# US 2025/0155536 A1 · Artificial intelligence distortion correction for magnetic resonance echo planar imaging

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

- Number: US 2025/0155536 A1
- Document ID: US20250155536A1
- Status: Published application
- Kind: A1
- Application number: 18/507,516
- Filed: 2023-11-13
- Published: 2025-05-15
- Earliest claimed: 2023-11-13
- Assignees: Siemens Healthineers AG
- Applicants: Siemens Healthineers AG
- Inventors: Shujaat Khan; Mahmoud Mostapha; Mariappan S. Nadar; Thorsten Feiweier; Bryan Clifford; Mengwei Ren
- Web page: https://patentgrep.com/patent/US20250155536A1

## CPC

- G01R 33/56554: {caused by acquiring plural, differently encoded echo signals after one RF excitation, e.g. correction for readout gradients of alternating polarity in EPI}
- 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)}
- G06T 12/20: Inverse problem, i.e. transformations from projection space into object space
- G06T 2210/41: Medical
- G06T 2211/441: AI-based methods, deep learning or artificial neural networks
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

For distortion correction in magnetic resonance (MR) echo planar imaging (EPI), a combination of supervised and unsupervised training provides deep learning with generalized applicability across different use cases as well as sufficient training data (e.g., less ground truth data is needed due to the inclusion of unsupervised learning). The neural network for displacement estimation may include deformable convolution and/or layers for a non-diffeomorphic displacement field. The network architecture features and/or combined supervised and unsupervised learning may be used together or individually.

## Claims

### Claim 1

A method of machine training for magnetic resonance (MR) distortion correction in echo planar imaging, the method comprising:
acquiring first MR training data including ground truth;
machine training a neural network to estimate MR distortion fields using the first MR training data in supervised training and differences of first images from second images in unsupervised training, the first images having opposing phase encoding directions than the second images; and
storing the neural network as machine trained.

### Claim 2 (depends on 1)

The method of claim 1 wherein machine training comprises machine training with a joint loss for (a) a difference of the estimated MR distortion fields from the ground truth and for (b) the difference of the first and second images with the opposing phase encoding directions.

### Claim 3 (depends on 1)

The method of claim 1 wherein the first MR training data comprises k-space data for a blip-up and blip-down MR echo planar acquisitions, the blip-up and blip-down MR echo planar acquisitions having the opposing phase encoding directions;
further comprising:
reconstructing first blip-up images and first blip-down images;
inputting the first blip-up images and first blip-down images to the neural network; and
applying estimated MR distortion fields to the first blip-up images and/or the first blip-down images, the application resulting in the first and second images.

### Claim 4 (depends on 3)

The method of claim 3 further comprising estimating B0 distortion with static field correction and correcting the first blip-up images and the first blip-down images with the static field correction prior to inputting to the neural network.

### Claim 5 (depends on 3)

The method of claim 3 wherein applying the estimated MR distortion fields comprises applying an unwarping followed by Jacobian modulation for intensity correction.

### Claim 6 (depends on 3)

The method of claim 3 wherein applying the estimated MR distortion field comprises intensity correction by a machine learning model.

### Claim 7 (depends on 1)

The method of claim 1 wherein machine training comprises learning parameters of adaptive mask and adaptive offset-based deformable convolution blocks without non-deformable convolution blocks prior to an output layer.

### Claim 8 (depends on 1)

The method of claim 1 wherein machine training comprises learning parameters of non-diffeomorphic layers for velocity prediction and/or integration of the velocity prediction as the estimated MR distortion field.

### Claim 9 (depends on 1)

The method of claim 1 wherein machine training comprises learning parameters for deformable and non-deformable convolutions layers with multi-resolution loss terms.

### Claim 10 (depends on 1)

The method of claim 1 wherein the first MR training data comprises k-space data for a blip-up and blip-down MR echo planar acquisitions, the blip-up and blip-down MR echo planar acquisitions having the opposing phase encoding directions;
further comprising:
first reconstructing first blip-up images and first blip-down images from the k-space data;
inputting the first blip-up images and first blip-down images to the neural network, the neural network outputting the MR distortion fields; and
second reconstructing second blip-up images and second blip-down images from the k-space data and the estimated MR distortion fields;
wherein machine training comprises machine training where the first images and second images are the second blip-up images and the second blip-down images, respectively.

### Claim 11 (depends on 10)

The method of claim 10 wherein second reconstructing comprises second reconstructing with a machine learning model, the k-space data and the estimated MR distortion fields input to the machine learning model, the machine learning model outputting the second blip-up images and the second blip-down images.

### Claim 12 (depends on 11)

The method of claim 11 wherein machine training comprises training the neural network and parameters of the machine learning model in a fine tuning with joint optimization.

### Claim 13 (depends on 11)

The method of claim 11 wherein machine training comprises end-to-end learning of the neural network, the neural network including the machine learning model.

### Claim 14 (depends on 1)

The method of claim 1 wherein acquiring comprises augmenting through change in orientation, acceleration factor, noise level, sign of displacement fields, and/or alignment.

### Claim 15 (depends on 1)

The method of claim 1 wherein acquiring comprises acquiring the first MR training data for different body regions and/or through simulation.

### Claim 16

A system for distortion correction in magnetic resonance echo planar imaging, the system comprising:
a medical scanner configured to scan a patient;
an image processor configured to reconstruct a representation of the patient from the scan, configured to estimate a displacement field with a machine-trained neural network, the neural network comprising at least one deformable convolution layer, and configured to correct the representation with the displacement field; and
a display configured to display an image of the patient generated from the representation as corrected.

### Claim 17 (depends on 16)

The system of claim 16 wherein the image processor is configured to correct by the reconstruction with the displacement field as an input to the reconstruction.

### Claim 18

A system for distortion correction in magnetic resonance echo planar imaging, the system comprising:
a medical scanner configured to scan a patient;
an image processor configured to reconstruct a representation of the patient from the scan, configured to estimate a diffeomorphic displacement field with a machine-trained neural network, the neural network comprising at least one velocity estimation layer, and configured to correct the representation with the displacement field; and
a display configured to display an image of the patient generated from the representation as corrected.

### Claim 19 (depends on 18)

The system of claim 18 wherein the image processor is configured to correct by the reconstruction with the displacement field as an input to the reconstruction.
