Estimating Spatially Resolved Radiation Fields Using Neural Networks

2025-12-19

AI Summary

This study addresses the challenge of real-time, high-resolution spatial characterization of scattered radiation fields in interventional radiology and cardiac catheterization. We propose a deep learning–based dose field reconstruction method. Methodologically, we systematically construct three progressively complex synthetic datasets using Geant4 Monte Carlo simulations to generate spatially and spectrally labeled training samples. A hybrid neural network architecture—integrating convolutional and fully connected layers—is designed to jointly reconstruct scattered radiation fluence and energy spectra. Our key contributions include establishing the first open-source, reproducible benchmark framework for spatial-spectral radiation field reconstruction, encompassing datasets, simulation protocols, and training pipelines. Experimental results demonstrate strong generalizability and clinically relevant accuracy: spatial resolution at the centimeter level and spectral reconstruction error below 8%. This work introduces a novel paradigm for real-time radiation protection dosimetry assessment.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Computer Vision: 3D Computer Vision

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

We present an in-depth analysis on how to build and train neural networks to estimate the spatial distribution of scattered radiation fields for radiation protection dosimetry in medical radiation fields, such as those found in Interventional Radiology and Cardiology. Therefore, we present three different synthetically generated datasets with increasing complexity for training, using a Monte-Carlo Simulation application based on Geant4. On those datasets, we evaluate convolutional and fully connected architectures of neural networks to demonstrate which design decisions work well for reconstructing the fluence and spectra distributions over the spatial domain of such radiation fields. All used datasets as well as our training pipeline are published as open source in separate repositories.

Problem

• Estimating scattered radiation fields spatially using neural networks
• Training networks on synthetic datasets from Monte-Carlo simulations
• Evaluating neural architectures for fluence and spectra reconstruction

Innovation

• Neural networks estimate scattered radiation spatial distribution
• Monte-Carlo Simulation generates synthetic training datasets
• Convolutional and fully connected architectures reconstruct fluence spectra