ESTRO 2025 - Abstract Book

S3023

Physics - Image acquisition and processing

ESTRO 2025

Figure 1

Figure 2

Conclusion: The findings suggest that generative models pre-trained on HEP simulations, leveraging the abundance and reliability of Monte Carlo-generated data, can be effectively adapted to medical imaging tasks with limited target domain data. This approach highlights a promising method for accelerating training in domains where data acquisition is expensive or constrained by privacy regulations.

Keywords: Deep Learning, CT scans, High-Energy Physics

References: [1] Isola P. et al. “Image-to-image translation with conditional adversarial networks”, 10.48550/arXiv.1611.07004. [2] Simsek E. et al. “CALPAGAN: Calorimetry for Particles using GANs”, 10.1093/ptep/ptae106. [3] Agostinelli S. et al. “Geant4 - A Simulation Toolkit”, 10.1016/S0168-9002(03)01368-8. [4] National Lung Screening Trial Research Team. “The National Lung Screening Trial: overview and study design”, 10.1148/radiol.10091808. [5] Pezoulas V. C. et al. “Synthetic data generation methods in healthcare: A review on open-source tools and methods”, doi: 10.1016/j.csbj.2024.07.005. [6] Kulathilake K. A. S. H. et al. “A review on Deep Learning approaches for low-dose Computed Tomography restoration”, 10.1007/s40747-021-00405-x.