U.S. patent number 11,417,423 [Application Number 16/986,787] was granted by the patent office on 2022-08-16 for multi-coil magnetic resonance imaging with artificial intelligence.
This patent grant is currently assigned to Shanghai United Imaging Intelligence Co., LTD.. The grantee listed for this patent is Shanghai United Imaging Intelligence Co., LTD.. Invention is credited to Terrence Chen, Xiao Chen, Zhang Chen, Shanhui Sun.
United States Patent |
11,417,423 |
Chen , et al. |
August 16, 2022 |
Multi-coil magnetic resonance imaging with artificial
intelligence
Abstract
A method includes acquiring magnetic resonance imaging (MRI)
data with multi-coil dimensions, compressing the coil dimensions to
a fixed and predetermined number of virtual coils, and utilizing
the fixed and predetermined number of virtual coils by an
artificial intelligence engine for artificial intelligence
applications.
Inventors: |
Chen; Xiao (Cambridge, MA),
Chen; Zhang (Cambridge, MA), Sun; Shanhui (Cambridge,
MA), Chen; Terrence (Cambridge, MA) |
Applicant: |
Name |
City |
State |
Country |
Type |
Shanghai United Imaging Intelligence Co., LTD. |
Shanghai |
N/A |
CN |
|
|
Assignee: |
Shanghai United Imaging
Intelligence Co., LTD. (Xuhui District, CN)
|
Family
ID: |
1000006499419 |
Appl.
No.: |
16/986,787 |
Filed: |
August 6, 2020 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20220044790 A1 |
Feb 10, 2022 |
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G01R
33/34 (20130101); G01R 33/5608 (20130101); G16H
30/20 (20180101); G06N 3/08 (20130101) |
Current International
Class: |
G16H
30/20 (20180101); G06N 3/08 (20060101); G01R
33/34 (20060101); G01R 33/56 (20060101) |
Field of
Search: |
;324/309 |
References Cited
[Referenced By]
U.S. Patent Documents
Primary Examiner: Fuller; Rodney E
Attorney, Agent or Firm: Ziegler IP Law Group, LLC
Claims
What is claimed is:
1. A method comprising: acquiring magnetic resonance imaging (MRI)
data with multi-coil dimensions; compressing the coil dimensions to
a fixed and predetermined number of virtual coils by: using a
central region of a k-space of each coil dimension as calibration
data; factorizing the calibration data using single value
decomposition to form a compression matrix; and multiplying each
coil dimension by the compression matrix; the method further
comprising utilizing the fixed and predetermined number of virtual
coils by an artificial intelligence engine for artificial
intelligence applications.
2. The method of claim 1, comprising acquiring the MRI data with
multi-coil dimensions from one or more of an MRI scanner, an MRI
data storage, a k-space storage, or an image storage.
3. The method of claim 1, wherein the multi-coil dimensions
comprise 12, 16, 24, 32, or 64 coils.
4. The method of claim 1, wherein the fixed and predetermined
number of virtual coils is between 4 and 20.
5. The method of claim 1, comprising compressing the coil
dimensions to a fixed and predetermined number of virtual coils
using one or more of principle component analysis, independent
component analysis, kernel principal component analysis, machine
learning, or deep learning.
6. The method of claim 1, comprising compressing the coil
dimensions to a fixed and predetermined number of virtual coils
using a coil compression engine.
7. The method of claim 6, wherein the coil compression engine is
incorporated as part of the artificial intelligence engine.
8. The method of claim 7, wherein the coil compression engine
comprises a convolutional layer of the artificial intelligence
engine.
9. The method of claim 1, wherein the artificial intelligence
engine comprises one or more of a deep learning model including one
or more gated recurrent units, long short term memory networks,
fully convolutional neural network models, generative adversarial
networks, back propagation neural network models, radial basis
function neural network models, deep belief nets neural network
models, or Elman neural network models.
10. The method of claim 1, wherein the artificial intelligence
applications comprise one or more of recovering image information
from undersampled multi-coil k-space data, denoising of multi-coil
k-space data, or denoising image data.
11. A system comprising: a multi-coil MRI data source; a coil
compression engine configured to compress multi-coil dimensioned
MRI data from the multi-coil MRI data source to a fixed and
predetermined number of virtual coils by: using a central region of
a k-space of each coil dimension as calibration data; factorizing
the calibration data using single value decomposition to form a
compression matrix; and multiplying each coil dimension by the
compression matrix; an artificial intelligence engine that utilizes
the fixed and predetermined number of virtual coils for artificial
intelligence applications.
12. The system of claim 11, wherein the MRI data is acquired from
one or more of an MRI scanner, an MRI data storage, a k-space
storage, or an image storage.
13. The system of claim 11, wherein the multi-coil dimensions
comprise 12, 16, 24, 32, or 64 coils.
14. The system of claim 11, wherein the fixed and predetermined
number of virtual coils is between 4 and 20.
15. The system of claim 11, wherein the coil compression engine is
configured to compress the coil dimensions to a fixed and
predetermined number of virtual coils using one or more of
principle component analysis, independent component analysis,
kernel principal component analysis, machine learning, or deep
learning.
16. The system of claim 11, wherein the coil compression engine is
incorporated as part of the artificial intelligence engine.
17. The system of claim 16, wherein the coil compression engine
comprises a convolutional layer of the artificial intelligence
engine.
18. The system of claim 11, wherein the artificial intelligence
engine comprises one or more of a deep learning model including one
or more gated recurrent units, long short term memory networks,
fully convolutional neural network models, generative adversarial
networks, back propagation neural network models, radial basis
function neural network models, deep belief nets neural network
models, or Elman neural network models.
19. The system of claim 11, wherein the artificial intelligence
applications comprise one or more of recovering image information
from undersampled multi-coil k-space data, denoising of multi-coil
k-space data, or denoising image data.
Description
BACKGROUND
The aspects of the present disclosure relate generally to Magnetic
Resonance imaging (MRI), and in particular to using coil
compression to enable application of artificial intelligence
methods to multi-coil MRI.
MRI is a widely used medical technique which produces images of a
region of interest using magnetic and radio frequency energy.
During an MRI scan, volume coils (for example, body coils) and
local coils (for example, surface coils) may acquire MR signals
produced by nuclear relaxation inside the object being
examined.
Most MRI scanners use multiple receiving coils to collect spatially
varying signals simultaneously, which greatly reduces scanning time
and increases image quality. For a targeted image resolution and
size, the use of multi-coil acquisition increases the size of the
data collected by a factor of the number of coils. For example, a
targeted image of N.sub.x.times.N.sub.y.times.N.sub.z will collect
N.sub.x.times.N.sub.y.times.N.sub.z.times.N.sub.coil data, where
N.sub.x, N.sub.y, N.sub.z are the spatial dimensions along the x,
y, and z axes, respectively, and N.sub.coil is the number of coils.
For contemporary MRI scanners, the coil number may be 64 or more,
with a general rule that as the number of coils increase, so does
image quality and acquisition speed. However, the increase in coils
also results in a challenging amount of data to be processed for
reconstruction and post-analysis.
Artificial intelligence, implemented for example using
deep-learning (DL) based neural networks (NN), has gained much
attention recently, given its huge success in general computer
vision. Multiple DL methods are proposed for MRI image
reconstruction and processing and has shown promising results.
However, for practical implementation of the algorithms for MRI,
the incoming data as inputs to the NN will have a large and unfixed
coil dimension, as the user may choose any number of coils smaller
than the system maximum. Most of the currently proposed DL methods
require a specific input size and a larger coil dimension may
require increased memory consumption during both inference and
training, which may limit NN complexity and capability.
One solution to the variable coil dimension is to apply the DL
method independently to each coil during training and inference.
However, this may introduce several potential problems: 1) images
of each coil during training may be quite different from those of
testing because the position of the coils can be arbitrarily
configured during a scan; 2) signals of each individual coil may be
noisier and correlation between coils is ignored resulting in a
difficult training problem; and 3) inference time is significantly
increased as a result of applying the DL method to each coil
independently.
Another solution may include combining multiple coil signals into a
single combined coil image using coil sensitivity maps, however,
this approach relies heavily on the quality of the estimation of
the coil sensitivity maps.
SUMMARY
It would be advantageous to provide a method and system that
provides a fixed coil dimension for artificial intelligence
applications independent of the number of coils utilized during MR
image acquisition.
According to an aspect of the present disclosure a method includes
acquiring MRI data with multi-coil dimensions, compressing the coil
dimensions to a fixed and predetermined number of virtual coils,
and utilizing the fixed and predetermined number of virtual coils
for artificial intelligence applications.
The method may include acquiring the MRI data with multi-coil
dimensions from one or more of an MRI scanner, an MRI data storage,
a k-space storage, or an image storage.
The multi-coil dimensions may include 12, 16, 24, 32, or 64
coils.
The fixed and predetermined number of virtual coils may be between
4 and 20.
The method may include compressing the coil dimensions to a fixed
and predetermined number of virtual coils using one or more of
principle component analysis, independent component analysis,
kernel principal component analysis, machine learning, or deep
learning.
The method may include compressing the coil dimensions to a fixed
and predetermined number of virtual coils using a coil compression
engine.
The coil compression engine may be incorporated as part of the
artificial intelligence engine.
The coil compression engine may be a convolutional layer of the
artificial intelligence engine.
The artificial intelligence engine may include one or more of a
deep learning model including one or more gated recurrent units,
long short term memory networks, fully convolutional neural network
models, generative adversarial networks, back propagation neural
network models, radial basis function neural network models, deep
belief nets neural network models, or Elman neural network
models.
According to an aspect of the present disclosure a system includes
a multi-coil MRI data source, a coil compression engine configured
to compress multi-coil dimensioned MRI data from the multi-coil MRI
data source to a fixed and predetermined number of virtual coils,
and an artificial intelligence engine that utilizes the fixed and
predetermined number of virtual coils for artificial intelligence
applications.
These and other aspects, implementation forms, and advantages of
the exemplary embodiments will become apparent from the embodiments
described herein considered in conjunction with the accompanying
drawings. It is to be understood, however, that the description and
drawings are designed solely for purposes of illustration and not
as a definition of the limits of the disclosed invention, for which
reference should be made to the appended claims. Additional aspects
and advantages of the invention will be set forth in the
description that follows, and in part will be obvious from the
description, or may be learned by practice of the invention.
Moreover, the aspects and advantages of the invention may be
realized and obtained by means of the instrumentalities and
combinations particularly pointed out in the appended claims.
BRIEF DESCRIPTION OF THE DRAWINGS
In the following detailed portion of the present disclosure, the
invention will be explained in more detail with reference to the
example embodiments shown in the drawings. These embodiments are
non-limiting exemplary embodiments, in which like reference
numerals represent similar structures throughout the several views
of the drawings, wherein:
FIG. 1 illustrates an exemplary process flow according to aspects
of the disclosed embodiments;
FIG. 2 illustrates an embodiment of an exemplary system
incorporating aspects of the disclosed embodiments;
FIG. 3 shows a schematic block diagram of an exemplary multi-coil
MRI data source according to the disclosed embodiments;
FIGS. 4A and 4B illustrate different MRI coil arrangements
according to the disclosed embodiments;
FIG. 5 shows various exemplary embodiments of MRI data sources for
implementing the disclosed embodiments;
FIG. 6 illustrates an exemplary architecture of a coil compression
engine according to the disclosed embodiments;
FIG. 7 depicts an exemplary neural network that may be utilized to
implement the disclosed embodiments; and
FIG. 8 shows an exemplary procedure for utilizing a multi-coil MRI
data source, a coil compression engine, and an artificial
intelligence engine according to the disclosed embodiments.
DETAILED DESCRIPTION
In the following detailed description, numerous specific details
are set forth by way of examples in order to provide a thorough
understanding of the relevant disclosure. However, it should be
apparent to those skilled in the art that the present disclosure
may be practiced without such details. In other instances, well
known methods, procedures, systems, components, and/or circuitry
have been described at a relatively high-level, without detail, in
order to avoid unnecessarily obscuring aspects of the present
disclosure. Various modifications to the disclosed embodiments will
be readily apparent to those skilled in the art, and the general
principles defined herein may be applied to other embodiments and
applications without departing from the spirits and scope of the
present disclosure. Thus, the present disclosure is not limited to
the embodiments shown, but to be accorded the widest scope
consistent with the claims.
It will be understood that the term "system," "unit," "module,"
and/or "block" used herein are one method to distinguish different
components, elements, parts, section or assembly of different level
in ascending order. However, the terms may be displaced by other
expression if they may achieve the same purpose.
It will be understood that when a unit, module or block is referred
to as being "on," "connected to" or "coupled to" another unit,
module, or block, it may be directly on, connected or coupled to
the other unit, module, or block, or intervening unit, module, or
block may be present, unless the context clearly indicates
otherwise. As used herein, the term "and/or" includes any and all
combinations of one or more of the associated listed items.
Generally, the word "module," "unit," or "block," as used herein,
refers to logic embodied in hardware or firmware, or to a
collection of software instructions. A module, a unit, or a block
described herein may be implemented as software and/or hardware and
may be stored in any type of non-transitory computer-readable
medium or another storage device. In some embodiments, a software
module/unit/block may be compiled and linked into an executable
program. It will be appreciated that software modules can be
callable from other modules/units/blocks or from themselves, and/or
may be invoked in response to detected events or interrupts.
Software modules/units/blocks configured for execution on computing
devices may be provided on a computer-readable medium, such as a
compact disc, a digital video disc, a flash drive, a magnetic disc,
or any other tangible medium, or as a digital download (and can be
originally stored in a compressed or installable format that needs
installation, decompression, or decryption prior to execution).
Such software code may be stored, partially or fully, on a storage
device of the executing computing device, for execution by the
computing device. Software instructions may be embedded in
firmware, such as an Erasable Programmable Read Only Memory
(EPROM). It will be further appreciated that hardware
modules/units/blocks may be included in connected logic components,
such as gates and flip-flops, and/or can be included of
programmable units, such as programmable gate arrays or processors.
The modules/units/blocks or computing device functionality
described herein may be implemented as software
modules/units/blocks, but may be represented in hardware or
firmware. In general, the modules/units/blocks described herein
refer to logical modules/units/blocks that may be combined with
other modules/units/blocks or divided into
sub-modules/sub-units/sub-blocks despite their physical
organization or storage. The description may be applicable to a
system, an engine, or a portion thereof.
The terminology used herein is for the purposes of describing
particular examples and embodiments only, and is not intended to be
limiting. As used herein, the singular forms "a," "an," and "the"
may be intended to include the plural forms as well, unless the
context clearly indicates otherwise. It will be further understood
that the terms "include," and/or "comprise," when used in this
disclosure, specify the presence of integers, devices, behaviors,
stated features, steps, elements, operations, and/or components,
but do not exclude the presence or addition of one or more other
integers, devices, behaviors, features, steps, elements,
operations, components, and/or groups thereof.
These and other features, and characteristics of the present
disclosure, as well as the methods of operation and functions of
the related elements of structure and the combination of parts and
economies of manufacture, may become more apparent upon
consideration of the following description with reference to the
accompanying drawings, all of which form a part of this disclosure.
It is to be expressly understood, however, that the drawings are
for the purpose of illustration and description only and are not
intended to limit the scope of the present disclosure. It is
understood that the drawings are not to scale.
The disclosed embodiments may generally utilize coil compression to
enable practical application of artificial intelligence methods for
multi-coil MRI. The disclosed embodiments may also utilize coil
compression to address the issues associated with using artificial
intelligence methods for multi-coil MRI, for example, memory
requirements, computational speed, training complexity and neural
network capability. The disclosed embodiments may further utilize
coil compression to exploit the redundancy among coils and
calculate a linear or non-linear transform that can sparsify the
data along a fixed, predetermined coil dimension, resulting in a
fixed and predetermined number of virtual coils that carry most of
the information. Where the MRI data is acquired from a number of
coils that is smaller than the pre-determined number of virtual
coils, exemplary operations such as padding zeros may be used to
increase the number of virtual coils to the pre-determined number.
Where the MRI data is acquired from a number of coils coil that is
the same as the pre-determined number of virtual coils, exemplary
operations may be utilized that output the data as input, or
operations may be utilized that may compress the MRI data to the
same number of virtual coils as input. Exemplary methods such as
dimension reduction by principal component analysis (PCA) may be
utilized for coil compression where the MRI data is acquired from a
number of coils that is larger than the pre-determined number of
virtual coils.
The disclosed embodiments are directed to a method comprising
acquiring multi-coil MRI data, using an algorithm to compress MRI
data from any number of coils to a fixed number of virtual coils,
and providing the fixed number of virtual coils to a neural network
for further processing.
The disclosed embodiments are further directed to a system
comprising a source of multi-coil MRI data, a coil compression
engine operating to compress the multi-coil MRI data to a fixed
number of virtual coils, and a neural network to process the fixed
number virtual coil data.
Referring to FIG. 1, a schematic block diagram of an exemplary
system 100 incorporating aspects of the disclosed embodiments is
illustrated. The system may include a multi-coil MRI data source
102 for providing MRI data from any number of coil assemblies. A
coil compression engine 104 may receive the multi-coil MRI data and
may operate to compress the multi-coil MRI data to a fixed number
of virtual coils whether the number of actual coils is lager,
smaller, or the same as the fixed number of virtual coils. The
multi-coil MRI data, compressed to a fixed number of virtual coils,
may be provided to an artificial intelligence engine 106 for
various processing operations. It should be understood that the
coil compression engine 104 and the artificial intelligence engine
106 of the exemplary system 100 may be implemented in hardware,
software, or a combination of hardware and software.
FIG. 2 illustrates an embodiment of an exemplary system 200
incorporating aspects of the disclosed embodiments. The system may
include a multi-coil MRI data source 102 for providing MRI data
from any number of coil assemblies. A coil compression engine 204
may receive the multi-coil MRI data and may operate to compress the
multi-coil MRI data to a fixed number of virtual coils, however in
this embodiment, the coil compression engine 204 may be
incorporated as part of the artificial intelligence engine 202 and
may provide the multi-coil MRI data, compressed to a fixed number
of virtual coils, to another section of the artificial intelligence
engine, for example, a deep learning model 206. The artificial
intelligence engine 202 may be implemented in hardware, software,
or a combination of hardware and software.
FIG. 3 shows a schematic block diagram of an exemplary multi-coil
MRI data source 102 in the form of an MRI apparatus 302 for
providing multi-coil MRI data according to the disclosed
embodiments. The MRI apparatus 302 may include an MRI scanner 304,
control circuitry 306 and a display 308. The function, size, type,
geometry, position, amount, or magnitude of the MRI scanner 304 may
be determined or changed according to one or more specific
conditions. For example, the MRI scanner 304 may be designed to
surround a subject (or a region of the subject) to form a tunnel
type MRI scanner, referred to as a closed bore MRI scanner, or an
open MRI scanner, referred to as an open-bore MRI scanner. The MRI
scanner 302 may include, as shown in cross section in FIG. 3, a
magnetic field generator 310, a gradient magnetic field generator
312, and a Radio Frequency (RF) generator 314, all surrounding a
table 316 on which subjects under study may be positioned. The MRI
scanner 304 may also include one or more coil arrays 318, an ECG
signal sensor 320 for capturing MRI data in the form of ECG signals
from the subject under study during MRI scanning, and a camera 322
for capturing MRI data in the form of video images of the subject
under study during MRI scanning.
In some embodiments, the MRI scanner 304 may perform a scan on a
subject or a region of the subject. The subject may be, for
example, a human body or other animal body. In some embodiments,
the subject may be a patient. The region of the subject may include
part of the subject. For example, the region of the subject may
include a tissue of the patient. The tissue may include, for
example, lung, prostate, breast, colon, rectum, bladder, ovary,
skin, liver, spine, bone, pancreas, cervix, lymph, thyroid, spleen,
adrenal gland, salivary gland, sebaceous gland, testis, thymus
gland, penis, uterus, trachea, skeletal muscle, smooth muscle,
heart, etc. In some embodiments, the scan may be a pre-scan for
calibrating an imaging scan. In some embodiments, the scan may be
an imaging scan for generating an image.
The main magnetic field generator 310 may create a static magnetic
field B.sub.0 and may include, for example, a permanent magnet, a
superconducting electromagnet, a resistive electromagnet, or any
magnetic field generation device suitable for generating a static
magnetic field. The gradient magnet field generator 312 may use
coils to generate a magnetic field in the same direction as B.sub.0
but with a gradient in one or more directions, for example, along
X, Y, or Z axes in a coordinate system of the MRI scanner 304.
In some embodiments, the RF generator 314 may use RF coils to
transmit RF energy through the subject, or region of interest of
the subject, to induce electrical signals in the region of
interest. The resulting RF field is typically referred to as the B1
field and combines with the B0 field to generate MR signals that
are spatially localized and encoded by the gradient magnetic field.
The coil arrays 318 may generally operate to sense the RF field and
convey a corresponding output to the control circuitry 306. In some
embodiments, the coil arrays may operate to both transmit and
receive RF energy, while in other embodiments, the coil arrays may
operate as receive only.
FIGS. 4A and 4B illustrate different MRI coil arrangements. The
coil arrangements may include phased array coil arrangements and
parallel array coil arrangements. FIG. 4A shows an exemplary phased
array coil arrangement where the coils overlap and are coupled
together to enhance gain and signal to noise properties. FIG. 4B
shows an exemplary parallel array arrangement where the coils are
decoupled and optimized for parallel imaging. The coil arrangements
may include any number of coils, depending on a particular
application. Exemplary numbers of coils may include 12, 16, 24, 32,
64 or more.
Returning to FIG. 3, the control circuitry 306 may control overall
operations of the MRI scanner 304, in particular, the magnetic
field generator 310, the gradient magnetic field generator 312, the
RF generator 314, and the coil arrays 318. For example, the control
circuitry 306 may control the magnet field gradient generator to
produce gradient fields along one or more of the X, Y, and Z axes,
and the RF generator to generate the RF field. In some embodiments,
the control circuitry 306 may receive commands from, for example, a
user or another system, and control the magnetic field generator
310, the gradient magnetic field generator 312, the RF generator
314, and the coil arrays 318 accordingly.
The control circuitry 306 may be connected to the MRI scanner 304
through a network 324. The network 324 may include any suitable
network that can facilitate the exchange of information and/or data
for the MRI scanner 304. The network 324 may be and/or include a
public network (e.g., the Internet), a private network (e.g., a
local area network (LAN), a wide area network (WAN)), etc.), a
wired network (e.g., an Ethernet network), a wireless network
(e.g., an 802.11 network, a Wi-Fi network, etc.), a cellular
network (e.g., a Long Term Evolution (LTE) network), a frame relay
network, a virtual private network ("VPN"), a satellite network, a
telephone network, routers, hubs, switches, server computers,
and/or any combination thereof. Merely by way of example, the
network 324 may include a cable network, a wireline network, a
fiber-optic network, a telecommunications network, an intranet, a
wireless local area network (WLAN), a metropolitan area network
(MAN), a public telephone switched network (PSTN), a Bluetooth.TM.
network, a ZigBee.TM. network, a near field communication (NFC)
network, or the like, or any combination thereof. In some
embodiments, the network 324 may include one or more network access
points. For example, the network 324 may include wired and/or
wireless network access points such as base stations and/or
internet exchange points through which one or more components of
the MRI scanner 402 may be connected with the network 324 to
exchange data and/or information.
According to some embodiments, the coil compression engine 104 and
the artificial intelligence engine 106 may be incorporated in the
control circuitry 306, while in other embodiments, one or both of
the coil compression engine 104 and the artificial intelligence
engine 106 may be located remotely from the control circuitry
306.
FIG. 5 shows various exemplary embodiments of MRI data sources for
implementing the disclosed embodiments. The sources of MRI data may
include, without limitation, one or more of the MRI scanner 304, a
storage of multi-coil MRI data 504, for example, MRI slices or
other MRI apparatus output, a storage of multi-coil k-space data
506 from any number of MRI scans, and an image storage 508 of
multi-coil MRI images, or any other suitable source of multi-coil
MRI data. The MRI data sources may further include any number of
local, remote, or cloud based sources.
FIG. 6 illustrates an exemplary architecture of the coil
compression engine 104 according to the disclosed embodiments. The
coil compression engine 104 may include computer readable program
code stored on at least one computer readable medium 602 for
carrying out and executing the process steps described herein. The
computer readable program code for carrying out operations for
aspects of the present disclosure may be written in any combination
of one or more programming languages, including an object-oriented
programming language such as Java, Scala, Smalltalk, Eiffel, JADE,
Emerald, C++, C#, VB. NET, Python or the like, conventional
procedural programming languages, such as the "C" programming
language, Visual Basic, Fortran 2103, Perl, COBOL 2102, PHP, ABAP,
dynamic programming languages such as Python, Ruby, and Groovy, or
other programming languages. The computer readable program code may
execute entirely on the coil compression engine 104, partly on the
coil compression engine 104, as a stand-alone software package,
partly on the coil compression engine 104 and partly on a remote
computer or server or entirely on the remote computer or server. In
the latter scenario, the remote computer may be connected to the
coil compression engine 104 through any type of network, including
those mentioned above with respect to network 324.
The computer readable medium 602 may be a memory of the coil
compression engine 104. In alternate aspects, the computer readable
program code may be stored in a memory external to, or remote from,
the coil compression engine 104. The memory may include magnetic
media, semiconductor media, optical media, or any media which is
readable and executable by a computer. The coil compression engine
104 may also include a computer processor 604 for executing the
computer readable program code stored on the at least one computer
readable medium 602. In at least one aspect, the coil compression
engine 104 may include one or more input or output devices,
generally referred to as a user interface 606 which may operate to
allow input to the coil compression engine 104 or to provide output
from the coil compression engine 104, respectively. The coil
compression engine 104 may be implemented in hardware, software or
a combination of hardware and software.
The coil compression engine 104 generally operates to linearly or
nonlinearly combine raw data from a variable number of multiple
coils, depending on the MRI scanner producing the MRI data, into a
fixed number of virtual coils. Exemplary fixed numbers of virtual
coils may generally be between 4 and 20, however, it should be
understood that any fixed number of virtual coils may be utilized.
The coil compression engine 104 may utilize any number of various
compression techniques including, but not limited to Principle
Component Analysis (PCA), Independent Component Analysis (ICA),
Kernel Principal Component Analysis (KPCA), Machine Learning (ML),
Deep Learning (DL). In some embodiments, to perform the coil
compression, a n.sub.calib.times.n.sub.calib central region of the
k-space of every coil representing low-spatial-frequency component
y.sub.calib.di-elect
cons.C.sup.n.sup.calib.sup.2.sup..times.n.sup.c is used as the
calibration data. The calibration data is factorized using singular
value decomposition and the first n.sub.vc columns of the
right-singular vectors are kept to form a compression matrix
M.sub.c.di-elect cons.C.sup.n.sup.c.sup..times.nvc. The acquired
n.sub.c-coil data is then compressed to n.sub.vc virtual coils
through y.sub.comp=yM.sub.c, where represents the matrix
multiplication and y is rearranged into shape n.times.n.sub.c
before multiplying.
Referring again to FIG. 2, when incorporated as part of the
artificial intelligence engine 202, the coil compression engine 205
can be implemented as a pre-learned convolutional layer with a
1.times.1 kernel size. Each column of the compression matrix is a
filter and there are total n.sub.vc filters, which convert the
input n.sub.c channel data to n.sub.vc channel features.
FIG. 7 depicts an exemplary simple neural network 700 that may be
utilized to implement the disclosed embodiments. While a simple
neural network is shown, it should be understood that the disclosed
embodiments may be implemented utilizing a deep learning model
including one or more gated recurrent units (GRUs), long short term
memory (LSTM) networks, fully convolutional neural network (FCN)
models, generative adversarial networks (GANs), back propagation
(BP) neural network models, radial basis function (RBF) neural
network models, deep belief nets (DBN) neural network models, Elman
neural network models, or any deep learning or machine learning
model capable of performing the operations described herein. The
multi-coil MRI data, compressed to a fixed number of virtual coils
may be used to train the neural network 700. In one embodiment, the
neural network 700 may operate to recover image information from
acquired multi-coil k-space data, where the data may be
undersampled for acceleration purpose. In another embodiment, the
neural network may operate to perform post-processing such as
denoising of the acquired multi-coil k-space or image data. When
incorporated as part of the artificial intelligence engine 202, the
coil compression engine 205 may be implemented as a special layer
in the deep learning model 206 or the neural network 700.
The number of the virtual coil dimensions can be determined by
experiments or experiences, or be learned from the data as a
hyperparameter. One example of deciding virtual coil number from
experiments or experiences is to calculate the total energy
maintained in the compressed data. The total energy can be
calculated using the Frobenius norm of the data matrix resulting
from the coil compression operation. The same technique may be used
to calculate the energy of the original uncompressed data from the
multi-coil MRI data source, and the total energy of the compressed
signal at different compression rates may also be calculated. Given
the nature of the compression, the remaining energy after
compression may be represented as a monotonic curve that increases
as the number of virtual coils increases. A threshold value of the
remaining energy after compression can be determined heuristically,
where the threshold value is sufficient for subsequent
applications, while at the same time allowing for filtering
unnecessary energy that is mostly noise. The number of virtual
coils that meet the determined threshold value of the remaining
energy after compression may then be selected. In some embodiments,
the selected number of virtual coils may be fixed and added to the
neural network 700 and the compression layer parameters may not
update during the training of the neural network. In another
embodiment, the number of virtual coils may not be fixed and added
to the neural network. The compression layer parameters are
initialized using the pre-calculated parameters and are updated
along with the other parts of the neural network 700 during
training.
Techniques that train to learn or to select a particular neural
network structure can be used to learn the hyperparameter of the
neural network for optimal performance. One example following a
reinforcement learning framework can be a searching neural network
that can act on the tested neural network by changing the
hyperparameters and observing the resulting performance. The
searching network can continuously perform trials of acting and
observing, and accumulate experiences through the trials. The
target of the searching network is to maximize some reward, which
can be defined as achieving better performance. The searching
network will eventually reach an optimal performance point, at
which the operations of the searching network may be
terminated.
FIG. 8 shows an exemplary procedure 800 for utilizing the
multi-coil MRI data source 302, the coil compression engine 104,
204 and the artificial intelligence engine 106, 202 according to
the disclosed embodiments. As shown in block 802, multi-coil MRI
data may be acquired from any suitable source, for example, one or
more of the MRI scanner 304, the storage of multi-coil MRI data
504, for example, MRI slices or other MRI apparatus output, the
storage of multi-coil k-space data 506 from any number of MRI
scans, and the image storage 508 of multi-coil MRI images. The
multi-coil MRI data may be derived from any suitable number of MRI
coils, for example, 12, 16, 24, 32, 64 or more. As shown in block
804, the coil dimensions of the multi-coil data may be compressed
to a fixed and pre-determined number of virtual coils. In some
embodiments, the fixed and predetermined number of virtual coils
may be any suitable number, for example, 10 or less. As shown in
block 806, the fixed and predetermined number of virtual coils may
be used by the artificial intelligence engine 106, 202 for various
applications, for example, recovering undersampled k-space data
808, denoising k-space data 810, and denoising image data 812.
The proposed method will result in no difference in the training or
testing of the artificial intelligence engine 106, 202 in order to
accomplish these exemplary applications. The neural network with
the proposed coil compression "layer" can be formed as supervised,
semi-supervised or unsupervised, can use any kinds of loss, and can
be trained using different kinds of training strategy, so long as
the application utilizes multi-coil data.
In order to accomplish artificial intelligence applications, the
artificial intelligence engine 106 may be trained to reconstruct MR
images from acquired multi-coil k-space data in a supervised
manner. The input to the artificial intelligence engine 106 may be
the multi-coil k-space data and the resulting output of the
artificial intelligence engine 106 may be estimated images. The
estimated images may be compared to ground truth images and the
differences between the estimated images and the ground truth
images may be back-propagated to update the parameters of the
neural network 700 in the artificial intelligence engine 106 to
improve the accuracy of the estimations. Testing of the neural
network 700 in the artificial intelligence engine 106 may be
performed without ground truth images and the input multi-coil
k-space data may be fed to the trained neural network which may
then output estimated reconstructions. For the artificial
intelligence engine 202 with the coil compression engine 204
implemented as a layer of the deep learning model 206 or the neural
network 700, the input, the output, the ground truths and the
training and testing processes may be the same as for the
artificial intelligence engine 106 separate from the coil
compression engine 104. The input multi-coil k-space data may be
provided to the compression layer where the dimension of the
multi-coil k-space data may be compressed to a fixed number. This
compressed data may be further fed to the preceding layers in the
neural network 700. During training, the difference between the
estimations and the ground truths may be back-propagated to update
the parameters of the neural network 700. During testing, the input
multi-coil k-space data may be fed to the trained neural network
which may then output estimated reconstructions.
The compression of a variable coil dimension to a fixed and
predetermined virtual coil dimension may generally result in less
constraints on the design of the neural network. The trained neural
network is presented with a fixed coil dimension, regardless of the
number of coils from which data may be collected, which greatly
increases the flexibility of the artificial intelligence engine
design. The compressed coil number is typically smaller, for
example, less than 10, than the number of coils utilized for the
MRI scan, requiring less memory for testing the neural network. The
smaller memory also allows for training a larger and more complex
neural network, and the smaller number of coils results in a faster
inference time for the neural network. The signals from each
virtual coil may exhibit a higher signal to noise ratio than the
actual physical coils and may translate to the ability of the
neural network to learn more representative features of the MRI
data and better image quality.
Thus, while there have been shown, described and pointed out,
fundamental novel features of the invention as applied to the
exemplary embodiments thereof, it will be understood that various
omissions, substitutions and changes in the form and details of
devices and methods illustrated, and in their operation, may be
made by those skilled in the art without departing from the spirit
and scope of the presently disclosed invention. Further, it is
expressly intended that all combinations of those elements, which
perform substantially the same function in substantially the same
way to achieve the same results, are within the scope of the
invention. Moreover, it should be recognized that structures and/or
elements shown and/or described in connection with any disclosed
form or embodiment of the invention may be incorporated in any
other disclosed or described or suggested form or embodiment as a
general matter of design choice. It is the intention, therefore, to
be limited only as indicated by the scope of the claims appended
hereto.
* * * * *