Learning from 3D image reconstruction: Reconstructing spatiotemporal distributions of radioactive iodine after...

Determining the spatial and temporal distribution of airborne radioactive iodine isotopes is crucial for effectively managing nuclear emergencies and implementing protective measures for the population. Existing methods rely on release estimates, weather data, and dispersion models-but they do not incorporate indirect data such as measurements from affected individuals. Our research specifically focused on using innovative AI methods to incorporate movement profiles and respective thyroid iodine activity measurements of individuals in an affected area as a new data source in order to reconstruct a spatial and temporal distribution of iodine concentration in the air. To our knowledge, this is the first study to use such data for reconstructing environmental contamination fields. In the absence of real-world emergency data, we developed a highly configurable simulator capable of generating realistic and error-prone movement profiles and related exposures. We then evaluated several AI-based reconstruction methods. The best results were achieved using Neural Radiance Fields and 3D Gaussian Splatting-state-of-the-art 3D image reconstruction techniques adapted here for environmental modeling. Our findings show that meaningful reconstructions are possible, but depend on the amount of personal data available during an emergency. The reconstruction quality strongly depends on the ratio between spatiotemporal resolution and the number of available movement profiles: high resolution with sparse data leads to increased problem difficulty and poor reconstruction results, while lower resolution combined with a higher number of paths enables more accurate and stable reconstructions. This work introduces a new class of reconstruction methods for environmental contamination scenarios.