Sign in


A Computational Answer to the Soaring MRI demand

Photo credit – Shutterstock

The steady growing demand for MRI scans is leading to unsustainable financial, human and operational costs, mainly caused by the typically long exam times. MR-STAT is a fundamentally new paradigm in MRI, which delivers multiple quantitative MR parameters in a single scan of just 5 minutes. The technology can be applied to existing MRI scanners increasing overall capacity while examinations become faster, more standardized, and comfortable. Although current pilot implementations of MR-STAT deliver usable quantitative data with a 5 minute scan, the processing step takes about 1-3 hours of computing time, thereby hampering its clinical implementation. This project aims at dramatically reducing the computing times to within 1 minute. To achieve this, scientists from the UMC Utrecht, CWI, Philips and eScience Center will create a highly-efficient distributed GPU implementation that parallelizes the computations across a GPU cluster. We will develop an open-source library using low-level GPU computing languages to provide highly-optimized GPU kernels for the basic building blocks of the MR-STAT reconstruction. By defining a high-level interface in Python and Julia, we will ensure it will be easy for researchers and students from a broader community to use the MR-STAT toolbox. This computing infrastructure will be tested in the UMCU radiology department at the end of this project. The resulting increased patient throughput will allow to meet the challenges of increased demand for MRI scans in the future. Furthermore, this project will represent an important ground of scientific exploration for the application of large-scale numerical methods in real-world scenarios.

Participating organisations

University Medical Center Utrecht
Netherlands eScience Center
Life Sciences


Alessandro Sbrizzi
Lead Applicant
Universitair Medisch Centrum Utrecht
Pablo Lopez-Tarifa
Pablo Lopez-Tarifa
Programme Manager
Netherlands eScience Center
Tristan van Leeuwen
Centrum Wiskunde en Informatica
Nico van den Berg
UMC Utrecht
Ben van Werkhoven
Ben van Werkhoven
Lead RSE
Netherlands eScience Center
Stijn Heldens
Stijn Heldens

Related tools

Kernel Tuner


Kernel Tuner greatly simplifies the development of highly-optimized and auto-tuned CUDA, OpenCL, and C code, supporting many advanced use-cases and optimization strategies that speed up the auto-tuning process.

Updated 1 month ago
40 14



Lightning: Fast data processing using GPUs on distributed platforms

Updated 10 months ago