Felix Lucka

welcome to my academic web site

Collection of MATLAB functions published

Over the years, I wrote a large number of all kinds of auxiliary MATLAB functions for my various projects in mathematical imaging. I recently decided to share them on GitHub as a toolbox and put some effort into commenting and standardizing them. Hope someone finds them useful, I will do the same with my collection of numerical optimization algorithms.

Deep Learning for cardiac MRI

Together with a team of researchers and clinicians from the Centre for Translational Cardiovascular Imaging, Great Ormond Street Hospital for Children we started an exciting project on using deep learning to improve cardiac imaging for children with congenital heart diseases. A first proof-of-concept study on using a U-Net-type CNN for real‐time reconstruction of dynamic, under-sampled cardiovascular MR just got published in Magnetic Resonance in Medicine. Big thanks to the team!

Numerical methods for acoustic holography

My colleague Brad Treeby from the Biomedical Ultrasound Group (BUG) at UCL got me involved in an interesting project that tries to find ways to numerically simulate the acoustic fields that ultrasound transducers generate in complex media. The idea is to find an equivalent interior source term that would reproduce field measurements taken in a homogeneous medium. The work let to a paper titled “Equivalent-Source Acoustic Holography for Projecting Measured Ultrasound Fields through Complex Media” which just got accepted for publication in IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control. Big thanks to Brad and the other co-authors!

Majorization-minimization from a hierarchical Bayesian perspective

A while ago, I got involved in an interesting collaboration with Alexandre Gramfort, Joseph Salmon & Yousra Bekhti from the Télécom ParisTech. We drew a connection between majorization-minimization methods for sparse non-convex regression and a particular kind of hierarchical Bayesian modeling. Furthermore, we demonstrate how it can be used to quantify the inherent uncertainty and ambiguity of ill-posed regression problems such as EEG/MEG source reconstruction. The paper can be found on arXiv, big thanks to all the co-authors! [Update: It has been published in Inverse Problems.]