A path has opened to diagnose prostate cancer early through the smell of urine, addressing the disease that has risen to become the most common cancer among Korean men.
Yonsei University Gangnam Severance Hospital said Wednesday that Professor Koo Kyo-chul of the Department of Urology, together with Professor Park Tai-hyun of the Department of Food Science and Nutrition at Ewha Womans University, developed an artificial intelligence (AI) model for early prostate cancer diagnosis based on an olfactory biosensor and proved its usefulness.
In the "2023 National Cancer Registration Statistics" released earlier this year by the Ministry of Health and Welfare and the Central Cancer Registry, prostate cancer overtook lung cancer to become the most common cancer among men. In 2023, there were 22,640 newly diagnosed cases of prostate cancer, about 15.6 times more than in 1999 (1,454 cases), when related statistics were first released. Prostate cancer can be diagnosed relatively easily simply by checking the level of prostate-specific antigen (PSA) in the blood. Generally, a PSA level of 4.0 ng/mL or higher raises suspicion of prostate cancer. However, it has a limitation in that the specificity of the standalone test for prostate cancer is not very high, since PSA can also rise in various prostate conditions such as benign prostatic hyperplasia and prostatitis, as well as in conditions like acute urinary retention.
The research team planned the study based on the judgment that a non-invasive diagnostic method was needed to complement the shortcomings of the PSA test, which had imposed physical and financial burdens by requiring additional painful biopsies even in patients without prostate cancer. The clue was found in past research data showing that a trained detection dog (German Shepherd) had distinguished prostate cancer with high accuracy by smelling patients' urine.
First, the team extracted six types of human olfactory receptor proteins and then bound them to a "Nanodisc," a microscopic artificial cell-membrane nanoparticle made of lipid components. This induced a phenomenon in which odor molecules emitted from urine affected the sensor, causing a reduction in the fluorescent signal. Afterward, by measuring the minute signal change values and combining them with a machine learning algorithm based on pattern analysis, they trained the system on the urine patterns of prostate cancer patients.
For thorough verification of the results, the research team applied strict criteria to select 73 people from a candidate pool of 143 participants, dividing them into a prostate cancer patient group (40 people) and a control group (33 people). In addition, to further improve the precision of the AI model, they built a set of 290 urine samples and performed cross-validation. As a result, they identified three key receptors (OR2W1, OR51E1, OR51E2) that play the most decisive role in detecting prostate cancer molecules, and based on these, distinguished patterns between cancer patients and healthy individuals.
According to the analysis, the accuracy of the model developed this time was 0.89. This means it can accurately distinguish between prostate cancer patients and healthy individuals with an 89% probability. When the hit rate of measurement, which simultaneously incorporates the model's sensitivity and specificity, was represented as a graph, the area under the curve (AUC) value was calculated at 0.964±0.01. This means urine from prostate cancer patients and healthy individuals can be distinguished with 96.4% accuracy, which belongs to the top grade in statistics. The research team plans to develop it into a diagnostic technology usable in actual clinical settings through large-scale clinical trials in the future.
Professor Koo Kyo-chul stressed, "It is highly significant in that it presented the possibility of diagnosing prostate cancer by analyzing urine in a non-invasive and simple manner," adding, "Not only is the hit rate high, but it can also obtain various related information, such as the degree of cancer aggressiveness (Gleason score), which could not be confirmed with the existing PSA test."
The results of this study were published in the latest issue of "ACS Biosensors," a top journal in the field of analytical chemistry.
