A review paper examining the use of trained medical detection dogs suggests that canine olfaction can support early cancer detection by identifying volatile organic compounds present in human biological samples.
Artificial intelligence applications have advanced diagnostic accuracy in imaging, genomics, and clinical decision support, but these systems require structured datasets and specialized infrastructure, which can limit implementation in some clinical and community settings. Trained dogs have previously identified disease-associated volatile compounds in conditions such as malaria, hypoglycemia, and COVID-19 in controlled investigations, with reported diagnostic performance varying by protocol and setting.
This paper – published in Frontiers in Medicine – offers a potential solution to Hong Kong’s cancer burden. Cancer accounted for approximately one quarter of deaths in Hong Kong in 2022, and population-based screening is currently limited to colorectal cancer for adults between 50 and 75 years of age. Screening programs for cancers such as lung, liver, breast, and pancreatic disease are not routinely implemented, contributing to a substantial proportion of diagnoses occurring at later stages.
Hong Kong’s existing use of detection dogs in customs operations demonstrates operational capacity for canine deployment. Medical biodetection, however, differs substantially from law-enforcement contexts given the variability of disease-related volatile organic compounds. Factors such as environmental odor complexity and air pollution in urban settings may further influence performance and require evaluation.
Investigations into canine detection have shown that dogs can be trained to recognize scent profiles associated with certain cancers using structured exposure to biological samples from patients with confirmed diagnoses, comparison groups with overlapping clinical features, and healthy controls. Training relies on repeated sample presentation and positive reinforcement. The approach has been described as conceptually similar to supervised machine learning, in which pattern recognition improves with the diversity and quality of training examples. Validation methods include blinded testing to measure accuracy in distinguishing target samples from nontarget samples.
The authors propose potential applications in Hong Kong that involve using detection dogs as an initial prescreening step in community or outpatient settings, followed by confirmatory imaging or laboratory testing when indicated. Considerations for feasibility include training standardization, handler expertise, environmental effects on volatile compound detection, animal welfare requirements, and operational costs.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
