Consistent 3D rendering in medical imaging

Patent metadata

Field	Value
Publication number	US-10957098-B2
Application number	US-202016789586-A
Country	US
Kind code	B2
Filing date	Feb 13, 2020
Priority date	Mar 10, 2017
Publication date	Mar 23, 2021
Grant date	Mar 23, 2021

Abstract

For three-dimensional rendering, a machine-learnt model is trained to generate representation vectors for rendered images formed with different rendering parameter settings. The distances between representation vectors of the images to a reference are used to select the rendered image and corresponding rendering parameters that provides a consistency with the reference. In an additional or different embodiment, optimized pseudo-random sequences are used for physically-based rendering. The random number generator seed is selected to improve the convergence speed of the renderer and to provide higher quality images, such as providing images more rapidly for training compared to using non-optimized seed selection.

First claim

We claim:

1. A method for three-dimensional rendering in a rendering system, the method comprising:
three-dimensionally rendering a reference image with path tracing using a first number of samples per pixel and a first random number generator seed;
three-dimensionally rendering other images with the path tracing using different ones of other random number generator seeds, the three-dimensionally rendering for each of the other images using a second number or less than the second number of samples per pixel, the second number less than the first number by at least a factor of ten;
measuring errors between each of the other images and the reference image;
identifying one of the other random number generator seeds with a lessor error of the errors than the others of the other random number generator seeds; and
configuring the rendering system for the path tracing with the identified one of the other random number generator seeds.

2. The method of claim 1 further comprising scanning a patient with a medical imaging system, and three-dimensionally rendering, by the rendering system as configured, an image from output of the scanning with the path tracing using the second number or less than the second number of the samples per pixel and the identified one of the other random number generator seeds.

3. The method of claim 2 further comprising performing the measuring and identifying as part of the three-dimensionally rendering of the image by the rendering system as configured being performed as interactive rendering from the output of the scanning.

4. The method of claim 1 wherein measuring the errors comprises measuring a number of the samples per pixel where the error reaches a threshold.

5. The method of claim 1 wherein three-dimensionally rendering the other images comprises emulating coherent scattering.

6. The method of claim 1 further comprising further rendering additional images as training renderings using the identified one of the other random number generator seeds, and training a neural network with deep learning to identify similar images with an architecture using a representation network connected to a decision network, the deep learning using the training renderings generated with the identified one of the other random number generator seeds.

Assignees

• Siemens Healthcare Gmbh

Inventors

• Petkov Kaloian
• Liu Chen
• Miao Shun
• Sudarsky Sandra
• Yu Daphne
• Mansi Tommaso

Classifications

• G06T12/30

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

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06T15/08 Primary

  Volume rendering · CPC title