US-20260126505-A1 - Quantitative Material Characterization of an Object

US20260126505A1US 20260126505 A1US20260126505 A1US 20260126505A1US-20260126505-A1

Abstract

For material characterization, first and second measured MRI data representing an object and corresponding first and second sequence descriptions may be received. For each iteration, first and second simulated MRI data may be generated according to the sequence descriptions based on a model including model values for at least one material parameter. An error may be determined, which depends on a deviation of the simulated MRI data from the measured MRI data. The model may be adapted based on the error. For an initial iteration, the model corresponds to an initial model and otherwise it corresponds to the adapted model of the preceding iteration. The quantitative material characterization may be determined based on the adapted model of a final iteration of the two or more iterations.

Inventors

Laura Pfaff
Tobias Würfl
Marcel Dominik Nickel
Moritz ZAISS
Simon Weinmüller
Jonathan Endres

Assignees

Siemens Healthineers Ag

Dates

Publication Date: 20260507
Application Date: 20251103
Priority Date: 20241104

Claims (15)

1 . A computer-implemented method for determining a quantitative material characterization of an object, the method comprising: receiving first measured magnetic resonance imaging (MRI) data representing the object and receiving a first sequence description specifying a first acquisition sequence usable for generating the first measured MRI data; receiving second measured MRI data representing the object and a second sequence description specifying a second acquisition sequence usable for generating the second measured MRI data; for each iteration of two or more iterations: generating first simulated MRI data by simulating an MRI acquisition according to the first sequence description based on a model for the object comprising respective model values for at least one material parameter of the object; generating second simulated MRI data by simulating an MRI acquisition according to the second sequence description based on the model; determining an error, which depends on a deviation of the first simulated MRI data from the first measured MRI data and on a deviation of the second simulated MRI data from the second measured MRI data; and adapting the model by adapting the model values depending on the error, wherein, based on a respective iteration being an initial iteration of the two or more iterations, the model of the respective iteration corresponds to an initial model for the object comprising respective predefined initial model values for the at least one material parameter, and, otherwise, the model of the respective iteration corresponds to the adapted model of a respective preceding iteration; and determining the quantitative material characterization based on the adapted model of a final iteration of the two or more iterations.
2 . The computer-implemented method according to claim 1 , wherein a type of the first acquisition sequence differs from a type of the second acquisition sequence.
3 . The computer-implemented method according to claim 2 , wherein: the type of the first acquisition sequence corresponds to a first T1-weighted MRI acquisition, a first T2-weighted MRI acquisition, a first proton density (PD)-weighted MRI acquisition, a first fluid attenuated inversion recovery (FLAIR) MRI acquisition, or a first diffusion weighted MRI acquisition; and/or the type of the second acquisition sequence corresponds to a second T1-weighted MRI acquisition, a second T2-weighted MRI acquisition, a second PD-weighted MRI acquisition, a second FLAIR MRI acquisition, or a second diffusion weighted MRI acquisition.
4 . The computer-implemented method according to claim 1 , wherein: the first sequence description specifies a first echo time and the second sequence description specifies a second echo time different from the first echo time; and/or the first sequence description specifies a first repetition time and the second sequence description specifies a second repetition time different from the first repetition time.
5 . The computer-implemented method according to claim 1 , wherein the model for the object comprises a respective spatially resolved parameter map for each material parameter of the at least one material parameter.
6 . The computer-implemented method according to claim 1 , wherein the at least one material parameter comprises: a T1-relaxation time, a T2-relaxation time, a T2*-relaxation time, a proton density (PD), and/or an apparent diffusion coefficient.
7 . The computer-implemented method according to claim 1 , wherein the generating the first simulated MRI data comprises: applying a differentiable MRI simulator to the model of the respective iteration and the first sequence description in order to generate the respective first simulated MRI data, and applying the differentiable MRI simulator the model of the respective iteration and the second sequence description in order to generate the respective second simulated MRI data.
8 . The computer-implemented method according to claim 1 , wherein determining the error comprises computing the respective error based on: a mean squared error of the deviation of the first simulated MRI data from the first measured MRI data, and a mean squared error of the deviation of the second simulated MRI data from the second measured MRI data.
9 . The computer-implemented method according to claim 1 , wherein the two or more iterations are terminated after the final iteration of the two or more iterations, wherein the error of the final iteration is equal to or less than a predefined maximum error.
10 . The computer-implemented method according to claim 1 , wherein the first measured MRI data and the second measured MRI data are respective raw data in k-space.
11 . The computer-implemented method according to claim 1 , further comprising: receiving third measured MRI data representing the object and receiving a third sequence description specifying a third acquisition sequence usable for generating the third measured MRI data; and for each iteration of the two or more iterations, generating third simulated MRI data by simulating an MRI acquisition according to the third sequence description based on the model of the respective iteration, the error being determined based on a deviation of the third simulated MRI data from the third measured MRI data.
12 . A method for quantitative magnetic resonance imaging (MRI), comprising: generating, by an MRI device according to a predefined first acquisition sequence, first measured MRI data representing an object; generating, by the MRI device according to a predefined second acquisition sequence, second measured MRI data representing the object; and performing the computer-implemented method according to claim 1 .
13 . A data processing system comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to perform the computer-implemented method according to claim 1 .
14 . A magnetic resonance imaging (MRI) system comprising: a data processing system including one or more processors and memory storing instructions that, when executed by the one or more processors, cause the system to perform the computer-implemented method according to claim 1 ; and an MRI device configured to generate the first measured MRI data according to the first acquisition sequence and the second measured MRI data according to the second acquisition sequence.
15 . At least one non-transitory computer-readable medium comprising instructions stored thereon, that when executed by one or more processors, cause the one or more processors to perform the computer-implemented method according to claim 1 .

Description

CROSS REFERENCE TO RELATED APPLICATIONS This patent application claims priority to European Patent Application No. 24210519.5, filed Nov. 4, 2024, which is incorporated herein by reference in its entirety. BACKGROUND The present disclosure is directed to a computer-implemented method for determining a quantitative material characterization of an object, wherein first measured magnetic resonance imaging, MRI, data representing the object is received. The disclosure is further directed to a data processing system for carrying out said computer-implemented method, to a corresponding method for quantitative MRI, to an MRI system comprising said data processing system, and to corresponding computer program products. Quantitative MRI, qMRI, plays an important role in medical imaging by providing quantitative, objective measurements of tissue properties, such as relaxation times or proton density. Unlike conventional MRI, which produces images with qualitative contrasts only, qMRI aims to derive material parameters that have direct relevance to the underlying tissue composition and microstructure. This shift towards quantification enhances the precision and reproducibility of MRI examinations and facilitates comparisons across different subjects and scanners. Various approaches have been suggested for the quantification of tissue parameters. However, most of them require excessive scan time, which poses challenges for their practical implementation in clinical settings and provides limited reproducibility. In acquisition processes, parametric mapping frequently involves a delicate balance between precision and accuracy versus the time required for measurement. Precise and accurate quantification often relies on “clean” acquisitions of multiple contrasts that only depend on few parameters. These parameters are later determined by fitting the different contrasts to a signal model. Prime examples include inversion or saturation recovery measurements for determining T1 and T2, which are often considered as gold standard. Accelerated acquisition methods frequently require trade-offs in contrast accuracy. For instance, variable flip angle, vFA, T1 mapping often relies on less precise steady states to expedite the process. Magnetic resonance fingerprinting, MRF, is an alternative approach in qMRI, which simultaneously acquires multiple parameters in a single scan. Instead of isolating specific sequences for individual measurements, MRF aims to generate a “fingerprint” of tissue properties, allowing for more efficient and comprehensive quantification. The underlying idea is based on a complete Bloch simulation of the applied acquisition sequence diagram for an imaged voxel. A significant compromise in MRF lies in the calculation of the signal evolution in a given voxel from the acquired k-space data. Since many samples for the signal evolution are required and a complete optimization of the whole volume is intractable with conventional reconstruction methods, typically simple regridding reconstructions are done with high undersampling. The method relies on the hope that corresponding undersampling artifacts average out in the voxel-wise dictionary matching. Moreover, establishing standardized dictionaries and protocols across different scanners and sites is a challenge for widespread clinical adoption. MRI simulators are software tools designed to replicate the processes and outcomes of MRI without the need for an actual MRI scanner. They work by using a model for an object and simulations to emulate the physical principles involved in MRI. Some simulators are based on solving the Bloch equations directly, as for example described in the publication of H. Benoit-Cattin et al.: “The SIMRI project: a versatile and interactive MRI simulator.”, Journal of Magnetic Resonance, 173(1), 97-115. Others rely on the extended phase graph algorithm, as explained in S. Rakshit et al.: “GPU-accelerated extended phase graph algorithm for differentiable optimization and learning.” Proc. Intl. Soc. Mag. Reson. Med. 29 (2021), available at https://somnathrakshit.github.io/projects/project-mri-sim-py-epg/3754.html (retrieved Oct. 11, 2024). MRzero is a comprehensive framework that emulates an MRI pipeline, encompassing sequence and phantom definition, signal simulation, and image reconstruction, as described in A. Loktyushin et al.: “MRzero-Automated discovery of MRI sequences using supervised learning.”, Magnetic Resonance in Medicine, 86: 709-724 and H. Dang et al.: “MR-zero meets RARE MRI: Joint optimization of refocusing flip angles and neural networks to minimize T2-induced blurring in spin echo sequences.”, Magnetic Resonance in Medicine, 90(4): 1345-1362. At its core, MRzero incorporates a phase distribution graph, PDG, simulator inspired by the EPG concept, as described in J. Endres et al.: “Phase distribution graphs for fast, differentiable, and spatially encoded Bloch simulations of arbitrary MRI sequences.” Magnetic Resonance in M