EP4256360A1

EP4256360A1 - Characterisation of distortions in magnetic resonance imaging

Info

Publication number: EP4256360A1
Application number: EP21840079.4A
Authority: EP
Inventors: Thomas PUISEUX; Anou SEWONU; Ramiro MORENO
Original assignee: Alara Expertise; Spin Up
Current assignee: Alara Expertise; Spin Up
Priority date: 2020-12-02
Filing date: 2021-12-01
Publication date: 2023-10-11
Also published as: FR3116905A1; WO2022117957A1; FR3116905B1; US20240012081A1

Abstract

The present invention relates to a method for calibrating a piece of MRI tomography equipment consisting in constructing a pair of phantoms comprising a physical phantom and a digital twin phantom, in determining the virtual image of the digital phantom on the basis of the characteristics of the piece of MRI equipment to be tested, in carrying out an MRI sequence with the physical phantom and in verifying the virtual and real images, characterized in that said digital phantom is produced by solving Bloch equations applied to the characteristics of said physical phantom as a function of the characteristics of said reference sequence.

Description

DESCRIPTION

TITLE: CHARACTERIZATION OF DISTORTIONS IN IMAGING BY

MAGNETIC RESONANCE

FIELD OF THE INVENTION

The present invention relates to the field of metrology and calibration of Magnetic Resonance Imaging (MRI) equipment.

MRI is a non-invasive, non-ionizing imaging technique that produces two- or three-dimensional tomographic images of the interior of the human body. Thanks to a wide variety of existing acquisition sequences, MRI offers the possibility of evaluating both the anatomical and functional characteristics of organs, with high contrast levels between tissues. It is very useful when a very fine analysis is necessary and certain lesions are not visible on standard x-rays, ultrasound or CT.

The principle of 1 MRI is based on the magnetic properties of hydrogen atoms, contained, to varying degrees, in all tissues of the human body.

By using strong magnets, the protons of the hydrogen atoms in the body are stimulated simultaneously, which has the effect of aligning their magnetic spin moments in the direction of the produced magnetic field. The sum of the magnetic moments contained in sample of spins called isochromates corresponds to the magnetization vector. Once this first step has been completed, the atoms are brought into resonance by subjecting them to excitation by a magnetic field (radiofrequency). When the stimulation stops, the atoms release energy, the intensity of which is measured and analyzed. Not all tissues contain the same amount of hydrogen atoms and the level of energy returned will differ depending on the tissue composition. The analysis of these data by computers makes it possible to reconstitute images in 2 and 3 dimensions and in the three planes of space.

The MRI machine contains gradient coils, placed on three axes of a 3D plane. The electric current which passes through these gradient coils produces a local distortion of the main magnetic field. This distortion is used for spatial encoding of images. It allows, on each axis of a 3D plane (x, y and z), to locate a particular point. For this, three gradient coils are used; one to reconstitute each of the three planes of the 3D MRI images.

To perform an MRI, the patient is placed in a magnet which emits a permanent magnetic field Bo typically of 1.5T or 3T. This stable environment to hydrogen nuclei scattered throughout the human body, to be in a state of balance. In order to initiate the phenomenon of resonance, the contribution of energy is carried out via the emission of another specific magnetic field called Bi which is a rotating field at the Larmor frequency of the hydrogen nuclei, i.e. a radiofrequency wave. 128 MHz for 3 Tesla magnets, and 64 MHz for 1.5 T magnets. All of the patient's hydrogen nuclei absorb this energy and therefore find themselves in an excited state. When the application of this magnetic field Bi is stopped, all the hydrogen nuclei dissipate the accumulated energy and gradually return to their state of equilibrium. It is at this moment that the acquisition of the image is carried out.

The return to equilibrium of the hydrogen nuclei (also called relaxation) is not instantaneous: it therefore requires time. This relaxation phenomenon is characterized by two time constants - Ti and T2 - reflecting this latency of the return to equilibrium. Then the hydrogen nuclei return more or less quickly in their state of equilibrium depending on the environment (or tissue) in which they find themselves. This therefore results in a relative decrease or increase in the relaxation times Ti and T2.

Excitation is done by applying a sequence of repeated pulsed RF signals, with echo times TE and repetition times TR forming a reference pattern.

The relaxation signal is picked up by the antennas of the MRI imager which deliver a digital signal which is processed to transform it into an image. The reconstruction of MRI images is based on a multidimensional Fourier transform.

The many instruments and the different processing steps required to form an image give rise to multiple sources of errors that are generally difficult to characterize. Indeed, the magnetic characteristics of the object (relaxation time, proton density), the instrumentation (permanent magnet, gradient coils, radio-frequency antennas), and the acquisition and reconstruction parameters (time of echo, spatio-temporal resolution, sampling frequency) are all elements that affect the quality of the MRI signal and can alter the image. In practice, the distortions result in variations of position and intensity of the voxels on the images compared to their real position.

However, the quality of the MRI image is essential for the relevance of the diagnosis and the use of this information for a therapy, in particular for the diagnosis and the precise localization of a tumor with a view to a therapeutic treatment and for the adaptation of the margins in the context of ablation or radiotherapy. Identify the elements responsible for Image distortions then represent a means of optimizing the MRI measurement.

The present invention relates to the calibration of MRI tomography equipment, in order to ensure the quality of the MRI images and to overcome the degradations observed either due to malfunction of the equipment itself (machine), or limitations of the sequence or the process of reconstructing the image.

PRIOR ART

It is known from the prior art to use a phantom making it possible to verify the image obtained and to compare it with a reference image obtained with perfectly calibrated MRI equipment. This solution is for example described in the brief https: // www .utc.fr/ tsibh/public/3abih/12 / stage/litov/ index . html 1.

Also known is the European Standard NF EN 62464-1 “Magnetic resonance apparatus for medical imaging Part 1 “determination of the main image quality parameters” defining protocols for verifying the parameters likely to influence the quality of MRI images.

Scientific publications relating to flow simulation algorithms are also known.

The publication "FORTIN ALEXANDRE ET AL: "Flow MRI simulation in complex 3D geometries: Application to the cerebral venous network", MAGNETIC RESONANCE IN MEDICINE, vol. 80, no. 4, 5 February 2018 (2018-02-05), pages 1655-1665, XP055839967" describes an example of the development of a computational tool for simulating the movement of spin flows in MRI for the study of flow artefacts in It mentions the comparison of simulations of simple flow models with real experiments on a physical flow phantom for the verification of the relevance of the digital simulation tool .

The aim of this work presented in the article is to provide a versatile tool for the simulation of MRI measurements on complex and arbitrary fluids and flow movements. The method described requires the recording of a limiting quantity of data (the trajectories of all the spins circulating over time), which makes the simulations very excessively greedy in computing resources.

Another article, MARSHALL IAN: "Computational simulations and experimental studies of 3D phase-contrast imaging of fluid flow in carotid bifurcation geometries", JOURNAL OF MAGNETIC RESONANCE IMAGING, vol. 31, no. 4, March 29, 2010 (2010-03-29), pages 928-934, XP055839940, also relates to the general field of MRI digital simulation for predicting the appearance of physiological flows.

We are also familiar with the applicant's patent application WO2019211556A1 which concerns the characterization of the degree of confidence in the MRI flow measurements by flow imaging by Magnetic Resonance Imaging (MRI) called "4D flow", by flow imaging. 'a ghost. It consists of acquiring MRI flow data by phase contrast imaging from a physical phantom fed with a reference fluidic feeding sequence, to record a second file of experimentally measured digital data, then performing steps of resetting, correcting and formatting the data, steps necessary for spatio-temporal compatibility, before carrying out a comparison of said digital data files to calculate at least one error indicator representative of the differences between said digital files.

DRAWBACKS OF THE PRIOR ART The solutions of the prior art do not concern the calibration of physical equipment intended to acquire 4D MRI images, but simulation tools. No metrological solution for calibrating MRI equipment is totally satisfactory because they mostly depend on the quality of the control image of a phantom produced on reference equipment, which itself may have multiple imperfections. Moreover, by acquiring an MRI image of such a phantom, it is impossible to determine the origin of any deterioration in quality.

DISCLOSURE OF THE INVENTION

The present invention aims to overcome the drawbacks of the state of the art by a solution for calibrating an MRI tomography equipment consisting, according to the most general meaning of the invention, in constructing a pair of phantoms comprising a physical phantom and a digital counterpart phantom, determining the virtual image of the digital phantom from the characteristics of the MRI equipment to be tested, performing an MRI sequence with the physical phantom and verifying the virtual and real image, said digital phantom is performed by solving the Bloch equations applied to the characteristics of said physical phantom as a function of the characteristics of said reference sequence.

Advantageously, a calculation of the mapping of the virtual isochromats of said digital phantom is carried out and a comparison of this mapping of the virtual isochromats with the mapping of the isochromats obtained by an MRI tomography of said physical phantom to determine the differences. Optimally, the processing of resolution of the Bloch equations is carried out “on the fly” without storing the data relating to the trajectories of the particles, in the case of a ghost of flux or animated by a movement. This originality avoids the massive storage of the spatiotemporal trajectories of the spins, unlocking the limits in terms of computing resources.

According to a variant, said step of calculating a map of the isochromats by the digital model consists of:

- model isochromats with Lagrangian particles deposited within any mesh, using the velocity from an MFN to update each particle position "on the fly"

- solve Bloch's equations independently for each particle.

GENERAL PRESENTATION OF THE INVENTION

The invention relates to a method and a system making it possible to verify the proper functioning of MRI equipment and to identify the sources of errors.

By identifying the sources of error, we mean to differentiate between the following types of distortions:

- object: image artefacts induced by the characteristics of the object to be imaged such as physiological movements (movement of the patient, breathing, heartbeat) or its material composition (presence of a metallic device).

- machine: imperfections and limitations of the instrumentation altering the expected characteristics of the acquisition system. Among these imperfections, we can cite the homogeneity of the permanent magnetic field, the linearity of the magnetic fields produced by the gradient coils, the effects of eddy currents, or even the spatial sensitivity in reception and transmission of the signal by the radio frequency antennas.

- sequence: distortions due to non-optimal operator-dependent acquisition parameters affecting image quality (echo time, encoding speed, sampling frequency) . Indeed, device manufacturers generally offer clinical protocols provided with pre-established sequence parameters depending on the type of imaging to be performed. These sequences can be adapted directly on site, or by the application engineer during software updates. It is therefore possible to see poorly calibrated or obsolete protocols still used in practice.

The limitations intrinsic to the sequence (imperfect character of the "spoilers", speed encoding) testifying to the ability of an acquisition to measure physical reality are also limitations of the sequence. image reconstruction: artifacts related to truncation effects and discrete sampling of Fourier space.

THE IDENTIFICATION OF ERROR SOURCES USES A SYSTEM CONSISTING OF:

A physical phantom (morphological or functional) whose image obtained by the MRI equipment to be characterized is intended for comparison with an image calculated digitally from:

- operating parameters of the MRI equipment to be characterized

- the virtual configuration of the physical phantom,

- and possibly disturbing parameters

A computer program for calculating a digital twin of the aforementioned physical phantom, based on the principle of numerical resolution of Bloch's equations. These equations govern the movement of nuclear magnetization subjected to a magnetic field and are at the center of the image formation process. This computer application is referred to as “synthetic MRI”.

The method proposed by the invention certainly uses the modeling of Bloch's equations, but does not record the trajectories of the spins. The method that is the subject of the invention progressively resolves (“on the fly”) the effect of these displacements in terms of magnetization, which eliminates the need to record colossal volumes of data. This method, unique in its kind, also makes it possible to solve any complex flow in the field of MRI, without being limited by computer resources.

MRI images of a physical phantom whose characteristics are perfectly controlled and known are acquired with an MRI sequence to be characterized.

The evolution of the magnetization within the same virtualized phantom subjected to this same sequence is resolved numerically with the computer program, thanks to the synthetic MRI program. The interest of the simulation here is to be able to reconstruct a virtual image corresponding to a given MRI sequence, but exempt from any error specific to the experimental measurement.

The comparison between the image acquired experimentally (experimental image) and the virtual image obtained by synthetic MRI makes it possible to highlight any imperfections in the instrumentation system (machine malfunctions).

The use of an experimental phantom whose anatomical and/or functional characteristics are known makes it possible to eliminate the image distortions induced by the above-mentioned characteristics of the object.

In order to identify the errors linked to the sequence and to the acquisition parameters, a third image called “reference image” is designed from the geometry known a priori of the MRI phantom (virtual phantom) and of the sequence. To ensure spatial correspondence with the MRI image, the geometry of the virtual phantom is resampled on a voxelized image, of the same size and spatial resolution as the acquisition matrix. In the case of a Cartesian filling of k-space, the effects of truncation and sampling of Fourier space are also taken into account by applying a spatial filtering as defined in (Haacke, ME, RW Brown, MR Thompson and R. Venkatesan Magnetic Resonance Imaging: Physical Principles and Sequence Design. Wiley-Liss. Haacke, 1999) during the virtual phantom resampling process. The errors associated with the reconstruction of the image can be estimated by comparing the reference image with and without the application of this filtering. In the case of a non-Cartesian filling of the k-space, these effects are taken into account by applying a Fourier transform to the resampled image in order to design a reference k-space.

The residual distortions observed by comparing the reference image to the virtual image obtained by simulated MRI are then the consequence of errors induced by the sequence.

DESCRIPTION OF FIGURES

Other characteristics and advantages of the invention will become apparent on reading the following detailed embodiment examples, with reference to the appended figures which represent respectively:

[Fig. 1] Figure 1 shows a perspective view of an example of a 4D flow MRI phantom for the implementation of the invention designed by 3D printing, and used to evaluate the 4D flow MRI sequences. b) .

[Fig. 2] Figure 2 shows a sectional view of the MRI phantom along the median plane of the XZ axis

[Fig. 3] Figure 3 shows a schematic view of the test bench [Fig. 4] Figure 4 shows a sectional view (see Fig. 1) of the velocity field norm obtained by MRI simulation (uSIRM), at peak systolic

[Fig. 5] Figure 5 shows a sectional view of the velocity field norm obtained by MFN simulation (uMFN), at the systolic peak.

[Fig. 6] Figure 6 shows a view of the error defined as | | uSMRI - uvox | | where uvox is the resampled MFN field on the voxelized grid.

[Fig. 7] Figure 7 shows a volume view of the regions of greatest MFN acceleration at the systolic peak.

[Fig. 8] Figure 8 shows a view of the corresponding MFN-SIRM error.

[Fig. 9] Figure 9 shows a view of the calculated error between MRI-MFN at peak systolic.

[Fig. 10] Figure 10 shows a view of the calculated error between SIRM-MFN at peak systolic.

[Fig. 11] Figure 11 shows a schematic view of the sequence error characterization process.

[Fig.12] Figure 12 shows a schematic view of the machine malfunction characterization process.

The invention relates to the field of metrology in general, by which a reference phantom, imaged by magnetic resonance (MRI), is compared with a simulation of a virtual MRI device characterized by the real MRI sequence played by the system of imaging during acquisition. The method then makes it possible to determine the origin of the deviations observed between reference and simulation. It may be a question of a bad design of the MRI sequence, or in the opposite case, of a so-called “machine” problem, relating rather to an electronic, mechanical or electromagnetic response failure. The aim of calibrating equipment is to compensate for the gain disparities of detectors based on avalanche photodiodes (PDA) which generate inter-channel differences in particular, to compensate for heterogeneities in the magnetic field, to compensate for rise time defects switched gradients and more generally to verify the origin of the degradations observed either due to malfunction of the equipment itself (machine), or limitations of the sequence or of the image reconstruction process.

It is not a question, contrary to the documents of the prior art, of developing a computer tool for simulating the movement of spin flows in MRI for the study of flow artefacts in angiography for modeling, but of the adjustment of functional exploration equipment.

DESCRIPTION OF AN EXAMPLE OF REALIZATION

FIG. 1 represents an example of a geometric configuration of a flow phantom (1), having the general shape of a U with a curved zone (11) of constant circular section extended on either side by two branches (12 , 13) . These two branches (12, 13) are connected by a transverse duct (14) with a hemispherical inlet volume (15) simulating an aneurysm. The branches of the U (12, 13) have an input connector (5) and an output connector (4).

The phantom (1) is connected to a rigid inlet tube (2) for the supply of a pulsed fluid coming from a pump (6) and by a rigid outlet tube (16). The length of these tubes (2, 16) is typically 80 centimeters, with a section of 26 millimeters. They are connected to the pump (6) by flexible conduits (17, 20) via connectors. The supply tube (2) comprises a "honeycomb" type flow stabilizer (10) which has the effect of laminating the flow penetrating into the phantom (1) by reducing its whirling motion. It also includes a drain valve and a valve. A flow meter (7) is placed in series to control the actual instantaneous flow supplying the phantom (1). The phantom has openings where pressure sensors are connected. The shape is chosen to generate a complex and realistic flow, such as that seen in the large circulation (aorta-heart) of the cardiovascular system.

The inside diameter of the phantom (1) is 26 mm and was designed with a radius of curvature of 50 mm to mimic the blood flows of the aortic arch.

A collateral branch (14) has been established by analogy with the collateral arteries. The size of this collateral branch was designed to reproduce the duplication observed in vivo between the aorta and the arteries of the supra-aortic trunk, celiac, renal and iliac arteries.

Finally, the protrusion (15) attached to the intersection between the collateral branch (14) and the main branch (13) simulates the blood flows in the aortic aneurysms, but also makes it possible to appreciate the vortices present in the cardiac cavities.

The shape of the phantom is first made digitally with computer-aided design software. The CAD file then constitutes the digital basis to prepare on the one hand a digital hemodynamic phantom and on the other hand the physical phantom, by a 3D printing technique based on stereolithography.

The physical phantom is placed in the MRI equipment, in place of the patient, after having been placed in a pocket filled with a gel. It is powered by a circuit comprising a pump (6) and a flow meter (7) as well as a buffer tank (8). The pump (6) is computer controlled to provide reproducible waveforms. The fluid has hydrodynamic characteristics similar to those of blood in the large circulation: viscosity = 4 cPoi and density = 1020 kg/m3 and the longitudinal and transverse relaxation times known at 1.5T (Tl = 0.85 and T2 = 0.17).

The pump delivers a pulsed flow which is controlled at all times by the flow meter (7). The pressure sensors make it possible to validate the hydrostatic pressure maps reconstructed from the speed maps.

From a clinical point of view, it is particularly interesting to characterize the measurement errors for 4D flow MRI sequences in order to better quantify the reliability and precision of the hemodynamic markers reconstructed using this technique. This methodology applied to 4D flow MRI therefore makes it possible to assess the ability of a clinical protocol to quantify flow.

To develop this example, it is necessary to consider the following two pre-existing elements:

An experimental test bench provided with an MRI phantom (1) in which a physiological pulsatile flow is known thanks to the numerical simulation of the Navier-Stokes equations (Computational fluid mechanics - CFD). Several comparisons with idealized MRI measurements made it possible to validate the simulations and the flow within the MRI phantom (1) . The steps and results of this validation are detailed in (Puiseux T, Sewonu A, Meyrignac O, et al. Reconciling PC-MRI and CFD: an in-vitro study. NMR in Biomedicine. 2019). This numerical flow is hereinafter referred to as “reference flow”.

A numerical simulation method of the magnetization process taking place in an MRI during a 4D Flow measurement, developed in the YALES2BIO software (https://imag.umontpellier.fr/~yales2bio/index.html), and of which a proof of concept, documentation and several validation cases are available. The software includes a method of numerical resolution of the equations which govern the movement of nuclear magnetization subjected to a magnetic field (Bloch's equations). This program is hereinafter referred to as “synthetic MRI”.

IMPLEMENTATION OF THE INVENTION

The implementation of the invention breaks down into two stages:

- the first part the errors specific to the sequence and to the image reconstructor are identified (FIGS. 4 to 6).

- the second part is devoted to the identification of machine malfunctions (Figures 9 and 10).

IDENTIFICATION OF ERRORS RELATED TO THE SEQUENCE

This first step consists in simulating said 4D flow MRI sequence from the reference flow in order to reconstruct the image and thus the velocity field which should be obtained by an ideal acquisition (without experimental errors).

To simulate the acquisition of a 4D flow MRI sequence, the volume of the MRI phantom (1) is discretized into a digital mesh consisting of tetrahedral elements. The Bloch equations are solved numerically on Lagrangian particles deposited periodically at each repetition time within the fluid domain, and simultaneously moved following the resolution of the Navier-Stokes equations. The magnetic field which appears in the Bloch equations is represented by the MRI sequence in question and updated at each iteration. The MRI signals resulting from the relaxation properties of the particles encoded in space and in velocity are collected sequentially during the application of the reading gradients, and stored in a Fourier space. The periodic pulsed flow imposed by the pump is split into several temporal phases, and a virtual image is reconstructed by 3D inverse Fourier transform to each of these phases. Each image reconstructed by the MRI simulation is finally compared with the reference image corresponding to the same phase.

To construct a reference image compatible with an image obtained by MRI, the reference flow is degraded in space and time. In order to reproduce the effect by means of the sequential filling of the space k, the solutions obtained by MFN over time are first phase-averaged over a period of 40 time periods.

The phase-averaged digital solution is then resampled on a Cartesian grid of the same size and resolution as the simulated image. To do this, to each node of the mesh is applied a spatial filtering which takes into account the effects of truncation and discrete sampling of the Fourier space, thus making it possible to reproduce the errors related to the process of image reconstruction.

The differences observed between the velocity field reconstructed by MRI simulation and the reference flow resampled on the voxelized grid (see Figure 4) then correspond to distortions purely caused by the limitations of the MRI sequence and its acquisition parameters. . For this case study, we observe a maximum error approximately equal to 20% of the encoding speed.

It is sometimes possible to deepen the level of characterization of these errors by a finer analysis of the characteristics of the sequence.

Figures 4 to 6 represent the norm of the velocity field obtained at peak systolic by MRI simulation (uSIRM) (Figure 4) and by MFN (uMFN) (Figure 5) as well as the error defined as | | uSMRI - uvox | | where uvox corresponds to the MFN field and resampled on the voxelized grid (Figure 6).

Figure 7 represents the spatial structures of greatest acceleration in the reference flow. Figure 8 shows the largest speed errors in the synthetic MRI at the corresponding time. The similarities observed between these two quantities suggest the presence of an acceleration artifact. Indeed, an assumption inherent in 4D flow MRI sequences is that there is a linear relationship between the phase and position of a nuclear spin. In other words, effects induced by spin acceleration are neglected in 4D flow sequences.

This first step makes it possible to quantify the level of error introduced by certain approximations of the sequence. By characterizing the image artifacts observed, this methodology can also be extended to the optimization of MRI sequences.

IDENTIFICATION OF MACHINE MALFUNCTIONS

Since the reference flow is similar to the experimental flow within the MRI phantom, the comparison between the experimentally acquired and simulated images using the same 4D flow MRI sequence allows the identification of possible malfunctions of the instrumentation system.

The experimentally reconstructed images are compared with the images obtained by MRI simulation. The differences observed between the two speed fields testify to machine errors. The velocities reconstructed by experimental and synthetic MRI are compared with the reference image in Figures 9 and 10. Similar error structures can be distinguished, showing that the experimental MRI reproduces well the acceleration artefact discussed in Figure 9 and 10. previous part. One can also note a systematically higher level of error of the experimental MRI, suggesting the presence of distortions due to the instrumentation. The evaluation of the operation of the instrumentation system does not require acquiring a complex sequence. Indeed, simple MRI sequences composed of magnetic pulses basic information is theoretically sufficient to warn of possible malfunctions. For example, this may allow different magnetic field gradient coils to be tested separately.

Claims

1 . Method for calibrating MRI tomography equipment consisting in constructing a pair of phantoms comprising a physical phantom and a digital twin phantom, in determining the virtual image of the digital phantom from the characteristics of the MRI equipment to be tested, in performing an MRI sequence with the physical phantom and verifying the virtual and real image, characterized in that said digital phantom is performed by on-the-fly resolution, without recording of spin trajectories, of Bloch's equations applied to the characteristics of said physical phantom according to the characteristics of said reference sequence.

2 . Method for calibrating an MRI tomography equipment according to claim 1 characterized in that a calculation of the mapping of the virtual isochromats of said digital phantom is carried out and a comparison of this mapping of the virtual isochromats with the mapping of the isochromats obtained by an MRI tomography of said physical phantom to determine the differences.

3 . Method for calibrating 4D flow tomography MRI equipment according to Claim 1, characterized in that the said step of calculating a map of the isochromats by the digital model consists of:

- model isochromats with Lagrangian particles deposited within any mesh, using the velocity from an MFN to update each position of particles "on the fly"

- solve Bloch's equations independently for each particle.