Automatic Learning of Image Features to Predict Disease

• US 20090310836A1
• Filed: 04/22/2009
• Published: 12/17/2009
• Est. Priority Date: 06/12/2008
• Status: Active Grant

• Current Assignee: Siemens Healthineers AG (Siemens AG)
• Original Assignee: Siemens AG, Siemens Medical Solutions USA Incorporated (Siemens AG)
• Inventors: Zhou, Xiang, Krishnan, Arun, Kelm, Michael, Freund, Joerg, Huber, Martin
• Granted Patent: US 7,949,167 B2
• Time in Patent Office: Days
• US Class Current: 382/128
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A method for training a computer system for automatic detection of regions of interest includes receiving patient records. For each of the received patient records a text field and a medical image are identified from within the patient record and the medical image is automatically segmented to identify a structure of interest. The text field is searched for one or more keywords indicative of a particular abnormality associated with the structure of interest. The medical image is added to a grouping representing the particular abnormality when the text field indicates that the patient has the particular abnormality and the medical image is added to a grouping representing the absence of the particular abnormality when the text field does not indicate that the patient has the particular abnormality. The groupings of medical images are used to automatically train a computer system for the subsequent detection of the particular abnormality.

Claims

1. A method for training a computer system for automatic detection of regions of interest in medical image data using a computer-based image processing device, comprising:

receiving a plurality of patient records from an electronic medical records database, and for each of the received patient records;

identifying a text field and a medical image from within the patient record;

automatically segmenting the medical image to identify a structure of interest;

searching the text field for one or more keywords indicative of a particular abnormality associated with the structure of interest;

determining whether the text field indicates that the patient has the particular abnormality; and

adding the medical image to a grouping of medical images representing the particular abnormality when it is determined that the text field indicates that the patient has the particular abnormality and adding the medical image to a grouping of medical images representing the absence of the particular abnormality when it is determined that the text field does not indicate that the patient has the particular abnormality, and

using the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality to automatically train a computer system for the subsequent detection of the particular abnormality.

2. The method of claim 1, wherein for all records, the medical image is a CT scan, an MRI, a PET or SPECT scan, photograph or an ultrasound image.

3. The method of claim 1, wherein for each record, the text field is from a radiology report corresponding to the medical image.

4. The method of claim 1, wherein the structure of interest is a particular organ.

5. The method of claim 4, wherein the particular abnormality is a particular disease, injury or defect of the particular organ.

6. The method of claim 1, wherein determining whether the text field indicates that the patient has the particular abnormality includes searching the text field for keywords that are indicative of a positive diagnosis for the particular abnormality.

7. The method of claim 1, wherein training the computer system for the subsequent detection of the particular abnormality includes training one or more classifiers based on the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality.

8. The method of claim 1, wherein using the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality to automatically train for the subsequent detection of the particular abnormality includes using the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality as training data for a machine-learning operation.

9. A method for automatic detection of regions of interest in medical image data using a computer-based image processing device, comprising:

receiving a plurality of patient records from an electronic medical records database, and for each of the received patient records;

identifying a text field and a medical image from within the patient record;

automatically segmenting the medical image to identify a structure of interest;

searching the text field for one or more keywords indicative of a particular abnormality associated with the structure of interest;

determining whether the text field indicates that the patient has the particular abnormality; and

adding the medical image to a grouping of medical images representing the particular abnormality when it is determined that the text field indicates that the patient has the particular abnormality and adding the medical image to a grouping of medical images representing the absence of the particular abnormality when it is determined that the text field does not indicate that the patient has the particular abnormality, and

using the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality to automatically train a computer-learning algorithm;

acquiring a subsequent medical image of a subsequent patient; and

using the trained computer-learning algorithm to analyze the subsequent medical image to aid in determining whether the subsequent patient has the particular abnormality.

10. The method of claim 9, wherein for all records, the medical image is a CT scan, an MRI, a PET or SPECT scan, photograph or an ultrasound image;

and the subsequent medical image is of the same modality as the records.

11. The method of claim 9, wherein for each record, the text field is from a radiology report corresponding to the medical image.

12. The method of claim 9, wherein the structure of interest is a particular organ.

13. The method of claim 9, wherein the particular abnormality is a particular disease, injury or defect of the particular organ.

14. The method of claim 9, wherein determining whether the text field indicates that the patient has the particular abnormality includes searching the text field for keywords that are indicative of a positive diagnosis for the particular abnormality.

15. The method of claim 9, wherein training the computer-learning algorithm includes developing one or more classifiers based on the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality.

16. A computer system comprising:

a processor; and

a program storage device readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for training a computer system for automatic detection of regions of interest, the method comprising;

receiving a plurality of patient records from an electronic medical records database, and for each of the received patient records;

identifying a text field and a medical image from within the patient record;

automatically segmenting the medical image to identify a structure of interest;

determining whether the text field indicates that the patient has a particular abnormality associated with the structure of interest; and

adding the medical image to a grouping of medical images representing the particular abnormality when it is determined that the text field indicates that the patient has the particular abnormality and adding the medical image to a grouping of medical images representing the absence of the particular abnormality when it is determined that the text field does not indicate that the patient has the particular abnormality, and

using the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality to automatically train a computer system for the subsequent detection of the particular abnormality.

17. The computer system of claim 16, wherein for each record, the text field is from a radiology report corresponding to the medical image.

18. The computer system of claim 16, wherein the structure of interest is a particular organ and the particular abnormality is a particular disease, injury or defect of the particular organ.

19. The computer system of claim 16, wherein determining whether the text field indicates that the patient has the particular abnormality includes searching the text field for keywords that are indicative of a positive diagnosis for the particular abnormality.

20. The computer system of claim 16, wherein training the computer system for the subsequent detection of the particular abnormality includes developing one or more classifiers based on the grouping of medical images representing the particular abnormality and the grouping of medical images representing the absence of the particular abnormality.