Computer-implemented method of building a database of pulse sequences for magnetic resonance imaging, and a method of performing magnetic resonance imaging using such a database
A computer-implemented method of building a database of pulse sequences for parallel-transmission magnetic resonance imaging, includes a) for each of a plurality of subjects, determining an optimal sequence for the subject; b) for each subject, computing the values of the or of a different cost or merit function obtained by playing the optimal sequences for all the subjects; c) aggregating the subjects into a plurality of clusters using a clustering algorithm taking the values, or functions thereof, as metrics; d) for each cluster, determining an averaged optimal sequence for the cluster; e) receiving, as input, a set of features characterizing an imaging subject, comprising at least a morphological feature of the subject; f) associating the subject to one pulse sequence of the database based on the set of features using the computer-implemented classifier algorithm; and g) performing magnetic resonance imaging using the pulse sequence. A magnetic resonance imaging apparatus for carrying out steps e)-g) of such a method is also provided.
1. A computer-implemented method of building a database of pulse sequences for magnetic resonance imaging, each pulse sequence comprising at least a magnetic field gradient waveform and a set of radio-frequency pulses, each pulse of said set being defined by a plurality of design parameters and being associated to a respective transmission channel (RFC1-RFC8) of a magnetic resonance imaging apparatus; the method comprising:
a) for each of a plurality of magnetic resonance imaging subjects, determining a pulse sequence, called optimal sequence for the subject, optimizing a cost or merit function representative of a difference between a spatial distribution of nuclear magnetization flip angles, obtained by simulating the pulse sequence on the subject, and a target distribution;
b) for each subject, computing the values of said cost or merit function or of a different cost or merit function representative of a difference between a spatial distribution of nuclear magnetization flip angles and a target distribution, said values being obtained by simulating the optimal sequences for all the subjects;
c) aggregating the subjects into a plurality of clusters using a clustering algorithm taking said values, or functions thereof, as metrics; and
d) for each cluster, determining a pulse sequence, called optimal sequence for the cluster, optimizing a cost or merit function representative of an averaged difference between the spatial distribution of nuclear magnetization flip angles of the subjects belonging to the cluster and the target distribution.
2. The method of claim 1 , wherein step a) comprises:
a1) acquiring inhomogeneity maps of radio-frequency fields generated within the subject by each transmission channel of the magnetic resonance imaging apparatus;
a2) computing, by taking the said inhomogeneity map or maps as inputs, a spatial distribution of the nuclear magnetization flip angles obtained by simulating said radio-frequency sequence, and
a3) determining the optimal pulse sequence for the subject by adjusting the design parameters of said radio-frequency pulses and of said gradient waveform to optimize said cost or merit function.
3. The method of claim 1 , further comprising:
e) training a classifier algorithm associating magnetic resonance imaging subjects to the clusters based at least on subject morphological features.
4. The method of claim 3 , further comprising:
f) using the classifier for associating a magnetic resonance imaging subject, not used for carrying out the preceding steps, to a cluster; and
g) performing magnetic resonance imaging of the subject using the optimal sequence for the cluster.
5. The method of claim 3 , further comprising a step of deriving the morphological features from at least a localizer image of the subject.
6. The method of claim 3 , wherein the classifier algorithm further uses, for associating a magnetic resonance imaging subject to the clusters, at least one feature chosen among:
identity data of the subject;
clinical data of the subject; and
operating parameters of the magnetic resonance imaging apparatus.
7. The method of claim 3 , wherein the classifier algorithm does not use, for associating a magnetic resonance imaging subject to the clusters, features derived by inhomogeneity maps of radio-frequency fields generated within the subject by each transmission channel of the transmission magnetic resonance imaging apparatus.
8. The method of claim 3 , wherein steps a) to d) are carried out using a first set of a plurality of subjects and step e) is carried out using, in addition to the first set, a second set of a plurality of subjects for which the cost or merit function representative of an averaged difference between the spatial distribution of nuclear magnetization flip angles of the subjects and the target distribution is also computed.
9. The method of claim 3 , further comprising a step α) of training a regression model to predict an average value of a nuclear magnetization flip angle as a function of said features, for the optimal pulse sequence of each cluster determined in step d).
10. The method of claim 1 , wherein said clustering algorithm is a hierarchical agglomerative clustering algorithm.
11. The method of claim 10 , wherein said hierarchical agglomerative clustering algorithm uses a complete linkage scheme.
12. The method of claim 1 , wherein the magnetic resonance imaging apparatus is a parallel magnetic resonance imaging apparatus having a plurality of transmission channels.
13. A method of performing magnetic resonance imaging of a subject, comprising:
providing a database of pulse sequences, each pulse sequence comprising at least a magnetic field gradient waveform and a set of radio-frequency pulses, each pulse of said set being defined by a plurality of design parameters and being associated to a respective transmission channel (RFC1-RFC8) of a magnetic resonance imaging apparatus, and each pulse sequence of the database optimizing a cost or merit function representative of an averaged difference between the spatial distribution of nuclear magnetization flip angles of the subjects belonging to the cluster and a target distribution;
receiving, as input, a set of features characterizing the subject, comprising at least a morphological feature of the subject;
associating the subject to one pulse sequence of the database based on said set of features using a computer-implemented classifier algorithm; and
performing magnetic resonance imaging using said pulse sequence.
14. The method of claim 13 , further comprising a step of deriving the morphological features from at least a localizer image of the subject.
15. The method of claim 13 , wherein the classifier algorithm further uses, for associating the magnetic resonance imaging subject to one pulse sequence of the database, at least one feature chosen among:
identity data of the subject;
clinical data of the subject; and
operating parameters of the magnetic resonance imaging apparatus.
16. The method of claim 13 , wherein the magnetic resonance imaging apparatus is a parallel magnetic resonance imaging apparatus having a plurality of transmission channels.
17. The method of claim 13 , further comprising a step β) of using a regression model for predicting, as a function of said set of features, an average nuclear magnetization flip angle induced in the subject by the associated pulse sequence, and a step γ) of modifying the associated pulse sequence as a function of the predicted average nuclear magnetization flip angle and of a target value thereof.
18. A magnetic resonance imaging apparatus comprising:
at least one radio-frequency transmission channel (RFC1-RFC8);
a set of gradient coils;
a memory storing a database of pulse sequences, each pulse sequence comprising at least a magnetic field gradient waveform and a set of radio-frequency pulses, each pulse of said set being defined by a plurality of design parameters and being associated to a respective transmission channel of the magnetic resonance imaging apparatus, and each pulse sequence of the database optimizing a cost or merit function representative of an averaged difference between the spatial distribution of nuclear magnetization flip angles of the subjects belonging to the cluster and a target distribution; and
a computer programmed for:
receiving, as input, a set of features characterizing a magnetic-resonance imaging subject, comprising at least a morphological feature of the subject;
associating the subject to one pulse sequence of the database based on said set of features using a classifier algorithm; and
driving the radio-frequency transmission channel or channels and the gradient coils for performing magnetic resonance imaging using said pulse sequence.
19. The apparatus of claim 18 , comprising a plurality of radio-frequency transmission channels, wherein the computer is programmed for driving the radio-frequency transmission channels and the gradient coils for performing parallel-transmission magnetic resonance imaging using said pulse sequence.