Florian Knoll

The goal of my research is the development and application of machine learning methods to medical imaging, and their translation into clinical practice so that they can help patients on a day-to-day level. In particular, I am interested in data acquisition and image reconstruction methods that make magnetic resonance imaging faster, more robust against image artifacts, allow imaging of new anatomical or pathological processes, make image interpretation easier and more standardized by moving from qualitative image contrasts to quantitative biomarkers for disease processes, and increase its global availability and accessibility. I serve as the deputy editor of Magnetic Resonance in Medicine for articles from this type of research.

 I serve as the principal investigator on multiple projects funded by the NIH:

• Machine learning MR image reconstruction for accelerated musculoskeletal imaging. With Thomas Pock (Graz University of Technology): R01EB024532.
• Quantitative multi-compartment MR-fingerprinting. With Jakob Asslaender: R21EB027241.
• TR&D 1 of the Center for Advanced Imaging and Innovation (CAI²R). With Dan Sodickson: P41EB017183.

As a strong supporter of reproducible research in the field of imaging, I currently serve as the chair for the ISMRM reproducible research study group. I also initiated and serve as the scientific lead of the fastMRI data sharing initiative, and the associated image reconstruction challenge. In collaboration with Facebook Artificial Intelligence Research, we made available a dataset of raw k-space data for more than 1300 knee MRI scans and more than 7000 brain MRI scans.

Code to reproduce the results for some of my papers can be obtained from:

• https://github.com/FlorianKnoll (Matlab and Python)
• Model based DTI reconstruction (Matlab)
• AGILE (CUDA)
• IRGN-TGV (Matlab)
• gpuNUFFT (CUDA and Matlab)
• Second order Total Generalized Variation (TGV) constrained reconstruction (Matlab)