# US 2024/0005504 A1 · Standardizing images of anatomical structures for analysis by machine learning systems

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

- Number: US 2024/0005504 A1
- Document ID: US20240005504A1
- Status: Published application
- Kind: A1
- Application number: 18/344,128
- Filed: 2023-06-29
- Published: 2024-01-04
- Earliest claimed: 2022-07-01
- Assignees: Peek Health S.A.
- Applicants: Peek Health S.A.
- Inventors: João Pedro de Araújo Ribeiro; Sara Isabel Azevedo Da Silva; Jaime Frederico Bastos De Campos
- Web page: https://patentgrep.com/patent/US20240005504A1

## CPC

- G06T 7/0014: {using an image reference approach}
- G06T 12/30: Image post-processing, e.g. metal artefact correction
- G06T 7/337: {involving reference images or patches}
- G06T 15/00: Three-dimensional [3D] image rendering
- G06T 2200/04: involving 3D image data
- G06T 2207/10116: X-ray image
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30008: Bone
- G06T 2211/441: AI-based methods, deep learning or artificial neural networks
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A computer-implemented method includes receiving a real two-dimensional medical image of a patient, processing the real two-dimensional medical image with a computer-implemented generator configured to transfer the real two-dimensional medical image from a real image domain into a synthetic image domain, and analyzing the two-dimensional medical image transferred into the synthetic image domain with a machine-learning model. The machine-learning model was trained with a collection of synthetic two-dimensional medical images generated from real three-dimensional medical images.

## Claims

### Claim 1

A computer-implemented method, the method comprising:
receiving a real two-dimensional medical image of a patient;
processing the real two-dimensional medical image with a computer-implemented generator configured to transfer the real two-dimensional medical image from a real image domain into a synthetic image domain; and
analyzing the two-dimensional medical image transferred into the synthetic image domain with a machine-learning model, wherein the machine-learning model was trained with a collection of synthetic two-dimensional medical images generated from real three-dimensional medical images.

### Claim 2 (depends on 1)

The method of claim 1, wherein the machine-learning model is a Generative Adversarial Network.

### Claim 3 (depends on 1)

The method of claim 1, wherein the generator is configured to perform one or more of the following:
a) align a plane of the anatomy in the real two-dimensional medical image with a standard image plane; or
b) select a standardized field of view in the real two-dimensional medical image; or
c) perform rigid transformation of an anatomical structure in the real two-dimensional medical image with respect to a standard image plane; or
d) perform rigid transformation of an anatomical structure in a first real two-dimensional medical image with respect to an anatomical structure in a second real two-dimensional medical image.

### Claim 4 (depends on 1)

The method of claim 1, wherein the generator includes a machine-learning model.

### Claim 5 (depends on 1)

The method of claim 1, further comprising training the machine-learning model of the generator to transfer the real two-dimensional medical image from the real image domain into the synthetic image domain.

### Claim 6 (depends on 5)

The method of claim 5, wherein training the machine-learning model of the generator comprises:
feeding back a penalty from a discriminator that is trained to distinguish between real-looking synthetic radiographs and synthetic-looking real radiographs into the generator; and
adapting the machine-learning model of the generator to reduce the penalty.

### Claim 7 (depends on 5)

The method of claim 5, further comprising training the discriminator.

### Claim 8 (depends on 1)

The method of claim 1, wherein the real three-dimensional medical images were projected into two-dimensional space to generate the synthetic two-dimensional medical images using an intensity-based decomposition analysis.

### Claim 9

A computer-implemented method for training a generator that is implemented in a computer-implemented artificial neural network, wherein the generator is trained to transfer real two-dimensional medical images from a real image domain into a synthetic image domain, the method comprising:
receiving a plurality of three-dimensional medical images of different patients and a plurality of two-dimensional medical radiographs of different patients;
using the generator, creating a plurality of synthetic two-dimensional radiographs based on the received plurality of three-dimensional medical images;
using a computer-implemented discriminator, discriminating the synthetic two-dimensional radiographs from the synthetic-looking two-dimensional radiographs; and
in response to the discriminator successfully discriminating between the real-looking synthetic radiographs and the synthetic-looking real radiographs, adapting the generator to reduce an ability of the discriminator to discriminate therebetween.

### Claim 10 (depends on 9)

The method of claim 9, wherein the computer-implemented artificial neural network is a Generative Adversarial Network.

### Claim 11 (depends on 9)

The method of claim 9, wherein adapting the generator comprises feeding back a penalty to the generator and adapting the weights of the generator to reduce the penalty.

### Claim 12 (depends on 9)

The method of claim 9, wherein adapting the generator comprises one or more of:
a) training the generator to align a plane of the anatomy in the real two-dimensional medical image with a standard image plane; or
b) training the generator to select a standardized field of view in the real two-dimensional medical image; or
c) training the generator for rigid transformation of an anatomical structure in the real two-dimensional medical image with respect to a standard image plane; or
d) training the generator for rigid transformation of an anatomical structure in a first real two-dimensional medical image with respect to an anatomical structure in a second real two-dimensional medical image.

### Claim 13 (depends on 9)

The method of claim 9, wherein creating the plurality of synthetic two-dimensional radiographs based on the received plurality of three-dimensional medical images comprises projecting the real three-dimensional medical images into two-dimensional space to generate the synthetic two-dimensional medical images using an intensity-based decomposition analysis.

### Claim 14 (depends on 9)

The method of claim 9, further comprising training the discriminator.

### Claim 15 (depends on 14)

The method of claim 14, wherein training the discriminator comprises:
inputting plane-projection corrected radiographs into a discriminator;
inputting plane-projection uncorrected radiographs into a discriminator; and
in response to the discriminator failing to discriminate between the plane-projection corrected radiographs and the plane-projection uncorrected radiographs, adapting the discriminator to improve an ability of the discriminator to discriminate therebetween.

### Claim 17 (depends on 4)

The method of claim 4, wherein the generator includes a Generative Adversarial Network

### Claim 18 (depends on 6)

The method of claim 6 wherein the method further comprises:
inputting real-looking synthetic radiographs into a discriminator;
inputting synthetic-looking real radiographs into a discriminator; and
in response to the discriminator successfully identifying the real-looking synthetic radiographs and the synthetic-looking real radiographs, adapting the machine-learning model of the generator to reduce an ability of the discriminator to discriminate therebetween.

### Claim 19 (depends on 7)

The method of claim 7, wherein training the discriminator comprises:
inputting real-looking synthetic radiographs and synthetic radiographs into a discriminator;
inputting synthetic-looking real radiographs and real radiographs into a discriminator; and
in response to the discriminators failing to discriminate between the real-looking radiographs and the synthetic-looking radiographs, adapting the discriminators to improve an ability of the discriminator to discriminate therebetween.
