Probablistic segmentation

Patent metadata

Field	Value
Publication number	US-12586200-B2
Application number	US-202017755151-A
Country	US
Kind code	B2
Filing date	Oct 20, 2020
Priority date	Oct 25, 2019
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

A machine learning system may be used for determining if a segmentation of a medical image is a reasonable segmentation in the sense that it is a segmentation that could be made by a human user and does not contain any impossible combinations of pixel values. The method is enhanced by user input to avoid the impossible combinations.

First claim

The invention claimed is:

1 . A computer-implemented method of determining an improved segmentation of a medical image, comprising: providing the medical image to a neural network based machine learning system that has been trained by means of training data sets comprising manually made segmentations, to determine if a segmentation is similar to the segmentations in the training dataset, wherein the medical image includes a set X of pixels or voxels having given pixel values x to a corresponding set of Y of segmentation values y; obtaining an initial segmentation of the medical image from the machine learning system in the form of a set Y of segmentation values y, said segmentation values y indicating for each pixel if it belongs to the structure, wherein the machine learning system approximates the unnormalized conditional probability distribution with a neural network; manually adjusting the initial segmentation by changing at least one of the segmentation values y to y1 and produce an adjusted set Y′ of segmentation values y′=[y1,y2]; and updating, by the machine learning system, the initial segmentation to produce an updated segmentation by optimizing the set Y of segmentation values based on the given pixel values x and the adjusted set Y′ of segmentation values, by solving the optimization problem.

2 . The method of claim 1 , further comprising repeating the manually adjusting and the updating, using the updated segmentation as the initial segmentation, until the updated segmentation is found to be good enough.

3 . The method of claim 1 , wherein the optimization problem is solved by means of simulated annealing.

4 . The method of claim 1 , wherein the optimization problem is solved by means of continuous approximation of the basic problem.

5 . The method of claim 1 , wherein the machine learning system is a generative adversarial network and the optimization is performed using a discriminator that has been trained to recognize a segmentation that are similar to those in a training dataset of segmentations that are or resemble manually made segmentations.

6 . The method of claim 1 , wherein the machine learning system is an auto encoder and the optimization is performed using an encoder and wherein the optimization is based on the score f(x,y)=pz(e(x,y)).

7 . A medical planning method including segmenting a medical image according to claim 1 and using the segmented medical image as a basis for planning.

8 . A computer program product for generating an improved segmentation of a medical image, the computer program product comprising computer readable code means which, when run in a computer will cause the computer to perform the following steps: receiving the medical image at a neural network based machine learning system that has been trained by means of training data sets comprising manually made segmentations, to determine if a segmentation is similar to the segmentations in the training dataset, wherein the medical image includes a set X of pixels or voxels having given pixel values x to a corresponding set of Y of segmentation values y; obtaining an initial segmentation of a structure in the image in the form of a set Y of segmentation values y, said segmentation values y indicating for each pixel if it belongs to the structure, wherein the machine learning system approximates the unnormalized conditional probability distribution with a neural network; receiving manual input for adjusting the initial segmentation by changing at least one of the segmentation values y to y1 and produce an adjusted set Y′ of segmentation values y′=[y 1 ,y 2 ] based on the manual input; updating, by the machine learning system the initial segmentation to produce an updated segmentation by optimizing the set Y of segmentation values based on the given pixel values x and the adjusted set Y′ of segmentation values by solving the optimization problem; and optionally repeating the receiving and updating, using the updated segmentation as the initial segmentation, until the updated segmentation is found to be good enough.

9 . A computer system for generating a segmentation of a medical image, comprising a processor, a program memory arranged to hold at least one computer program to be run in the processor, and a data memory, wherein the program memory holds a computer program product according to claim 8 .

Assignees

• Raysearch Laboratories Ab Publ
• Raysearch Lab Ab

Inventors

• Nordstrom Marcus
• Soderberg Jonas
• Lidberg David

Classifications

• G06T2207/30004

  Biomedical image processing · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/20076

  Probabilistic image processing · CPC title

• G06T2207/10081

  Computed x-ray tomography [CT] · CPC title