Patients with hepatocellular carcinoma (HCC) have poor outcomes as a result of late detection of the disease. We investigated the possibility of using smell detection by dogs for detecting HCC from the breath of patients. Patients whose diagnosis of HCC was confirmed histologically or radiologically according to the American Association for the Study of Liver Diseases criteria had breaths collected using face masks and transported to the study test site. The numbering of the HCC samples was sent in a sealed envelope to blind the dog trainer during testing but allow for correct rewarding of the dog afterwards. One golden retriever was trained to detect HCC with positive feedback using known samples of HCC and healthy controls in a step-wise manner. The controls were selected from hospital staff and relatives of patients who were not involved in the study. They were questioned about the risks of their disease before selection. When the trainer was confident that the dog could recognize the HCC scent, blind testing was performed using 1 HCC : 3 healthy controls per test run. Once the dog signaled on a specimen, it was given a reward. The correct-detection rate was compared to the theoretical detection rate expected based on chance of 25% using the statistical one-sample test of proportions. Thirty-seven HCC patients were tested. The patients had a mean age of 58 years and 21/37 were male. Seventeen patients had hepatitis B and 14 patients had hepatitis C. Twenty-six patients had one HCC lesion; four patients had two lesions in the liver, whilst seven had many lesions. The number of patients in the very early, early, intermediate, advanced, and terminal stages of the Barcelona Clinic Liver Cancer classification was 5, 9, 21, 1, and 1, respectively. The dog detected correctly in 29 runs. The sensitivity for canine detection was 78% (95% CI: 62%-90%). Compared to the 25% correct indication expected based on chance, this was statistically significant (p < 0.001).
Conclusion: This is the first study to look at the possibility of detecting HCC from breath using canine olfaction. Our results show that this is possible with an accuracy of 78% (p < 0.001 when compared to chance alone), and are thus a proof of concept. Further refinement of the process of detection will be needed before clinical application.