Improved metrology for quantitative MRI

The same patient's MR image can look different depending on the scanner it was taken on, so results from different hospitals cannot be compared directly. iMet-MRI laid the metrological foundation needed to turn MRI from a device that produces pictures into one that makes reliable measurements. SI-traceable phantom materials, standardised measurement procedures and uncertainty analyses were developed for all five measurands: T1, T2, fat fraction, iron content and diffusion. Before the project, only NIST could offer traceability anywhere in the world, and only for three of them. The methods were tested on scanners from the three major manufacturers, from 0.55 to 7 T, in a multi-site study across four countries. By the end of the project, the IEC had accepted that new standards are needed for quantitative MRI.

At a glance


Project number	20NRM05
Short name	iMet-MRI
Programme	EMPIR, 2020 call (Normative)
Duration	June 2021 – May 2024 (36 months)
Coordinator	Matt G. Hall, NPL (United Kingdom)
Chief stakeholder	International Electrotechnical Commission (IEC)

Why it was needed

Cancer, heart disease, dementia and stroke can all be detected and monitored with MRI. European health services run thousands of scanners from different manufacturers, of different ages and field strengths. Conventional MR images are qualitative: they show relative contrast between tissues, and that contrast varies greatly from one scanner to another. Images acquired at different sites therefore cannot be compared directly.

Quantitative MRI (qMRI) instead forms images by measuring physical quantities. This makes it possible to calibrate scanners and compare their performance directly. Lower data variability also means smaller cohorts in clinical studies. But it requires well-characterised test objects (phantoms) and measurement protocols. Existing standards either did not address quantitative measurements (IEC 62464-1) or exempted MRI from measurement requirements (IEC 60601-2-33).

Objectives

Develop traceable phantom materials and manufacturing methods for T₁, T₂, fat fraction, iron content and diffusion, together with reference methods that provide traceability to these materials.
Develop standardised measurement procedures and user guides, using experimental and digital methods, for data analysis and uncertainty quantification of qMRI parameter maps.
Demonstrate the reproducibility of the methods in a multi-site study with at least five sites.
Publish the methods and at least two good practice guides, and develop harmonisation guidelines for standards bodies and clinical trial organisers.
Work with the technical committees IEC TC62/SC62A and BSI CH/62/2 so that the project outputs feed into future standards.

Results

Traceable phantom materials

Materials and characterisation methods were developed for all five qMRI measurands. Solutions of each solute were prepared over ranges representative of human tissue, and the target concentrations were verified by NMR spectroscopy. T₁ and T₂ relaxation times were measured on a 60 MHz benchtop NMR system, traceable to the TÜBİTAK rubidium atomic clock. The fat phantoms used NIST PS1 benzoic acid reference material as an internal standard.

Material	Formula	Measurand
Nickel(II) chloride hexahydrate	NiCl₂·6H₂O	T₁
Manganese(II) chloride tetrahydrate	MnCl₂·4H₂O	T₂
Iron(III) chloride hexahydrate	FeCl₃·6H₂O	Iron content
Nitric acid	HNO₃	Iron content (solvent)
tert-Butanol	(CH₃)₃COH	Fat fraction
Ultra-pure water	H₂O	Solvent for all measurands except iron

The materials were placed in a phantom consisting of a watertight outer container and two 3D-printable inserts with orientable fiducial markers. Because diffusion imaging is more sensitive to susceptibility-induced distortion, a second holder with larger vials was designed. Ten phantom sets were produced and distributed to the consortium.

Measurement procedures and analysis

Standard acquisition procedures were written for all five measurands. For every measurand except iron content, several sequences were compared, from a best-possible protocol to ones that fit clinical scan times:

Measurand	Pulse sequence
T₁	Inversion recovery, variable flip-angle, MOLLI
T₂	Single echo, multi-echo
ADC	PGSE
Fat fraction	2-, 3- and 6-point Dixon
Iron content	Gradient echo

Analysis code that extracts the parameters from the images was written in Python, and an uncertainty analysis was carried out for each measurand. A Bloch-equation MR simulator covering the full sequence and readout, and a digital twin of the phantom, were also developed. The digital model provided a known ground truth for testing the analysis methods.

Multi-site study

The phantoms and procedures were tested on ten 1.5 T and 3 T scanners from Siemens, Philips and GE at clinical sites in the UK, Italy and Ireland, with additional scans at 7 T in the UK and Switzerland and at 0.55 T in Switzerland. Key findings:

T₁ values above 600 ms do not depend on field strength, but shorter T₁ values differ significantly. T₁ values from MRI agree well with NMR.
T₂ measurements differ significantly between field strengths and agree poorly with NMR values.
The relationship between T₂* and iron concentration is similar across scanners at the same field strength.
Fat fraction agrees with nominal values on all scanners at 3 T; at 1.5 T, fat contents below 40% tend to be underestimated.
ADC values from basic clinical sequences agree well with nominal values.

The study also showed the need to balance accuracy against scan time, and to record information that DICOM headers do not always contain, such as coil loading.

Guides and standards

All results were combined in a Good Practice Guide describing test objects and test procedures for assessing scanner performance in qMRI, published as a draft for consultation. Two free e-learning courses on uncertainty evaluation for fat fraction and relaxation parameters were published on NPL’s training portal. After the results were presented to the IEC SC62 MT40 working group, the IEC accepted that new standards for quantitative MRI are needed, and the qMRI standards gap was included in the current redrafting of IEC 60601-1.

Impact

Before the project, NIST was the only national metrology institute able to provide traceable characterisation of MR-measurable quantities, and only for three of the five measurands. iMet-MRI defined routes to traceability for all five, and traceable reference materials for MRI are now available in Europe. The phantom developed is lower in cost than those currently on the market. Results were presented to RSNA QIBA, EIBALL and ISMRM, a symposium was held at the ISMRM meeting, and three stakeholder workshops took place. Following the interest the project generated, NPL committed more than £2M to a new laboratory.

Partners

National metrology institutes: NPL (United Kingdom, coordinator), TÜBİTAK UME (Türkiye), LGC (United Kingdom), INRIM (Italy), IMBiH (Bosnia and Herzegovina)
Other partners: AOUC (Italy), BHSCT (United Kingdom), UCL (United Kingdom), UHBW (United Kingdom), Verlab (Bosnia and Herzegovina)

Publications

Hall, M. G. et al. (2023). The perfect diagnostic imaging machine and what it means for quantitative MRI reproducibility. IPEM-Translation, 1, 100019. doi:10.1016/j.ipemt.2023.100019
Tofts, P. S. (2024). Realistic phantoms for quantitative MR: the need and a proposal (the B1 sleeve). IPEM-Translation, 9, 100026. doi:10.1016/j.ipemt.2024.100026