Granted Patent B2

US 12,511,754 · App. 18/162,852 · Granted Dec 30, 2025

Inventors: Marc Vornehm (Erlangen, DE); Daniel Giese (Erlangen, DE); Jens Wetzl (Erlangen, DE); Elisabeth Preuhs (Erlangen, DE)

Assignee: SIEMENS HEALTHINEERS AG

G06T7/11A61B5/055G06T2207/20084

Abstract

Claims (43)

1 . A method for training an artificial neural network for determining scar segmentation on contrast agent enhanced medical images with a reduced contrast agent dose, the method comprising:

acquiring a first set of medical images after a low-dose of a contrast agent has been applied;

receiving input training data, the input training data including the first set of medical images of objects to be segmented acquired after the application of the low-dose of the contrast agent, the low-dose of the contrast agent including less of the contrast agent than a full-dose of the contrast agent;

acquiring a second set of medical images after an additional dose of the contrast agent has been applied such that the low-dose and the additional dose equal the full-dose;

receiving output training data, the output training data including reference scar segmentation masks determined from the second set of medical images of the objects acquired after the application of the full-dose of the contrast agent; and

training the artificial neural network using the input training data and the output training data to determine the scar segmentation on the contrast agent enhanced medical images with the reduced contrast agent dose.

2 . The method of claim 1 , further comprising:

determining the reference scar segmentation masks via a full width at half maximum method on the second set of medical images.

3 . A non-transitory computer-readable medium including instructions which, when executed by a processing unit of a medical imaging device, cause the medical imaging device to perform the method of claim 2 .

4 . The method of claim 1 , wherein

at least one of the input training data or the output training data comprise real medical images and simulated medical images.

5 . The method of claim 4 , wherein a number of the real medical images is increased by at least one of oversampling or augmentation.

6 . The method of claim 5 , further comprising:

performing an image registration of the first set of medical images and the second set of medical images to transform the first set of medical images and the second set of medical images into a common coordinate system.

7 . The method of claim 1 , further comprising:

performing an image registration of the first set of medical images and the second set of medical images to transform the first set of medical images and the second set of medical images into a common coordinate system.

8 . An imaging method for determining a scar area of an object, the method comprising:

acquiring a medical image of the object after a low-dose of a contrast agent has been applied to the object, the low-dose of contrast agent includes less of the contrast agent than a full-dose of the contrast agent; and

determining a scar segmentation mask by applying a trained artificial neural network to the medical image, the trained artificial neural network being trained according to the method of claim 1 .

9 . A non-transitory computer-readable medium including instructions which, when executed by a processing unit of a medical imaging device, cause the medical imaging device to perform the method of claim 1 .

10 . The method of claim 1 , wherein the scar segmentation is a segmentation of a myocardial scar.

11 . A computer-implemented method for determining scar segmentation, the method comprising:

receiving a medical image of an object to be segmented acquired after an application of a low-dose of a contrast agent, the low-dose of the contrast agent including less of the contrast agent than a full-dose of the contrast agent;

determining borders of the object to be segmented on the medical image; and

determining a scar segmentation mask by applying a trained artificial neural network to the medical image after the determining the borders of the object to be segmented on the medical image, the determining the scar segmentation by the trained artificial neural network being restricted to the borders.

12 . The method of claim 11 , wherein the trained artificial neural network is trained for scar segmentation on contrast agent enhanced medical images with a reduced contrast agent dose.

13 . The method of claim 11 , wherein

the contrast agent is a gadolinium-based contrast agent, and

the medical image is acquired via MR imaging with late gadolinium enhancement.

14 . The method of claim 11 , wherein the low-dose comprises 20% to 90% of an amount of the contrast agent of the full-dose.

15 . The method of claim 11 , wherein a U-Net based artificial neural network is used.

16 . A system for determining a scar area of an object, the system comprising:

a first interface configured to receive a medical image of the object to be segmented acquired after an application of a low-dose of a contrast agent, the low-dose of the contrast agent including less of the contrast agent than a full-dose of the contrast agent;

a processing unit configured to determine a scar segmentation mask by applying a trained artificial neural network, the trained artificial neural network being trained by

receiving input training data, the input training data including a first set of medical images of objects to be segmented acquired after an application of the low-dose of the contrast agent,

receiving output training data, the output training data including reference scar segmentation masks determined from a second set of medical images of the objects acquired after an application of the full-dose of the contrast agent, and

training an artificial neural network using the input training data and the output training data to obtain the trained artificial neural network to determine scar segmentation on contrast agent enhanced medical images with a reduced contrast agent dose; and

a second interface configured to output the scar segmentation mask.

17 . A computer-implemented method for determining scar segmentation, the method comprising:

receiving a medical image of an object to be segmented acquired after an application of a low-dose of a contrast agent, the low-dose of the contrast agent including less of the contrast agent than a full-dose of the contrast agent; and

determining a scar segmentation mask by applying a trained artificial neural network to the medical image,

wherein the low-dose of the contrast agent comprises 50% to 75% of an amount of the contrast agent of the full-dose.

18 . The method of claim 17 , wherein the low-dose comprises 60% to 70% of the amount of the contrast agent of the full-dose.

Assignments (2)

Priority Claims (1)

Continuity (1)

References Cited (43)