Redundant spatial ensemble for computer-aided detection and image understanding

• US 8,494,238 B2
• Filed: 02/24/2011
• Issued: 07/23/2013
• Est. Priority Date: 12/21/2007
• Status: Active Grant

• Current Assignee: Siemens Healthcare GMBH (Siemens AG)
• Original Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Inventors: Zhou, Xiang Sean
• Primary Examiner(s): Fitzpatrick, Atiba O
• Application Number:
  US13/033,886
• Publication Number:
  US 20110188706A1
• Time in Patent Office: 880 Days
• Field of Search: 382128-134, 128920-925, 356 39- 49, 600407-414, 600424-426, 345581-618
• US Class Current: 382/128
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods

Abstract

Described herein is a technology for facilitating computer-aided detection and image understanding. In one implementation, an input set of training images of a target structure, such as an anatomical structure, is received. The input set of training images is spatially realigned to different landmarks to generate multiple bags of training images. At least one of the multiple bags comprises substantially all the training images in the input set, but realigned to a landmark. The multiple bags of training images may be used to train a spatial ensemble of detectors, which can be employed to generate an output result by automatically detecting a target structure in an input image.

Claims

1. A method of computer-aided detection, comprising:

(i) receiving an input set of training images of a target structure;

(ii) spatially realigning the input set of training images to different landmarks to generate multiple bags of training images, wherein at least one of the multiple bags comprises substantially all the training images in the input set realigned to at least one of the different landmarks;

(iii) training a spatial ensemble of detectors using the multiple bags of training images; and

(iv) generating an output result by using the trained spatial ensemble of detectors to automatically detect the target structure in an input image.

2. The method of claim 1 wherein the step (iv) comprises determining a presence, location, size, orientation, or combination thereof, of the target structure.

3. The method of claim 1 wherein the step (ii) comprises generating a number of the multiple bags that is more than a minimum number of landmarks required for determining a presence, location, size, orientation, or combination thereof, of the target structure.

4. The method of claim 1 wherein the step (iii) comprises performing an AdaBoost-based or random forest learning method.

5. The method of claim 1 further comprises performing consensus learning based on the spatial ensemble of detectors.

6. The method of claim 5 wherein the performing the consensus learning comprises performing majority voting to detect the target structure.

7. The method of claim 5 wherein the performing the consensus learning comprises generating one or more sparse configuration models by implementing a sparse voting scheme to detect any outliers.

8. The method of claim 7 wherein the sparse voting scheme comprises determining a vote received by a landmark point from a sparse set of landmarks points based on a conditional distribution estimated using the input set of training images.

9. The method of claim 7 wherein the sparse voting scheme is applied on or around the target structure.

10. The method of claim 7 wherein the sparse voting scheme is applied on a distal structure away from the target structure.

11. The method of claim 1 wherein the step (iii) comprises training the spatial ensemble of detectors at multiple scales ranging from a coarse level to a fine level.

12. The method of claim 1 wherein the step (iv) comprises determining a presence, location, size, orientation, or combination thereof, of the target structure.

13. The method of claim 1 wherein the step (iv) comprises predicting an imaging plane that is aligned for optimal visualization of at least a portion of the target structure in the input image.

14. The method of claim 1 wherein the step (iv) comprises predicting a two-dimensional or three-dimensional field of view for locating or imaging at least a portion of the target structure in the input image.

15. The method of claim 14 wherein the field of view is stationary or moving in time to align with any motion.

16. The method of claim 1 wherein the step (iv) comprises classifying and annotating at least a portion of the target structure in the input image.

17. The method of claim 1 further comprises generating a measure of reliability based on the output result.

18. A non-transitory computer readable medium embodying a program of instructions executable by machine to perform steps for computer-aided detection, the steps comprising:

(i) receiving an input set of training images of a target structure;

(ii) spatially realigning the input set of training images to different landmarks to generate multiple bags of training images, wherein at least one of the multiple bags comprises substantially all the training images in the input set realigned to at least one of the different landmarks;

(iii) training a spatial ensemble of detectors using the multiple bags of training images; and

(iv) generating an output result by using the trained spatial ensemble of detectors to automatically detect the target structure in an input image.

19. A computer-aided detection system, comprising:

a non-transitory memory device for storing computer readable program code; and

a processor in communication with the memory device, the processor being operative with the computer readable program code to;

(i) receive an input set of training images of a target structure;

(ii) spatially realign the input set of training images to different landmarks to generate multiple bags of training images, wherein at least one of the multiple bags comprises substantially all the training images in the input set realigned to at least one of the different landmarks;

(iii) train a spatial ensemble of detectors using the multiple bags of training images; and

(iv) generate an output result by using the trained spatial ensemble of detectors to automatically detect the target structure in an input image.