DOI: https://doi.org/10.1016/S1473-3099(19)30220-8
The Journal of Urology(2015), DOI: https://doi.org/10.1016/j.juro.2014.09.099