SYSTEM AND METHOD FOR AUTOMATIC RECOGNITION AND LABELING OF ANATOMICAL STRUCTURES AND VESSELS IN MEDICAL IMAGING SCANS

• US 20100054525A1
• Filed: 08/27/2008
• Published: 03/04/2010
• Est. Priority Date: 08/27/2008
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Gong, Leiguang, Rohrer, Jonathan B., Iyengar, Giridharan R.
• Granted Patent: US 8,385,688 B2
• Time in Patent Office: Days
• US Class Current: 382/100
• CPC Class Codes: G06T 2207/30012 Spine; Backbone
• CPC Class Codes: G06T 2207/30101 Blood vessel; Artery; Vein; Vascular
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/149 involving deformable models, e.g. active contour models
• CPC Class Codes: G06T 7/187 involving region growing; involving region merging; involving connected component labelling
• CPC Class Codes: G06V 10/754 involving a deformation of the sample pattern or of the reference pattern; Elastic matching
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images

Abstract

A system and method for recognizing and labeling anatomical structures in an image includes creating a list of objects such that one or more objects on the list appear before a target object and setting the image as a context for a first object on the list. The first object is detected and labeled by subtracting a background of the image. A local context is set for a next object on the list using the first object. The next object is detected and labeled by registration using the local context. Setting a local context and detecting and labeling the next object are repeated until the target object is detected and labeled. Labeling of the target object is refined using region growing.

Claims

1. A method for recognizing and labeling anatomical structures in an image, comprising:

creating a list of objects such that one or more objects on the list appear before a target object;

setting an entire image as a context for a first object on the list;

detecting and labeling the first object by subtracting a background from the image;

setting a local context for a next object on the list using the first object;

detecting and labeling the next object by registration using the local context;

repeating the steps of setting a local context and detecting and labeling the next object until the target object is detected and labeled; and

refining labeling of the target object using region growing.

2. The method as recited in claim 1, wherein detecting and labeling the next object by registration using the local context includes detecting and labeling the next object using a non-rigid registration process.

3. The method as recited in claim 1, wherein the non-rigid registration process provides an approximate mapping for a current object.

4. The method as recited in claim 3, further comprising refining the local context of an object by employing the approximate mapping to define a refined context by expanding a labeled region for the current object by a margin of pixels.

5. The method as recited in claim 4, further comprising reregistering the current object in the refined context.

6. The method as recited in claim 1, further comprising redefining labeling on the target object using a region growing method.

7. The method as recited in claim 6, wherein the region growing method further comprises automatically selecting a seed pixel in the target object;

and growing the region from the seed pixel.

8. The method as recited in claim 7, wherein the seed pixel includes a mask and the region growing results in a more accurate detection and labeling of the target object.

9. A computer readable medium comprising a computer readable program for recognizing and labeling anatomical structures in an image, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:

creating a list of objects such that one or more objects on the list appear before a target object;

setting an entire image as a context for a first object on the list;

detecting and labeling the first object by subtracting a background from the image;

setting a local context for a next object on the list using the first object;

detecting and labeling the next object by registration using the local context;

repeating the steps of setting a local context and detecting and labeling the next object until the target object is detected and labeled; and

refining labeling of the target object using region growing.

10. The computer readable medium as recited in claim 9, wherein detecting and labeling the next object by registration using the local context includes detecting and labeling the next object using a non-rigid registration process.

11. The computer readable medium as recited in claim 9, wherein the non-rigid registration process provides an approximate mapping for a current object.

12. The computer readable medium as recited in claim 11, further comprising refining the local context of an object by employing the approximate mapping to define a refined context by expanding a labeled region for the current object by a margin of pixels.

13. The computer readable medium as recited in claim 12, further comprising reregistering the current object in the refined context.

14. The computer readable medium as recited in claim 9, further comprising redefining labeling on the target object using a region growing method.

15. The computer readable medium as recited in claim 14, wherein the region growing method further comprises automatically selecting a seed pixel within a region labeled by a registration process for the target object;

and growing the region from the seed pixel.

16. The computer readable medium as recited in claim 15, wherein the seed pixel selection includes a mask and the region growing results in a more accurate detection and labeling of the target object.

17. A method for recognizing and labeling objects in an image, comprising:

providing a first image with anatomical objects with labels of the objects, a second image which is the same as the first image but without labels, and a target image having one or more anatomical structures to be recognized and labeled;

storing all the images in indexable data structures in memory;

subtracting a background of the second image and the target image by setting the pixels in the background to a unique pre-specified value;

setting an initial alignment of the second image and the target image at a pre-selected point and angle;

performing a progressively focused non-rigid registration with the second image and the target image according to a list of ordered reference objects;

applying a transformation or deformation field produced in the registration to the first image to produce a transformed labeled image;

projecting labels from the transformed labeled image to the target image;

selecting one or more seed pixels; and

region growing the seed pixels to produce a final recognition and labeling of the target image.

18. A computer readable medium comprising a computer readable program for recognizing and labeling objects in an image, wherein the computer readable program when executed on a computer causes the computer to perform the steps as recited in claim 17.

19. The method as recited in claim 17, further comprising sending the final recognition and labeling of the target image to at least one of a display device and a storage media.

20. A system for recognizing and labeling anatomical structures in an image, comprising:

a preprocessing unit configured to normalize a non-labeled atlas image and a target image, which is an image in which one or more anatomical structures are to be recognized and labeled, by mapping intensity values to new values within a pre-specified range to perform a background subtraction;

a registration unit configured to perform a progressively focused non-rigid registration process that results in a registration between the non-labeled atlas image and the target image; and

a segmentation unit configured to region grow a selected seed pixel or set of seed pixels using the results of the progressively focused registration process to output a final recognition and labeling of the target image with the labels of a labeled atlas image.

21. The system as recited in claim 20, wherein the preprocessing unit provides a background subtracting process, which dynamically computes a threshold value for each of a plurality of image slices, and then compares every pixel intensity value against the threshold value, and pixels with intensity value lower than the threshold value, are defined as background and assigned a same unique value, all other pixels will keep their original value.

23. The system as recited in claim 20, wherein after the registration unit performs the registration between the second image and the target image, the first image with labels is employed to label the target image.

24. The system as recited in claim 20, wherein the preprocessing unit is configured to perform region grouping to produce a list of independent regions of connected non-zero value pixels in the images to be used in the background subtraction.

25. The system as recited in claim 20, wherein the segmentation unit is configured to region grow the selected seed pixel or set of seed pixels to result in a more accurate detection and labeling of the target image.

22. The system as recited in claim 22, wherein the registration unit is configured to perform the registration between a second image without atlas labels and the target image, where the second image is a same image as the first image.