Neural computed tomography reconstruction

Patent metadata

Field	Value
Publication number	US-2025104299-A1
Application number	US-202318728259-A
Country	US
Kind code	A1
Filing date	Jan 13, 2023
Priority date	Jan 13, 2022
Publication date	Mar 27, 2025
Grant date	—

Abstract

Methods and systems that pertain to an image reconstruction of motion-corrupted images are disclosed. In some embodiments of the disclosed technology, an image reconstruction method includes obtaining an initial estimate of object boundaries from motion-corrupted images, creating an implicit representation of the motion-corrupted images, updating the implicit representation of the motion-corrupted images using acquired imaging data to generate an updated implicit representation of the motion-corrupted images, and converting the updated implicit representation of the motion corrupted images to an explicit set of motion-corrected images.

First claim

1 - 6 . (canceled)

7 . An image reconstruction method, comprising:
obtaining an implicit signed distance function image of an object; updating the implicit signed distance function image to produce an updated implicit signed distance function image that matches acquired imaging data; and converting the updated implicit signed distance function image into an explicit signed distance function image of the object.

8 . The method of claim 7 , wherein obtaining the implicit signed distance function image of the object includes: generating, from a filtered back projection image of the object, a binary classification image of the object; and converting the binary classification image into the implicit distance function image.

9 . The method of claim 8 , wherein the binary classification image is generated by: identifying the object in an image reconstructed through a filtered back projection by performing an image segmentation; and
encoding, using a signed distance function, a segmentation image of the object that is obtained by performing the image segmentation.

10 . The method of claim 7 , wherein the implicit signed distance function image is represented by a neural network.

11 . The method of claim 7 , wherein the acquired imaging data includes acquired sinogram data.

12 . The method of claim 11 , wherein the acquired sinogram data includes data that is acquired by a computed tomography (CT) scanner.

13 . The method of claim 7 , wherein converting the updated implicit signed distance function image into the explicit signed distance function image of the object includes sampling the updated implicit signed distance function image over a grid at a predetermined spatial resolution to generate a spatiotemporal intensity image of the object.

14 . An image reconstruction method, comprising:
performing an image segmentation on an initial reconstructed image to obtain binary classification images for identifying an object of interest in the initial reconstructed image; converting the binary classification images into a first set of signed distance function (SDF) images configured to explicitly represent the object of interest; converting the first set of signed distance function into a second set of SDF images configured to implicitly represent the object of interest; training the second set of SDF images to match acquired sinogram data; and converting the trained second set of SDF images into a third set of SDF images configured to explicitly represent the object of interest.

15 . The method of claim 14 , further comprising performing a filtered back projection (FBP) on the object of interest to obtain the initial reconstructed image of the object.

16 . The method of claim 14 , wherein the second set of SDF images are represented by a neural network.

17 . The method of claim 16 , wherein training the second set of SDF images comprises: converting the second set of SDF images to a spatiotemporal intensity map and projecting to a sinogram domain to determine a sinogram estimate;
determining a difference between the sinogram estimate and the acquired sinogram data; and updating the second set of SDF images based on the difference by backpropagating the difference through neural representations of the SDF images.

18 . The method of claim 14 , wherein the binary classification images represent pixels of each label across time.

19 . The method of claim 14 , wherein the acquired sinogram data includes data that is acquired by a computed tomography (CT) scanner.

20 . The method of claim 14 , wherein the acquired sinogram data includes an attenuation accumulated by x-rays traversing from a light source to a specific detector position associated with the object of interest.

21 . The method of claim 14 , wherein converting the trained second set of SDF images into the third set of SDF images includes sampling the trained second set of SDF images over a grid at a predetermined spatial resolution.

22 . The method of claim 14 , further comprising creating occupancy images by binarizing the third set of SDF images.

23 . An image reconstruction method, comprising:
performing a first explicit-to-implicit representation conversion to obtain an implicit signed distance function image of an object by: obtaining an initial binary classification image of the object; and converting the binary classification image into the implicit distance function image; performing a first training operation on the implicit signed distance function image by updating the implicit signed distance function image to produce a first updated implicit signed distance function image that matches an acquired sinogram; performing a first implicit-to-explicit representation conversion from the updated implicit signed distance function image to generate a first binary image; performing a second explicit-to-implicit representation conversion by feeding the first binary image as the binary classification image of the object; and performing a second training operation and a second implicit-to-explicit representation conversion to generate a second binary image.

24 . The method of claim 23 , wherein the initial binary classification image of the object is obtained from a filtered back projection image of the object.

25 . The method of claim 23 , wherein the second training operation includes updating an implicit signed distance function image obtained by performing the second explicit-to-implicit representation conversion to produce a second updated implicit signed distance function image that matches an acquired sinogram.

26 . The method of claim 23 , wherein the first implicit-to-explicit representation conversion includes sampling the updated implicit signed distance function image over a grid at a predetermined spatial resolution.

27 . The method of claim 26 , wherein the first binary image is generated by binarizing the sampled updated implicit signed distance function image.

28 . The method of claim 23 , wherein the initial binary classification image is generated by:
identifying the object in an image reconstructed through a filtered back projection by performing an image segmentation; and encoding, using a signed distance function, a segmentation image of the object that is obtained by performing the image segmentation.

29 . (canceled)

Assignees

• Univ California

Inventors

• Contijoch Francisco
• Gupta Kunal
• Colvert Brendan

Classifications

• G06T12/30

  Image post-processing, e.g. metal artefact correction · CPC title

• G06T12/20 Primary

  Inverse problem, i.e. transformations from projection space into object space · CPC title

• G06T2210/41

  Medical · CPC title

• G06T2211/441

  AI-based methods, deep learning or artificial neural networks · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title