Routine blood test can reveal past radiation exposure, in mice

A routine blood test, the kind run in almost every clinic, may carry a trace of radiation a person received months earlier. At Lawrence Berkeley National Laboratory, a team has built a machine learning tool that reads that trace from a complete blood count, or CBC, the standard panel that measures the cells circulating in the blood. Their first tests were done on mice.

The stakes come from how hard exposure is to detect today. The method used, called biodosimetry, looks for measurable biological changes in the body to establish whether exposure happened and, when possible, how large the absorbed dose was. The existing tests are time-intensive, can only be run by specialized laboratories, and often give results that are inconclusive or hard to read, particularly after low-dose exposure, which is more common and more likely to be overlooked. Jamie Inman, a research scientist at the lab who co-led the study, said the team was surprised by how much information the CBC patterns kept across a wide range of doses and times after exposure.

The approach works differently from earlier studies. Previous work looked at individual CBC parameters, or the ratios between a few of them. Chang's model searches the full set of measurements for combined signals that no single number reveals, and the team says this gives it better translational power when applied to different groups. The training doses were produced by co-authors Kei Nakamura and Lieselotte Obst-Huebl, from the laboratory's accelerator and applied physics division, who established a low-dose radiation environment and refined procedures to quantify doses close to the lower detection limits of standard methods.

The limits are clear, and the researchers state them. The genetic background of the animals changed how well the model worked: performance was lower and varied across mouse strains. A standardized approach that works for anyone remains the central challenge of biodosimetry, since two people can respond very differently to the same dose. Inman's own caveat is that the findings suggest CBC data could contribute to rapid exposure screening, but confirmatory testing would still be needed. The team is optimistic that the model can be adapted for humans, but that step has not yet been tested.

What changes, and for whom? For now, the direct beneficiaries are researchers, who can use the openly published model and its data to test their own approaches. For clinicians and emergency services, the promise is a cheap first screen, run on a test that already exists in most hospitals, that would flag who needs specialized analysis. The study does not give a timeline for that to happen, and it does not say how the approach would perform on people, whose genetic diversity is at least as wide as that of the mouse populations tested here. The next step is to build models trained on that diversity, then to see whether the signal holds in human blood.