Preclinical PET imaging relies on anatomical information—traditionally obtained from CT—for attenuation and scatter correction and for accurately mapping radiotracer biodistribution. Recently, a strong trend toward low-dose CT has emerged, driven by the need to reduce radiation burden and streamline high-throughput imaging workflows. However, even low-dose CT adds complexity and exposes animals to unnecessary irradiation, especially when structural information is required only for anatomical mapping and not for diagnosis. Despite this, there is currently no solution capable of providing CT-equivalent anatomical information directly from PET data without any CT scan.
SynthesizeCT aims to go beyond the state-of-the-art by developing a physics-informed Generative AI (GenAI) model that synthesises 3D anatomical maps directly from PET acquisitions, completely eliminating the need for CT irradiation. The approach exploits PET images, statistical properties of sinograms, and Monte Carlo simulated ground-truth scatter and attenuation maps to reconstruct high fidelity anatomical structures with synthetic CT images.
The methodology integrates:
i) Monte Carlo simulations to generate accurate ground-truth scatter maps and attenuation distributions for the training dataset,
ii) Physics-informed GenAI models to develop 3D CT anatomical maps directly from PET projection data,
iii) High-Performance Computing (HPC) resources to support large-scale Monte Carlo production and computationally demanding deep-learning training,
iv) Preclinical PET datasets acquired on BIOEMTECH’s commercial “eyes-series” 3D imaging systems, ensuring robust training and direct applicability.
BIOEMTECH, a leading Greek SME in biomedical technology and bioinformatics, has extensive experience in developing computational phantoms, AI-driven imaging tools, and integrating computational innovation into commercial imaging platforms. The company has demonstrated strong success in EuroHPC initiatives:
• PediDose (FF4EuroHPC Success Story) – HPC & AI based personalized pediatric dosimetry platform.
• ΔosimetrEYE (FFplus Success Story) – GenAI based tool for 3D preclinical dosimetry from 2D dynamic images.
Building on this experience, SynthesizeCT will deliver a fully innovative, CT-free anatomical synthesis module that will be integrated into BIOEMTECH’s “eyes-series TM” systems, enabling real-time AI-enhanced imaging and direct industrial exploitation. This positions the project at the core of the EuroHPC AI Factory Large-Scale vision: combining HPC with advanced AI to create deployable, high-impact products from European SMEs.
SynthesizeCT offers a disruptive advance: a radiation-free, physics-guided, AI-based anatomical mapping solution that enhances PET quantification, reduces animal burden, accelerates workflows, and unlocks a new generation of intelligent, AI-empowered imaging systems ready for competitive global markets.
Valentina Paneta, BIOEMTECH, Greece