Systems and methods for calcium-free computed tomography angiography

Patent metadata

Field	Value
Publication number	US-12530765-B2
Application number	US-202017428937-A
Country	US
Kind code	B2
Filing date	Feb 18, 2020
Priority date	Feb 19, 2019
Publication date	Jan 20, 2026
Grant date	Jan 20, 2026

Abstract

A method of analyzing computed tomography (CT) images comprises receiving an initial CT image of an object, identifying calcium-free regions and a calcified region in the initial CT image of the object, generating a calcium-free image patch, and applying the calcium-free image patch to the initial CT image patch to produce a final CT image. The initial CT image shows a calcium deposit and a target structure in the object. The calcified region in the initial CT image shows the calcium deposit in the object obscuring a portion of the target structure. The calcium-free regions show the remaining portions of the target structure. The calcium-free image patch corresponds to the calcified region in the initial CT image. The final CT image shows the calcium-free image patch and the calcium-free region from the initial CT image. The calcium-free image patch is generated and applied using a convolutional neural network.

First claim

What is claimed is:

1 . A method of analyzing computed tomography (CT) images, comprising:
receiving, by a processing device, an initial CT image of a portion of an object, the initial CT image showing one or more calcium deposits in the portion of the object and at least some of a target structure in the portion of the object; identifying, using the processing device, a calcified region in the initial CT image and one or more calcium-free regions in the initial CT image, the calcified region in the initial CT image showing the calcium deposits in the portion of the object; applying an inpainting mask corresponding to the calcified region in the initial CT image to the initial CT image to produce a modified CT image, the modified CT image showing the inpainting mask and the one or more calcium-free regions; inputting the modified CT image into a neural network that is trained to:
generate a calcium-free image patch from the inpainting mask in the modified CT image, the calcium-free image patch corresponding to the calcified region in the initial CT image; and
apply the calcium-free image patch to modified CT image to produce a final CT image, the final CT image showing the calcium-free image patch and the calcium-free regions from the initial CT image; and
receiving the final CT image from the neural network, wherein: the neural network is convolutional neural network that includes a contracting path, an expanding path, and a transition stage between the contracting path and the expanding path, the contracting path being configured to hierarchically extract features of the one or more calcium-free regions from the modified CT image, the expanding path being configured generate the calcium-free image patch from the extracted features and merge the calcium-free image patch with the calcium-free regions; the expanding path includes:
(i) a first stage with one or more 1×1×1 convolution layers and one or more 3×3×3 deconvolution layers;
(ii) a second stage with a dense expanding block, and one or more 3×3×3 deconvolution layers;
and (iii) a third stage with one or more 1×1×1 convolution layers; and
the output of the transition stage is input into the first stage of the expanding path, the output of the first stage of the expanding path is input into the second stage of the expanding path, the output of the second stage of the expanding path is input into the third stage of the expanding path, and the output of the third stage of the expanding path is the final CT image.

2 . The method of claim 1 , wherein the target structure includes at least a first portion and a second portion, the second portion being adjacent to and continuous with the first portion.

3 . The method of claim 2 , wherein one of the one or more calcium-free regions shows the first portion of the target structure, and wherein the calcified region is adjacent to the one calcium-free region showing the first portion and does not show the adjacent second portion of the target structure, such that the second portion of the target structure is obscured or is not visible in the initial CT image.

4 . The method of claim 3 , wherein the generated calcium-free image patch shows the second portion of the target structure such that the final CT image shows the second portion of the target structure adjacent to and continuous with the first portion of the target structure.

5 . The method of claim 1 , further comprising determining whether a value of a property of each of a plurality of voxels in the initial CT image is greater than a threshold value, the value of the property of a respective voxel representing a value of a property of a material within the portion of the object being represented by the respective voxel.

6 . The method of claim 5 , wherein the value of the property of the material being represented by the respective voxel is indicative of the type of material being represented by the respective voxel.

7 . The method of claim 6 , wherein the value of the property of the respective voxel being greater than the threshold value indicates that the respective voxel represents calcium.

8 . The method of claim 5 , wherein the property of the material is a radiodensity of the material measured in Hounsfield units (HU), and wherein the threshold value of the property of the material is 700 HU.

9 . The method of claim 5 , wherein the property of each respective voxel is a brightness of each respective voxel on a gray scale.

10 . The method of claim 1 , wherein the identifying the calcified region in the CT image comprises generating a probability map indicating a probability of each voxel in the CT image representing calcium, and wherein the target structure is an arterial lumen.

11 . The method of claim 1 , wherein one or more of the dense contracting block, the dense transition block, and the dense expanding block is a 12-layer dense block with a growth rate of four, or wherein the transition layer includes a batch normalization, a rectified linear unit (ReLU) activation, and a 3×3×3 convolutional layer.

12 . The method of claim 1 , wherein the output of the first stage of the contracting path is concatenated with the output of the second stage of the expanding path and input into the third stage of the expanding path, or wherein the output of the second stage of the contracting path is concatenated with the output of the first stage of the expanding path and input into the second stage of the expanding path.

13 . The method of claim 1 , wherein the neural network is trained using a plurality of images with no calcified regions prior to inputting the modified CT image.

14 . The method of claim 1 , wherein applying the inpainting mask includes modifying a brightness of at least each voxel in the calcified region of the initial CT image to be black.

15 . The method of claim 1 , further comprising:
identifying, using the processing device, an additional calcified region in the final CT image and one or more calcium-free regions in the final CT image, the one or more calcium-free regions in the final CT image including the calcium-free image patch and the one or more calcium-free regions from the initial CT image; applying an additional inpainting mask corresponding to the additional calcified region in the final CT image to the final CT image to produce a modified final CT image, the modified final CT image showing the inpainting mask and the one or more calcium-free regions from the final CT image; inputting the modified final CT image into the neural network that is trained to:
generate an additional calcium-free image patch from the additional inpainting mask in the modified final CT image, the additional calcium-free image patch corresponding to the additional calcified region in the final CT image; and
apply the additional calcium-free image patch to modified final CT image to produce an updated final CT image, the updated final CT image showing the calcium-free image patch and the calcium-free regions from the final CT image; and
receiving the updated final CT image from the neural network.

16 . The method of claim 1 , further comprising determining, using the processing device, whether a value of a property of each of a plurality of voxels in the initial CT image is greater than a threshold value, the value of the property of a respective voxel representing a value of a property of a material within the portion of the object being represented by the respective voxel, the value of the property of a respective voxel being greater than the threshold value indicating that the respective voxel r

Assignees

• Cedars Sinai Medical Center

Inventors

• Xie Yibin
• Li Debiao
• Dey Damini
• Chen Yuhua

Classifications

• G06T2207/30101

  Blood vessel; Artery; Vein; Vascular · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/10081

  Computed x-ray tomography [CT] · CPC title

• G06T7/136

  involving thresholding · CPC title