I’m a physicist and physician working on quantitative MRI, inverse problems and machine learning at PTB and TU Berlin.
I combine physical models with learned methods to reconstruct images and tissue parameters, and build open-source tools for working with real scanner data. Before MRI, I worked on X-ray fluorescence and diffraction imaging at Stanford/SLAC and SACLA.
Open MR reconstruction in PyTorch
I lead the development of MRpro: a shared framework for model-based reconstruction, learned methods and quantitative MRI, with common data structures and differentiable operators. The MRpro team received second prize for relaxometry in the 2024 ISMRM QMRI Study Group Challenge.
A faster-moving continuation of MRpro
My continuation of MRpro for developing new reconstruction and processing ideas. It keeps a familiar interface while extending the research toolbox.
Physics-Informed Machine Learning for Robust Quantitative MRI
Research at PTB and TU Berlin combining MR acquisition and signal models with learned priors. My work spans differentiable optimization, quantitative parameter estimation, uncertainty and low-field imaging.
Learning with the physics of quantitative MRI
End-to-end quantitative reconstruction through unrolled optimization, differentiable solvers and learned regularization. The signal and acquisition models are part of the network.
Ultra-low-field MRI enhancement
Diverse augmentations and auxiliary learning tasks for enhancing 64 mT brain scans with limited paired training data. My ULF-EnC submission placed fourth on the final MICCAI 2025 test leaderboard.
Transmit-field mapping from localizers
Learning spatial transmit-field maps from localizer images, with training and evaluation code for whole-brain B₁⁺ prediction at ultra-high field.
Semi-supervised T₁, B₀ and B₁ mapping
Spatially regularized parameter estimation from WASABIT! data, combining synthetic pretraining with supervised and self-supervised fine-tuning. Includes trained models and a Pulseq sequence. Presented orally at ISMRM 2023 and recognized with a Magna Cum Laude award.
Cardiac reconstruction without explicit sensitivity maps
Learned, unrolled reconstruction for the CMRxRecon 2023 challenge, developed with Andreas Kofler. NoSENSE works without a separate coil-sensitivity-map estimation step.
Data-adapted regularization for reconstruction
Collaborative work on learning spatially and temporally varying regularization parameters, pairing neural networks with variational reconstruction and algorithm unrolling.
Raw-data conversion, directly in the browser
Convert Siemens Twix measurements to ISMRMRD, inspect and edit headers, and merge metadata. Conversion runs locally in the browser; measurement data is not uploaded to a server.
Write and run MRI sequences in the browser
A browser workspace for PyPulseq code using Pyodide. Run Python, inspect plots and download generated .seq files without a local Python setup.
Bring reconstruction to Siemens Open Recon
An Open Recon server using MRpro for scanner-side reconstruction, with configurable processing and a scanner interface.
Contributions to open MRI sequence design
Contributions and sequence-development work with the community PyPulseq package for vendor-neutral MRI pulse sequences in Python. It provides the foundation for my sequence experiments and browser-based editor.
N-dimensional operations for PyTorch
Convolution, padding, cropping, interpolation and adjoints with flexible dimension layouts, including complex tensors where supported.
Optimization as a network layer
A compact example of differentiating through nonlinear optimization and using regression inside a neural network. An early building block for work such as PINQI.
Differentiable MR signal simulation
Extended phase graph simulations for fast spin echo and magnetic resonance fingerprinting, implemented with batched PyTorch operations and autograd.
Practical building blocks for MRI research
A collection of PyTorch tools including adjoint grid sampling, separable convolutions, coil-sensitivity estimation and sliding-window views.
Incoherent diffraction imaging
Software for simulating, reconstructing and analyzing X-ray fluorescence data, developed alongside experimental imaging work at Stanford/SLAC and SACLA.
Incoherent Imaging with X-Ray Fluorescence
My physics master’s work at TU Berlin studied fluorescence-intensity correlations for X-ray imaging: simulations, detector alignment, photon counting and experiments at SACLA.
X-ray diffraction imaging
GPU-accelerated simulations of X-ray scattering from three-dimensional particles, comparing iterative reconstruction and in-flight holography.
Planning rural electrification
An interactive Leaflet map of population clusters and geodata for electrification planning in Tanzania, built with open-source tools and open data.
LaTeX in the browser
TeXlive compiled to JavaScript with Emscripten, based on the work of manuels. Write and compile LaTeX without installing a local TeX distribution.
Master stability function and cluster synchronization
A nonlinear-dynamics project analyzing the stability of cluster synchronization in complex networks.
A motorized camera dolly
An inexpensive Atmel AVR controller for a camera dolly.
ULF-EnC at MICCAI 2025, Springer, pp. 28–36
arXivNMR in Biomedicine
IEEE Transactions on Biomedical Engineering
Nature Communications
Magnetic Resonance in Medicine
IEEE Transactions on Biomedical Engineering
Measurement: Sensors
SPIE, X-Ray Free-Electron Lasers VII PC135360Z
IEEE Transactions on Computational Imaging
arXivQuantification of Biophysical Parameters in Medical Imaging
arXivSTACOM at MICCAI 2023, Springer, pp. 454–466
arXivISMRM, #0659, oral presentation
ISMRM, #1166, oral presentation, Magna Cum Laude award
SPring-8/SACLA Research Report 11(5), pp. 387–389
APS DAMOP, X11.005
APS DAMOP, S04.007
