Probabilistic refinement of model-based segmentation

Patent metadata

Field	Value
Publication number	US-9659364-B2
Application number	US-201113581512-A
Country	US
Kind code	B2
Filing date	Feb 14, 2011
Priority date	Mar 11, 2010
Publication date	May 23, 2017
Grant date	May 23, 2017

Abstract

A system for segmenting current diagnostic images includes a workstation ( 30 ) which segments a volume of interest in previously generated diagnostic images of a selected volume of interest generated from a plurality of patients. One or more processors ( 32 ) are programmed to register the segmented previously generated images and merge the segmented previously generated images into a probability map that depicts a probability that each voxel represents the volume of interest ( 24 ) or background ( 26 ) and a mean segmentation boundary ( 40 ). A segmentation processor ( 50 ) registers the probability map with a current diagnostic image ( 14 ) to generate a transformed probability map ( 62 ). A previously-trained classifier ( 70 ) classifies voxels of the diagnostic image with a probability that each voxel depicts the volume of interest or the background. A merge processor ( 80 ) merges the probabilities from the classifier and the transformed probability map. A segmentation boundary processor ( 84 ) determines the segmentation boundary for the volume of interest based on the current image based on the merge probabilities.

First claim

The invention claimed is:

1. A system for segmenting current diagnostic images comprising:
one or more workstations which segment a volume of interest in previously generated diagnostic images of a selected volume of interest generated from a plurality of patients;
one or more processors programmed to:
register the segmented previously generated images, and
merge the segmented previously generated images into a probability map which depicts a probability that each voxel represents the volume of interest, a probability that each voxel represents background, and a mean segmentation boundary;

a segmentation processor which registers the probability map with a current diagnostic image of the volume of interest in a current patient to generate a transformed probability map, the segmentation processor being programmed to register the probability map with the current image by performing the steps of:
registering the mean segmentation boundary to the volume of interest of one of the current image and a model registered to the current image;
determining a transform by which the mean segmentation boundary was transformed to be registered to the current image or model;
transforming the probability map with the determined transform to generate the transformed probability map; and

a segmentation boundary processor which determines a segmentation boundary for the volume of interest based on the transformed probability map.

2. The system according to claim 1 , further including:
a classifier which has been previously trained to classify diagnostic images of the volume of interest based on characteristics of the diagnostic image to determine a probability that at least selected voxels of the current image depict the volume of interest, and the background; and
a merge processor or computer routine which merges the probabilities from the classifier and the probabilities of the corresponding voxels from the transformed probability map as registered with the current image.

3. The system according to claim 2 , further including:
a user input device by which a relative weighting with which of the classifier and probability map probabilities are merged.

4. The system according to claim 2 , wherein the segmentation boundary processor further determines the segmentation boundary for the volume of interest in the current image based on the merged probabilities, and further including:
a processor or computer routine which combines the determined segmentation boundary with the current image.

5. The system according to claim 1 , further including:
a thresholding device or processor which assigns voxels of the current image which, in the transformed probability map, have probabilities above a threshold value to one of the volume of interest and the background and if the voxels have probabilities below the threshold to the other of the volume of interest and the background.

6. The system according to claim 5 , wherein the segmentation boundary processor further determines the segmentation boundary from an interface between the voxels assigned to the volume of interest and the voxels assigned to the background, and further including:
an image processor programmed to combine the segmentation boundary with the current image;
a display on which the segmented current image is displayed; and
an input device by which a user adjusts a probability threshold to adjust the segmentation boundary and adjust the segmentation of the displayed segmented current image.

7. The system according to claim 1 , further including:
a medical database in which the segmented current image is stored; and
a radiation therapy planning system which uses the segmented current image to perform a radiation therapy planning process.

8. A system for segmenting current diagnostic images comprising:
one or more workstations which segment a volume of interest in previously generated diagnostic images of a selected volume of interest generated from a plurality of patients;
one or more processors programmed to:
register the segmented previously generated images, and
merge the segmented previously generated images into a probability map which depicts a probability that each voxel represents the volume of interest, a probability that each voxel represents background, and a mean segmentation boundary;
a segmentation processor which registers the probability map with a current diagnostic image of the volume of interest in a current patient to generate a transformed probability map, the segmentation processor being programmed to register the probability map with the current image by performing the steps of:
registering the mean segmentation boundary to the volume of interest of one of the current image and a model registered to the current image;
determining a transform by which mean segmentation boundary was transformed to be registered to the current image or model;
transforming the probability map with the determined transform to generate the transformed probability map; and

a segmentation boundary processor which determines a segmentation boundary for the volume of interest based on the transformed probability map;
wherein each of the previously segmented images has a segmented image boundary; and
wherein the probability that each voxel represents the volume of interest and the probability that each voxel represents the background is based on how many of the segmented image boundaries each voxel is inside of or outside of.

9. A method for segmenting diagnostic images comprising:
segmenting a volume of interest in prior diagnostic images of a selected volume of interest generated from a plurality of patients to define a plurality of segmentation boundaries;
registering the segmented prior images including the segmentation boundaries;
merging the registered segmented prior images into a probability map which depicts a probability that each voxel represents one of the volume of interest or background based on a distribution of the segmentation boundaries in the prior diagnostic image, and a mean segmentation boundary; and
registering the probability map with a current diagnostic image of the volume of interest from a current patient to generate a transformed probability map.

10. The method according to claim 9 , wherein registering the probability map to the current image includes:
registering the mean segmentation boundary to one of the volume of interest of the current image and a model registered to the current image;
determining a transform by which the mean segmentation boundary was transformed into registration with the current image; and
transforming the probability map with the determined transform to generate the transformed probability map.

11. The method according to claim 9 , further including:
classifying the current image of the volume of interest with a previously trained classifier routine based on characteristics of the current image to determine a probability that at least selected voxels of the current image depict the volume of interest, and the background; and
merging the probabilities from the classifier routine and the probabilities from the transformed probability map as registered with the current image.

12. The method according to claim 11 , further including:
determining a segmentation boundary for the volume of interest of the current image based on the merged probabilities; and
combining the determined segmentation boundary with the current image.

13. The method according to claim 9 , further including:
assigning voxels of the current image which have probabilities in the transformed probability map above a threshold to one of the volume o

Assignees

• Pekar Vladimir
• Qazi Arish A
• Koninklijke Philips Nv

Inventors

• Pekar Vladimir
• Qazi Arish A

Classifications

• G06T7/0012 Primary

  Biomedical image inspection · CPC title

• G06T7/143

  involving probabilistic approaches, e.g. Markov random field [MRF] modelling · CPC title

• G06T2207/10072

  Tomographic images · CPC title

• G06T2207/20128

  Atlas-based segmentation · CPC title