AUTOMATIC GROUND TRUTH GENERATION FOR MEDICAL IMAGE COLLECTIONS

• US 20160232658A1
• Filed: 02/06/2015
• Published: 08/11/2016
• Est. Priority Date: 02/06/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Syeda-Mahmood, Tanveer F.
• Granted Patent: US 9,842,390 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 10/464 using a plurality of salient features, e.g. bag-of-words [BoW] representations
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/772 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
• CPC Class Codes: G06V 20/47 Detecting features for summarising video content
• CPC Class Codes: G16H 15/00 ICT specially adapted for medical reports, e.g. generation or transmission thereof
• CPC Class Codes: G16H 30/40 for processing medical images, e.g. editing

Abstract

Methods and arrangements for automatic ground truth generation of medical image collections. Aspects include receiving a plurality of imaging studies, wherein each imaging study includes one or more images and a textual report associated with the one or more images. Aspects also include selecting a key image from each of the one or more images from each of the plurality of imaging studies and extracting one or more discriminating image features from a region of interest within the key image. Aspects further include processing the textual report associated with the one or more images to detect one or more concept labels, assigning an initial label from the one or more concept labels to the one or more discriminating image features, and learning an association between each of the one or more discriminating image features and the one or more concept labels.

Claims

1. A method for automatic ground truth generation of medical image collections, the method comprising:

receiving a plurality of imaging studies, wherein each imaging study comprises one or more images and a textual report associated with the one or more images;

selecting a key image from each of the one or more images from each of the plurality of imaging studies;

extracting one or more discriminating image features from a region of interest within the key image;

processing the textual report associated with the one or more images to detect one or more concept labels;

assigning an initial label from the one or more concept labels to the one or more discriminating image features; and

learning an association between each of the one or more discriminating image features and the one or more concept labels.

2. The method of claim 1 further comprising:

creating a set of frequently occurring concept labels based on the one or more concept labels detected in the textual reports of the plurality of imaging studies.

4. The method of claim 1, wherein processing the textual report associated with the one or more images to detect the one or more concept labels comprises executing a longest common subfix algorithm on the textual report.

5. The method of claim 1, wherein the plurality of imaging studies are echocardiogram studies and the region of interest is a Doppler spectrum.

6. The method of claim 5, wherein extracting one or more discriminating image features from the Doppler spectrum includes densely sampling both an envelope and an interior of the Doppler spectrum using one or more scale-invariant feature transform features.

7. The method of claim 1, wherein learning the association between each of the one or more discriminating image features and the one or more concept labels comprises executing a convex optimization learning algorithm.

3. The method of claim 3 wherein learning the association between each of the one or more discriminating image features and the one or more concept labels comprises determining if the initial label assigned to the one or more discriminating image features is within the set of frequently occurring concept labels.

8. A computer program product for automatic ground truth generation of medical image collections, the computer program product comprising:

a computer readable storage medium having program code embodied therewith, the program code executable by a computer to;

receive a plurality of imaging studies, wherein each imaging study comprises one or more images and a textual report associated with the one or more images;

select a key image from each of the one or more images from each of the plurality of imaging studies;

extract one or more discriminating image features from a region of interest within the key image;

process the textual report associated with the one or more images to detect one or more concept labels;

assign an initial label from the one or more concept labels to the one or more discriminating image features; and

learn an association between each of the one or more discriminating image features and the one or more concept labels.

9. The computer program product of claim 8, wherein the program code executable by the computer to further creates a set of frequently occurring concept labels based on the one or more concept labels detected in the textual reports of the plurality of imaging studies.

10. The computer program product of claim 9, wherein learning the association between each of the one or more discriminating image features and the one or more concept labels comprises determining if the initial label assigned to the one or more discriminating image features is within the set of frequently occurring concept labels.

11. The computer program product of claim 8, wherein processing the textual report associated with the one or more images to detect the one or more concept labels comprises executing a longest common subfix algorithm on the textual report.

12. The computer program product of claim 8, wherein the plurality of imaging studies are echocardiogram studies and the region of interest is a Doppler spectrum.

13. The computer program product of claim 12 wherein extracting one or more discriminating image features from the Doppler spectrum includes densely sampling both an envelope and an interior of the Doppler spectrum using one or more scale-invariant feature transform features.

14. The computer program product of claim 8, wherein learning the association between each of the one or more discriminating image features and the one or more concept labels comprises executing a convex optimization learning algorithm.

15. A computer system for automatic ground truth generation of medical image collections, the computer system comprising:

a processor configured to execute a program code causing the computer system to;

receive a plurality of imaging studies, wherein each imaging study comprises one or more images and a textual report associated with the one or more images;

select a key image from each of the one or more images from each of the plurality of imaging studies;

extract one or more discriminating image features from a region of interest within the key image;

process the textual report associated with the one or more images to detect one or more concept labels;

assign an initial label from the one or more concept labels to the one or more discriminating image features; and

learn an association between each of the one or more discriminating image features and the one or more concept labels.

16. The computer system of claim 15, wherein the program code further causes the computer system to create a set of frequently occurring concept labels based on the one or more concept labels detected in the textual reports of the plurality of imaging studies.

17. The computer system of claim 16, wherein learning the association between each of the one or more discriminating image features and the one or more concept labels comprises determining if the initial label assigned to the one or more discriminating image features is within the set of frequently occurring concept labels.

18. The computer system of claim 15, wherein processing the textual report associated with the one or more images to detect the one or more concept labels comprises executing a longest common subfix algorithm on the textual report.

19. The computer system of claim 15, wherein the plurality of imaging studies are echocardiogram studies and the region of interest is a Doppler spectrum.

20. The computer system of claim 19 wherein extracting one or more discriminating image features from the Doppler spectrum includes densely sampling both an envelope and an interior of the Doppler spectrum using one or more scale-invariant feature transform features.