Functional imaging features from computed tomography images

Patent metadata

Field	Value
Publication number	US-12586181-B2
Application number	US-202117924382-A
Country	US
Kind code	B2
Filing date	May 12, 2021
Priority date	May 15, 2020
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

Methods, apparatus and computer readable media are provided for identifying functional features from a computed tomography (CT) image. The CT image may be a contrast-enhanced CT image or a non-contrast CT image. According to some examples, methods, apparatus and computer readable media are also provided for using machine learning to identify functional features from CT images. According to some examples, simulated functional image datasets such as simulated PET images or simulated SUV images are generated from a received CT image.

First claim

The invention claimed is:

1 . A method for identifying one or more functional features in a computed tomography (CT) image, the method comprising:
providing the CT image to a trained generator model, the generator model trained to translate an input CT image showing a target region of a subject to a positron emission tomography (PET) scan image indicating one or more functional features in the target region represented in the input CT image, wherein the generator model has been trained on a plurality of CT images and feedback from a discriminator; outputting, from the trained generator model, a PET scan image corresponding to the provided CT image, the PET scan image indicating one or more functional features in a target region represented in the provided CT image.

2 . A method according to claim 1 , wherein the CT image is a non-contrast CT (NCT) image.

3 . A method according to claim 1 , wherein the CT image is a contrast CT (CCT) image.

4 . A method according to claim 1 , wherein the PET scan image further indicates structural features in the target region.

5 . A method according to claim 1 , wherein the PET scan image comprises a visualisation indicating the one or more functional features in the target region represented in the CT image.

6 . A method according to claim 1 , wherein the trained generator model has been trained using a generative adversarial network.

7 . A method according to claim 1 , wherein the trained generator model comprises a trained image segmentation model.

8 . A method according to claim 1 , wherein the one or more functional features comprise one or more tumours, and wherein the method further comprises:
sampling, from the PET scan image, radiomic feature values for a set of radiomic features; providing the radiomic feature values to a trained classification model, the classification model trained to take as input a set of radiomic feature values and to output a classification indicating a predicted clinical outcome for the subject having the one or more tumours.

9 . A method according to claim 8 , wherein the classification model comprises a regression model or a random forest.

10 . A method according to claim 8 , wherein the predicted clinical outcome comprises locoregional tumour recurrence, distant metastasis, or death.

11 . A non-transitory computer-readable medium having stored thereon:
computer-readable code representative of a trained generator model or classification model; and instructions which, when executed by one or more processors, cause the one or more processors to implement a method according to claim 1 to identify one or more functional features in a computed tomography (CT) image.

12 . A computing apparatus for identifying functional features in a computed tomography (CT) image, the apparatus comprising:
one or more memory units; and one or more processors configured to execute instructions stored in the one or more memory units to perform the method of claim 1 .

13 . A method for training a generative adversarial network (GAN) to generate a simulated functional image dataset from a computed tomography (CT) image, the GAN comprising a generator network and a discriminator network, the method comprising:
receiving a training set comprising:
a plurality of CT images, each CT image showing a target region of a subject; and
a plurality of functional image datasets, each functional image dataset indicating functional features in a target region of a subject;
training the GAN, wherein training the GAN comprises:
training the generator network, using the plurality of CT images and feedback from the discriminator network, to generate simulated functional image datasets;
training the discriminator network, using the generated simulated functional image datasets and the plurality of functional image datasets, to classify received image datasets as simulated functional image datasets or genuine functional image datasets, and to provide feedback to the generator network; and
outputting a trained generator model to translate an input CT image to a simulated functional image dataset indicating one or more functional features in the target region shown in the input CT image.

14 . A method according to claim 13 , wherein the GAN is a cycle-GAN.

15 . A method according to claim 13 , wherein the plurality of functional image datasets comprises a plurality of PET scan images, PET-CT scan images, or SUV images, and wherein the trained generator model is to translate an input CT image to a simulated PET scan image, PET-CT scan image, or SUV image.

16 . A non-transitory computer-readable medium having instructions stored thereon which, when executed by one or more processors, cause the one or more processors to implement a method for training a GAN, for training a machine learning image segmentation algorithm, or for training a machine learning classification algorithm, according to claim 13 .

17 . A computing apparatus for training a GAN, for training a machine learning image segmentation algorithm, or for training a machine learning classification algorithm, the apparatus comprising:
one or more memory units; and one or more processors configured to execute instructions stored in the one or more memory units to perform the method of any of claim 13 .

18 . A method according to claim 1 , wherein the one or more functional features in a target region indicate infected, inflamed or cancerous tissue.

19 . A method according to claim 13 , wherein the one or more functional features in a target region indicate infected, inflamed or cancerous tissue.

Assignees

• Univ Oxford Innovation Ltd

Inventors

• Lee Regent
• Chandrashekar Anirudh
• Grau Vicente

Classifications

• G06T2207/30096

  Tumor; Lesion · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/10104

  Positron emission tomography [PET] · CPC title

• G06T2207/10081

  Computed x-ray tomography [CT] · CPC title

• G06V20/698

  Matching; Classification · CPC title