Quantitative MR-based imaging of physical biomarkers

At a glance

Project number	18HLT05
Short name	QUIERO
Programme	EMPIR, 2018 call (Health)
Duration	June 2019 – November 2022 (42 months)
Coordinator	Luca Zilberti, INRIM (Italy)

Why it was needed

With more than 30 million scans a year in Europe, MRI is one of the most important imaging tools in clinical practice. Yet standard MRI results are qualitative: they depend on a specialist’s visual interpretation, are hard to compare across time and sites, and give no direct information about the nature of a pathology.

Electrical Properties Tomography (EPT) computes the electrical conductivity and permittivity of tissues from MR data. Magnetic Resonance Fingerprinting (MRF) measures several tissue parameters, such as T₁ and T₂, in a single fast acquisition. Both promise to remove inter-observer variability, reduce the need for invasive procedures such as biopsies, and support early disease detection. But when the project began, their reliability had not been characterised systematically, and clinicians needed to know how far these results could be trusted.

Objectives

1. Develop, improve and implement numerical algorithms for EPT and MRF, and characterise their performance.
2. Make EPT and MRF usable in “high impact” clinical conditions such as cerebral ischaemia and heart disease, including suppression of motion artefacts in the heart.
3. Develop heterogeneous phantoms mimicking the electrical and relaxation properties of human tissue, and evaluate accuracy in controlled experiments (target uncertainty 20% for EPT and 10% for MRF).
4. Determine inter- and intra-subject physiological variability and minimum detection thresholds under real clinical conditions, and develop computer-aided diagnosis approaches.
5. Facilitate uptake by accredited laboratories, MRI manufacturers, technical committees and hospitals.

Results

Open-source algorithms

EPT methods were collected in EPTlib, an open-source C++ library developed by INRIM. EPTlib provided the first public implementations of EPT algorithms, including Helmholtz-EPT, Convection-Reaction EPT and gradient-EPT, with a user guide and b1map-sim, a tool that generates synthetic MR data. For MRF, alongside classical dictionary-based methods, the project developed a new technique that estimates tissue parameters directly from raw MR data (k-space) and a Bayesian approach that returns results together with their uncertainty.

The algorithms were compared in virtual experiments based on realistic electromagnetic and Bloch simulations, and the best operating conditions were identified. As a side result, EPT was shown to be potentially important for estimating subject-specific radiofrequency exposure (SAR) during MRI.

Adapting to brain and heart

For the brain, a population of about 150 healthy and 200 pathological virtual human models was built. A post-processing approach based on agreement between two EPT methods allowed white-matter lesions to be detected in a large share of the virtual models. To use MRF in the heart, a simulation environment modelling heartbeat and respiration was released as open source and a motion-correction method was developed. The optimised cardiac MRF protocol reached relative errors of (5.6 ± 3.4)% for T₁ and (8.3 ± 8.4)% for T₂. On myocardium models with fibrosis, MRF-based classification reached an AUC of 0.992 (a common threshold for clinically useful classification is 0.8).

Tissue-mimicking phantoms

Materials mimicking white and grey matter were prepared according to a protocol agreed by INRIM, PTB, TUD and TÜBİTAK, using GdCl₃ for T₁, NaCl for conductivity, PVP and glycerol for permittivity, and NaN₃ against fungal growth. Several generations of phantoms were produced, from simple cylinders to brain-shaped phantoms cast in moulds made from 3D-printed structures. TU Dresden designed a new low-cost soft-matter 3D printer and printed a triphasic brain phantom with grey matter, white matter and cerebrospinal fluid.

Relaxation times of the materials were measured on MRI scanners and NMR spectrometers; T₁ and T₂ measurements made by TÜBİTAK on a 1.4 T NMR spectrometer agreed with the expected values. Relaxation times monitored by benchtop NMR changed by less than 2% over more than three years, and dielectric properties stayed within measurement uncertainty for over two years.

Clinical studies

• Brain: 100 scans were acquired at 1.5 T and 7 T from 53 patients, mostly children with white-matter diseases, and 32 healthy controls. Using a new EPT technique that derives conductivity from MRF-based T₁ maps, an age-corrected increase of +34.3 mS/m in white-matter conductivity was found to indicate disease at a 95% confidence level.
• Heart: 15 patients with hypertrophic cardiomyopathy, muscular dystrophy or severe aortic stenosis and 4 healthy volunteers were scanned. Myocardial T₁ above 1478 ms or below 1062 ms, and T₂ above 52 ms or below 16 ms, were found to reliably indicate these conditions.
• Repeatability: A study with 12 healthy volunteers at eight sites showed that 3D MRF measurements in grey and white matter are highly repeatable and reproducible; the data are openly available on Zenodo.

Impact

During the project, 13 open-access papers were published and 45 presentations (six invited) were given at scientific meetings. The work describing EPTlib received a Magna Cum Laude award at ISMRM 2021, and the library was recognised by the ISMRM Electro-Magnetic Tissue Properties Study Group as the first EPT package available on the internet. The consortium worked with GE, Philips and Siemens, and the expertise gained in characterising the dielectric and relaxation properties of phantoms enables INRIM, PTB and TÜBİTAK to offer measurement services in this field. EPT-based subject-specific SAR assessment was shared with the committee responsible for the MRI safety standard IEC 60601-2-33, and EIBALL, QIBA and ISMRM were kept informed throughout.

Partners

• National metrology institutes: INRIM (Italy, coordinator), PTB (Germany), LNE (France), LGC (United Kingdom), TÜBİTAK UME (Türkiye), IMBiH (Bosnia and Herzegovina)
• Other partners: Charité – Universitätsmedizin Berlin (Germany), Fondazione Stella Maris (Italy), TU Dresden (Germany), University of Turin (Italy), University of Ljubljana (Slovenia)

Selected publications

• Arduino, A. (2021). EPTlib: An open-source extensible collection of electric properties tomography techniques. Applied Sciences. doi:10.3390/app11073237
• Buonincontri, G. et al. (2021). Three dimensional MRF obtains highly repeatable and reproducible multi-parametric estimations in the healthy human brain at 1.5T and 3T. NeuroImage. doi:10.1016/j.neuroimage.2020.117573
• Metzner, S. et al. (2021). Bayesian uncertainty quantification for magnetic resonance fingerprinting. Physics in Medicine and Biology. doi:10.1088/1361-6560/abeae7
• Kilian, D. et al. (2022). 3D extrusion printing of biphasic anthropomorphic brain phantoms mimicking MR relaxation times based on alginate-agarose-carrageenan blends. ACS Applied Materials & Interfaces. doi:10.1021/acsami.2c12872
• Hufnagel, S. et al. (2022). 3D model-based super-resolution motion-corrected cardiac T1 mapping. Physics in Medicine and Biology. doi:10.1088/1361-6560/ac9c40
• Martinez, J. et al. (2023). Evaluation and correction of B1+-based brain subject-specific SAR maps using electrical properties tomography. IEEE Journal of Electromagnetics, RF, and Microwaves in Medicine and Biology. doi:10.1109/JERM.2023.3236153
• Gatefait, C. et al. (2023). Optimisation of data acquisition towards continuous cardiac magnetic resonance fingerprinting applications. Physica Medica. doi:10.1016/j.ejmp.2022.102514
• Arduino, A. et al. (2023). Repeatability and reproducibility uncertainty in magnetic resonance-based electric properties tomography of a homogeneous phantom. Tomography. doi:10.3390/tomography9010034

Full list: EURAMET publications repository

Links

• EURAMET project page
• Project website
• EPTlib · GitHub
• Project video (YouTube)
• MRF repeatability dataset (Zenodo)