Viewpoint recognition in computer tomography images

• US 9,652,846 B1
• Filed: 10/22/2015
• Issued: 05/16/2017
• Est. Priority Date: 10/22/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Codella, Noel, Moradi, Mehdi, Syeda-Mahmood, Tanveer
• Primary Examiner(s): NAKHJAVAN, SHERVIN K
• Application Number:
  US14/920,604
• Publication Number:
  US 20170116728A1
• Time in Patent Office: 572 Days
• Field of Search: None
• CPC Class Codes: A61B 6/032 Transmission computed tomography [CT]
• CPC Class Codes: A61B 6/503 for diagnosis of the heart
• CPC Class Codes: A61B 6/5217 extracting a diagnostic or physiological parameter from medical diagnostic data
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30048 Heart; Cardiac
• CPC Class Codes: G06T 2207/30244 Camera pose
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06T 7/41 based on statistical description of texture
• CPC Class Codes: G06T 7/74 involving reference images or patches
• CPC Class Codes: G16H 50/30 for calculating health indices; for individual health risk assessment

Abstract

A solution is presented for cardiac CT viewpoint recognition to identify the desired images for a specific view and subsequent processing and anatomy recognition. A new set of features is presented to describe the global binary pattern of cardiac CT images characterized by the highly attenuating components of the anatomy in the image. Five classic image texture and edge feature sets are used to devise a classification approach based on SVM classification, class likelihood estimation, and majority voting, to classify 2D cardiac CT images into one of six viewpoint categories that include axial, sagittal, coronal, two chamber, four chamber, and short axis views. Such an approach results in an accuracy of 99.4% in correct labeling of the viewpoints.

Claims

1. A system for computed tomography (CT) viewpoint recognition in CT images comprising:

a feature engine comprising a global binary pattern feature calculation unit and at least one image feature calculation unit, said global binary pattern feature calculation unit;

(i) preprocessing said CT images with histogram equalization to maximize contrast in one or more pre-determined regions, (ii) extracting connected components from said preprocessed CT images, and (iii) forming a feature vector based on said extracted connected components;

a cognitive engine comprising;

(i) a plurality of classifiers, one for each of said global binary calculation unit, and (ii) at least one image feature calculation unit; and

a voting unit picking a class label for each of the CT images based on a majority voting scheme determining the most frequently returned label among said classifiers.

2. The system of claim 1, wherein said CT images are cardiac CT images.

3. The system of claim 2, wherein said class label is any of the following:

short axis view (SHA), two chamber view (2C), and four chamber view (4C).

4. The system of claim 1, wherein said connected components are extracted by applying Otsu thresholding.

5. The system of claim 4, wherein said Otsu thresholding is performed with four levels.

6. The system of claim 5, wherein said Otsu thresholding step comprises:

calculating optimal thresholds to minimize intra level variance, applying component clustering to a highest intensity level, and filtering and outputting connected components as said extracted connected components.

7. The system of claim 1, wherein said classifiers are support vector machine (SVM) classifiers.

8. A system for computed tomography (CT) viewpoint recognition in CT images comprising:

a feature engine comprising;

one or more image feature calculation units selected from a group consisting of;

a statistical image texture feature calculation unit, a curvelet feature calculation unit, a wavelet feature calculation unit, an edge histogram features calculation unit, and a local binary pattern (LBP) features calculation unit; and

a global binary pattern feature calculation unit;

(i) preprocessing said CT images with histogram equalization to maximize contrast in one or more pre-determined regions, (ii) extracting connected components from said preprocessed CT images, and (iii) forming a feature vector based on said extracted connected components;

a cognitive engine comprising a plurality of classifiers, (i) one for each of said global binary calculation unit, and (ii) one for each of said image feature calculation units; and

a voting unit picking a class label for each of the CT images based on a majority voting scheme determining the most frequently returned label among said classifiers.

9. The system of claim 8, wherein said CT images are cardiac CT images.

10. The system of claim 9, wherein said class label is any of the following:

short axis view (SHA), two chamber view (2C), and four chamber view (4C).

11. The system of claim 8, wherein said connected components are extracted by applying Otsu thresholding.

12. The system of claim 11, wherein said Otsu thresholding is performed with four levels.

13. The system of claim 12, wherein said Otsu thresholding step comprises:

calculating optimal thresholds to minimize intra level variance, applying component clustering to a highest intensity level, and filtering and outputting connected components as said extracted connected components.

14. The system of claim 8, wherein said classifiers are support vector machine (SVM) classifiers.

15. A method for computed tomography (CT) viewpoint recognition in CT images comprising:

preprocessing said CT images with histogram equalization to maximize contrast in one or more pre-determined regions;

extracting connected components from said preprocessed CT images;

forming a feature vector based on said extracted connected components;

applying a plurality of classifiers, (i) one for said feature vector formed from extracted connected components, and (ii) at least one additional classifier for a calculated image feature; and

picking a class label for each of the CT images based on a majority voting scheme determining the most frequently returned label among said classifiers.

16. The method of claim 15, wherein said CT images are cardiac CT images.

17. The method of claim 16, wherein said class label is any of the following:

short axis view (SHA), two chamber view (2C), and four chamber view (4C).

18. The method of claim 15, wherein said connected components are extracted by applying Otsu thresholding.

19. The method of claim 18, wherein said Otsu thresholding is performed with four levels and said Otsu thresholding step comprises:

calculating optimal thresholds to minimize intra level variance, applying component clustering to a highest intensity level, and filtering and outputting connected components as said extracted connected components.

20. The method of claim 15, wherein said classifiers are support vector machine (SVM) classifiers.

21. A method for use with a 3-D computed tomography (CT) image of anatomy relevant to a medical condition, the method comprising:

generating a binary image of the CT image, said binary image delineating bone from soft tissue;

analyzing said binary image with a machine learning component to train a classifier within said machine learning component and generating a taxonomy related to anatomical viewpoints of the CT image;

ranking a plurality of 2-D slices of the CT image based on said generated taxonomy; and

selecting a 2-D slice among said plurality of 2-D slices that is most relevant to the medical condition based on said ranking.

22. The method of claim 21, wherein said CT image is of a human chest, including the heart, and the viewpoints correspond to portions of the heart.

23. The method of claim 21, wherein said machine learning component includes support vector machines (SVMs).

24. The method of claim 23, wherein said SVMs vote to determine said selected 2-D slice among said plurality of 2-D slices.