# US 11,992,289 B2 · Fast real-time cardiac cine MRI reconstruction with residual convolutional recurrent neural network

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,992,289 B2
- Document ID: US11992289B2
- Status: Granted
- Kind: B2
- Application number: 17/060,988
- Filed: 2020-10-01
- Published: 2021-06-03
- Granted: 2024-05-28
- Earliest claimed: 2019-11-29
- Assignees: Shanghai United Imaging Intelligence Co., Ltd.
- Applicants: Shanghai United Imaging Intelligence Co., Ltd.
- Inventors: Zhang Chen; Xiao Chen; Shanhui Sun; Terrence Chen
- Web page: https://patentgrep.com/patent/US11992289B2

## CPC

- A61B 5/0044: {for the heart}
- A61B 5/004: {adapted for image acquisition of a particular organ or body part (A61B5/0082 takes precedence; arrangements for optical scanning A61B5/0062)}
- A61B 5/7264: {Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems}
- A61B 5/7267: {involving training the classification device}
- G01R 33/3642: {Mutual coupling or decoupling of multiple coils, e.g. decoupling of a receive coil from a transmission coil, or intentional coupling of RF coils, e.g. for RF magnetic field amplification}
- G01R 33/5608: {Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels (image data processing or generation, in general G06T)}
- G01R 33/5612: {Parallel RF transmission, i.e. RF pulse transmission using a plurality of independent transmission channels}
- G01R 33/5615: {Echo train techniques involving acquiring plural, differently encoded, echo signals after one RF excitation, e.g. using gradient refocusing in echo planar imaging [EPI], RF refocusing in rapid acquisition with relaxation enhancement [RARE] or using both RF and gradient refocusing in gradient and spin echo imaging [GRASE]}
- G01R 33/56325: {Cine imaging}
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0475: Generative networks
- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/092: Reinforcement learning
- G06N 3/094: Adversarial learning
- G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn
- G01R 33/5611: {Parallel magnetic resonance imaging, e.g. sensitivity encoding [SENSE], simultaneous acquisition of spatial harmonics [SMASH], unaliasing by Fourier encoding of the overlaps using the temporal dimension [UNFOLD], k-t-broad-use linear acquisition speed-up technique [k-t-BLAST], k-t-SENSE (structural details of arrays of sub-coils G01R33/3415)}
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method includes using fully sampled retro cine data to train an algorithm, and applying the trained algorithm to real time MR cine data to yield reconstructed MR images.

## Claims

### Claim 1

A method comprising:
obtaining subsampled real-time magnetic resonance imaging (MRI) data associated with multiple coils; and
reconstructing, using at least a residual convolutional recurrent neural network, the subsampled real-time MM data into a reconstructed MRI image, wherein the reconstructed MRI image reflects information acquired via the multiple coils and wherein the residual convolutional recurrent neural network comprises one or more bi-directional convolutional recurrent neural network cells.

### Claim 2 (depends on 1)

The method of claim 1, wherein the subsampled real-time MRI data is based on a sub-sampling mask configured to achieve an acceleration rate of ten times or more.

### Claim 3 (depends on 2)

The method of claim 2, wherein the residual convolutional recurrent neural network is trained using subsampled MRI data generated based on a sub-sampling mask that is substantially similar to the sub-sampling mask on which the subsampled real-time MRI data is based.

### Claim 4 (depends on 1)

The method of claim 1, wherein the subsampled real-time MM data is acquired without electrocardiogram (ECG) gating and breath-holding, and wherein the residual convolutional recurrent neural network is trained using retro cine data acquired based on ECG gating and breath-holding.

### Claim 5 (depends on 1)

The method of claim 1, wherein the residual convolutional recurrent neural network comprises one or more data consistency layers configured to ensure consistency between the reconstructed MRI image and the subsampled real-time MM data.

### Claim 6 (depends on 1)

The method of claim 1, wherein the residual convolutional recurrent neural network comprises one or more residual connections configured to add high-frequency details to the reconstructed MM image.

### Claim 7 (depends on 1)

The method of claim 1, wherein the residual convolutional recurrent neural network is configured to receive complex values associated with the subsampled real-time MRI data as a two-channel tensor.

### Claim 8 (depends on 1)

The method of claim 1, wherein the subsampled real-time MRI data comprises respective subsampled real-time MRI data associated with multiple individual coils of an MRI wherein reconstructing the subsampled real-time MM data into the reconstructed MM image comprises:
reconstructing corresponding single-coil MM images based on the respective subsampled real-time MRI data associated with the multiple individual coils; and
generating the reconstructed MRI image based on a combination of the single-coil MRI images.

### Claim 9 (depends on 8)

The method of claim 8, wherein the reconstructed MRI images is generated based on a root sum of squares or one or more coil sensitivity maps associated with the multiple individual coils of the MRI scanner.

### Claim 10 (depends on 8)

The method of claim 8, wherein the single-coil MM images are generated independently from each other based on the respective subsampled real-time MRI data associated with the multiple individual coils.

### Claim 11

A system comprising:
computing circuitry configured to:
obtain subsampled real-time magnetic resonance imaging (MRI) data associated with multiple coils; and
reconstruct, using at least a residual convolutional recurrent neural network, the subsampled real-time MM data into a reconstructed MRI image, wherein the reconstructed MRI image reflects information acquired via the multiple coils and wherein the residual convolutional recurrent neural network comprises one or more bi-directional convolutional recurrent neural network cells.

### Claim 12 (depends on 11)

The system of claim 11, wherein the subsampled real-time MRI data is based on a sub-sampling mask configured to achieve an acceleration rate of ten times or more.

### Claim 13 (depends on 12)

The system of claim 12, wherein the residual convolutional recurrent neural network is trained using subsampled MM data generated based on a sub-sampling mask that is substantially similar to the sub-sampling mask on which the subsampled real-time MRI data is based.

### Claim 14 (depends on 11)

The system of claim 11, wherein the subsampled real-time MRI data is acquired without electrocardiogram (ECG) gating and breath-holding, and wherein the residual convolutional recurrent neural network is trained using retro cine data acquired based on ECG gating and breath-holding.

### Claim 15 (depends on 11)

The system of claim 11, wherein the residual convolutional recurrent neural network comprises one or more data consistency layers configured to ensure consistency between the reconstructed MRI image and the subsampled real-time MM data.

### Claim 16 (depends on 11)

The system of claim 11, wherein the residual convolutional recurrent neural network comprises one or more residual connections configured to add high-frequency details to the reconstructed MM image.

### Claim 17 (depends on 11)

The system of claim 11, wherein the residual convolutional recurrent neural network is configured to receive complex values associated with the subsampled real-time MRI data as a two-channel tensor.

### Claim 18 (depends on 11)

The system of claim 11, wherein the subsampled real-time MM data comprises respective subsampled real-time MRI k-data associated with multiple individual coils of an MRI scanner and wherein the computing circuitry being configured to reconstruct the subsampled real-time MRI data into the reconstructed MRI image comprises the computing circuitry being configured to:
reconstruct corresponding single-coil MM images based on the respective subsampled real-time MRI data associated with the multiple individual coils; and
generate the reconstructed MRI image based on a combination of the single-coil MM images.

### Claim 19 (depends on 18)

The system of claim 18, wherein the computing circuitry is configured to generate the reconstructed MRI images based on a root sum of squares or one or more coil sensitivity maps associated with the multiple individual coils of the MRI scanner.

### Claim 20 (depends on 18)

The system of claim 18, wherein the computing circuitry is configured to generate the single-coil MM images independently from each other based on the respective subsampled real-time MM data associated with the multiple individual coils.