Methods and apparatus for applying user corrections to medical fact extraction

• US 9,922,385 B2
• Filed: 02/14/2014
• Issued: 03/20/2018
• Est. Priority Date: 02/18/2011
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Yegnanarayanan, Girija
• Primary Examiner(s): Kanaan, Maroun P
• Application Number:
  US14/180,525
• Publication Number:
  US 20140164023A1
• Time in Patent Office: 1,495 Days
• CPC Class Codes: G06Q 10/10 Office automation; Time management
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/228 of application context
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16Z 99/00 Subject matter not provided for in other main groups of this subclass

Abstract

Techniques for applying user corrections to medical fact extraction may include extracting a first set of one or more medical facts from a first portion of text documenting a patient encounter. A correction to the first set of medical facts may be received from a user. The correction may identify a fact that should be associated with the first portion of the text. A second set of one or more medical facts may be extracted from a second portion of the text based at least in part on the user'"'"'s correction to the first set of medical facts. Extracting the second set of facts may include extracting one or more facts similar to the identified fact from the second portion of the text.

Claims

1. A method of analyzing a text documenting a patient encounter, the method comprising:

extracting from a first portion of the text, using a fact extraction component implemented via at least one processor, a first set of one or more medical facts representing one or more abstract semantic concepts;

receiving information identifying a change made by a user to the first set of medical facts extracted from the first portion of the text, wherein the information identifying the change identifies at least one user-specified abstract semantic concept;

updating the fact extraction component based on the at least one user-specified abstract semantic concept and the first portion of the text, wherein the updating comprises;

identifying features included in the first portion of the text, the features including one or more tokens and/or one or more N-grams appearing in the first portion of the text;

identifying, from among the features identified in the first portion of the text, a first set of features to associate with the at least one user-specified abstract semantic concept, the first set of features including one or more features, the one or more features of the first set including one or more tokens and/or one or more N-grams; and

updating the fact extraction component to associate the first set of features with the at least one user-specified abstract semantic concept; and

extracting, using the fact extraction component following the updating and using the at least one processor, a second set of one or more medical facts from a second portion of the text, wherein extracting the second set comprises, in response to identifying in the second portion of the text a second set of features corresponding to the first set of features, extracting from the second portion of the text one or more facts corresponding to the at least one user-specified abstract semantic concept, the second set of features including one or more features.

2. The method of claim 1, wherein:

the at least one user-specified abstract semantic concept corresponds to a particular fact type, and

extracting the second set of facts comprises labeling one or more tokens in the second portion of the text as corresponding to the particular fact type.

3. The method of claim 1, wherein at least one of the one or more facts corresponding to the at least one user-specified abstract semantic concept has a same normalized form as the at least one user-specified abstract semantic concept.

4. The method of claim 1, wherein:

updating the fact extraction component comprises establishing one or more rules correlating the first set of features of the first portion of the text with the at least one user-specified abstract semantic concept, and

extracting using the fact extraction component following the updating comprises applying at least one of the one or more rules to the second portion of the text.

5. The method of claim 4, wherein extracting the one or more facts in response to identifying the second set of features in the second portion of the text comprises:

in response to determining that the second set of features satisfies at least one of the one or more rules correlating the first set of features with the at least one user-specified abstract semantic concept, extracting the one or more facts.

6. The method of claim 1, wherein:

updating the fact extraction component comprises establishing a rule assigning at least one feature of the first set of features correlated with the at least one user-specified abstract semantic concept to a token in the first portion of the text, and

extracting using the fact extraction component following the updating comprises, in response to applying the rule to the second portion of the text and determining that the second portion includes the token, including the at least one feature in the second set of features.

7. The method of claim 1, wherein updating the fact extraction component comprises re-training a statistical model used to extract the medical facts, the statistical model having been trained using original training data prior to extracting the first set of medical facts, the re-training comprising:

creating updated training data by adding the first portion of the text, labeled with the identified at least one user-specified abstract semantic concept, to the original training data; and

re-training the statistical model using the updated training data.

8. The method of claim 7, wherein extracting the one or more facts in response to identifying the second set of features in the second portion of the text comprises extracting the one or more facts at least in part in response to identifying that the at least one feature is included in the second set of features.

9. The method of claim 1, wherein correlating the set of features of the first portion of the text to the one or more abstract semantic concepts comprises applying one or more rules and/or one or more statistical models to the set of features.

10. The method of claim 1, wherein the second set of features is at least a subset of the first set of features.

11. The method of claim 1, wherein:

the second set of features comprises at least one of the one or more tokens and/or at least one of the one or more N-grams of the first set.

12. The method of claim 1, wherein identifying the second set of features corresponding to the first set of features comprises identifying that the second set of features corresponds to the first set of features.

13. The method of claim 1, wherein identifying the second set of features corresponding to the first set of features comprises identifying a second set of features that is a match to the first set of features.

14. The method of claim 13, wherein identifying the second set of features that is a match to the first set of features comprises determining that the second set of features is a match to the first set of features.

15. The method of claim 14, wherein determining that the second set of features is a match to the first set of features comprises:

with a statistical model that applies different weights to different features, generating a score indicating a similarity of the first set and the second set; and

determining whether the first set is a match to the second set based on the score.

16. The method of claim 1, wherein updating the fact extraction component based on the at least one user-specified abstract semantic concept and the first portion of the text comprises updating the fact extraction component based on an association between the at least one user-specified abstract semantic concept and the first portion of the text.

17. The method of claim 1, wherein identifying features included in the first portion of the text comprises identifying only some of the first portion of the text as the one or more features of the first set of features.

18. Apparatus comprising:

at least one processor; and

at least one memory storing processor-executable instructions that, when executed by the at least one processor, perform a method of analyzing a text documenting a patient encounter, the method comprising;

extracting from a first portion of the text, using a fact extraction component, a first set of one or more medical facts representing one or more abstract semantic concepts;

receiving information identifying a change made by a user to the first set of medical facts extracted from the first portion of the text, wherein the information identifying the change identifies at least one user-specified abstract semantic concept;

updating the fact extraction component based on the at least one user-specified abstract semantic concept and the first portion of the text, wherein the updating comprises;

identifying features included in the first portion of the text, the features including one or more tokens and/or one or more N-grams appearing in the first portion of the text;

identifying, from among the features identified in the first portion of the text, a first set of features to associate with the at least one user-specified abstract semantic concept, the first set of features including one or more features, the one or more features of the first set including one or more tokens and/or one or more N-grams; and

updating the fact extraction component to associate the first set of features with the at least one user-specified abstract semantic concept; and

extracting, using the fact extraction component following the updating and using the at least one processor, a second set of one or more medical facts from a second portion of the text, wherein extracting the second set comprises, in response to identifying in the second portion of the text a second set of features corresponding to the first set of features, extracting from the second portion of the text one or more facts corresponding to the at least one user-specified abstract semantic concept, the second set of features including one or more features.

19. The apparatus of claim 18, wherein:

updating the fact extraction component comprises establishing a rule assigning at least one feature of the first set of features correlated with the at least one user-specified abstract semantic concept to a token in the first portion of the text, and

extracting using the fact extraction component following the updating comprises, in response to applying the rule to the second portion of the text and determining that the second portion includes the token, including the at least one feature in the second set of features.

20. At least one non-transitory computer-readable storage medium encoded with computer-executable instructions that, when executed, perform a method of analyzing a text documenting a patient encounter, the method comprising:

extracting from a first portion of the text, using a set of one or more rules and/or one or more statistical models implemented via at least one processor, a first set of one or more medical facts representing one or more abstract semantic concepts;

receiving information identifying a change made by a user to the first set of medical facts extracted from the first portion of the text, wherein the information identifying the change identifies at least one user-specified abstract semantic concept;

updating the set of rules and/or the one or more statistical models based on the at least one user-specified abstract semantic concept and the first portion of the text, wherein the updating comprises;

identifying features included in the first portion of the text, the features including one or more tokens and/or one or more N-grams appearing in the first portion of the text;

identifying, from among the features identified in the first portion of the text, a first set of features to associate with the at least one user-specified abstract semantic concept, the first set of features including one or more features, the one or more features of the first set including one or more tokens and/or one or more N-grams; and

updating the set of rules and/or the one or more statistical models to associate the first set of features with the at least one user-specified abstract semantic concept; and

extracting, using the set of rules and/or the one or more statistical models following the updating and using the at least one processor, a second set of one or more medical facts from a second portion of the text, wherein extracting the second set comprises, in response to identifying in the second portion of the text a second set of features corresponding to the first set of features, extracting from the second portion of the text one or more facts corresponding to the at least one user-specified abstract semantic concept, the second set of features including one or more features.

21. The at least one non-transitory computer-readable storage medium of claim 20, wherein:

updating the set of rules and/or the one or more statistical models comprises establishing one or more rules correlating the first set of features of the first portion of the text with the at least one user-specified abstract semantic concept, and

extracting using the set of rules and/or the one or more statistical models following the updating comprises applying at least one of the one or more rules to the second portion of the text.