Reading Bladder Cancer in the Smell of Urine: An Early Sensor Study

Can a machine learn to recognise disease from the smell of urine? This 2013 pilot study set out to test exactly that, using a sensor device to analyse the gases rising from urine samples.

What the researchers did

The team collected 98 urine samples — 24 from people with bladder cancer and 74 from controls with urinary symptoms but no cancer. They captured the volatile organic compounds (VOCs) in the headspace (the air above each sample), separated them, and fed the resulting gas-sensor patterns into a statistical classifier. The model was then validated using a leave-one-out cross-check, where each sample is tested against a model trained on all the others.

What they found

• The full model correctly identified 24 out of 24 cancer cases (100% sensitivity) and 70 of 74 controls (about 95% specificity).
• Under the stricter cross-validation, performance remained strong: roughly 96% of cancer cases and 93% of controls were correctly classified.
• Remarkably, just nine sensor features were enough to separate the groups — meaning the discriminating signal does not require exhaustive chemical detail.

Why it matters

This was one of the earliest demonstrations that a relatively simple sensor approach — rather than a full laboratory spectrometer — could read meaningful health signals from urine vapour. While this is research into disease detection and not a consumer diagnostic, it underlines a broader point relevant to gut-health science: urine’s volatile chemistry is information-rich, and many of the compounds involved share an origin with gut-microbial metabolism. The authors were careful to note the obvious next step.

“The sensor device showed potential for the diagnosis of bladder cancer, though the data needed to be reproduced in a larger study.”