research (cards) | Georg Schramm

Our group at KU Leuven develops the mathematical models, reconstruction algorithms, and machine-learning methods that turn raw scanner data into accurate, quantitative medical images. We work across molecular imaging modalities — primarily PET and SPECT, and increasingly MR — with a strong emphasis on open, reproducible, and computationally sustainable software.

Research themes

Fast, scalable PET/SPECT reconstruction

Accelerated, model-based reconstruction algorithms and the high-performance projectors behind them, for state-of-the-art image quality at clinically practical speed.

parallelproj →

Deep learning for low-dose & accelerated PET

Dose- and time-reducing deep learning for PET — anatomy-guided denoising and listmode DL reconstruction. Winner of the PETRIC and ultra-low-dose PET challenges.

pyapetnet →

Sustainable, compute-efficient ML

Memory- and compute-efficient training, inference, and preconditioning — making advanced imaging ML practical, reproducible, and energy-aware.

Sodium (²³Na) MR reconstruction

Extending our reconstruction expertise to sodium (²³Na) MRI — quantitative methods for a challenging, low-SNR modality that probes tissue viability and metabolism.

Funded projects