Jun 7 – 11, 2026
Prague, Czechia
Europe/Prague timezone

CFPD-Informed Reduced-Order and Artificial Intelligence Framework for Mechanistic Inhalation Dosimetry Across Emergency Response and Population Applications

Jun 11, 2026, 11:00 AM
15m
Auditorium 115

Auditorium 115

Břehová 7, Prague 1
Oral Presentation Computational methods and modelling in dosimetry Computational methods and modelling in dosimetry

Speaker

Shaheen Dewji (Georgia Institute of Technology)

Description

Mechanistic inhalation dosimetry for airborne radionuclides requires spatially resolved particle deposition profiles within anatomically realistic respiratory tract geometries. Compartmental approaches, including those of the International Commission on Radiological Protection (ICRP), assume regionally uniform deposition and population-averaged morphometry, limiting resolution for emergency response and individualized exposure assessment. High-fidelity Computational Fluid and Particle Dynamics (CFPD) simulations address these limitations but remain computationally intensive.

This work presents an integrated framework in which verified subject-specific CFPD simulations form the mechanistic basis for reduced-order and AI-driven surrogate modeling. Three-dimensional airway geometries reconstructed from computed tomography scans were processed through an automated workflow for segmentation, meshing, and solver preparation. Transient airflow and Lagrangian particle tracking were solved under realistic breathing conditions to generate detailed particle deposition profiles for comparison with ICRP reference predictions.

To reduce computational burden while preserving dominant flow and deposition structure, Dynamic Mode Decomposition (DMD) was applied to reconstruct respiratory velocity fields from reduced modal representations. In parallel, geometry-conditioned neural network models, including conditional diffusion-based approaches, were trained to emulate particle trajectory and deposition behavior directly from anatomical descriptors and breathing parameters. These reduced-order models reproduce regional deposition trends while enabling rapid prediction across diverse airway morphologies.

The combined CFPD-ROM-AI framework provides a scalable platform for emergency response dose estimation following airborne radionuclide release. Beyond emergency applications, the methodology supports occupational, environmental, and medical internal dosimetry by coupling subject-specific deposition profiles with radiation transport solvers for absorbed dose calculation. This approach balances anatomical specificity, mechanistic structure, and computational efficiency in advanced inhalation dosimetry assessment.

Author

Shaheen Dewji (Georgia Institute of Technology)

Co-authors

Mr Ignacio Bartol (Georgia Institute of Technology) Mr Martin Graffigna (Georgia Institute of Technology) Dr Mauricio Tano (Idaho National Laboratory)

Presentation materials

There are no materials yet.