Anonymization of heterogenous clinical reports

Patent metadata

Field	Value
Publication number	US-2020387635-A1
Application number	US-202016868377-A
Country	US
Kind code	A1
Filing date	May 6, 2020
Priority date	Jun 5, 2019
Publication date	Dec 10, 2020
Grant date	—

Abstract

For anonym izing or other keyword identification medical patient data, a conditional random field sequence classifier is used for the NER model for NLP, providing a technical solution to help the computer perform better at identifying PHI from context and reduce manual anonym ization efforts of medical reports. One tool or executable integrates report format conversion, annotation, training, and application. These operations may be selected, or the tool configured for anonymization or keyword identification. Different files from each stage may be exported or used by others operating on other computers, allowing collaboration or sequential burden sharing for anonym ization.

First claim

I (we) claim:

1 . A method for anonymizing medical patient data with a machine-learned system, the method comprising:
inputting the medical patient reports for multiple patients of a medical institution to a machine-learned condition random field sequence classifier, the medical patient reports including protected health information for the multiple patients; outputting, by the machine-learned condition random field sequence classifier in response to the inputting, anonymized patient data free of the protected health information; and transmitting the anonymized patient data to an entity other than the medical institution.

2 . The method of claim 1 wherein outputting comprises labeling and removing the protected health information by the machine-learned condition random field sequence having been trained as a linear chain condition random field sequence classifier.

3 . The method of claim 1 further comprising converting the medical patient reports from heterogenous formats to a common format, and wherein inputting comprises inputting the medical patient reports in the common format.

4 . The method of claim 3 wherein the converting, inputting, and outputting are part of a single executable with a library of multiple functions.

5 . The method of claim 1 wherein the machine-learned condition random field sequence classifier was trained on other reports of the medical institution.

6 . The method of claim 1 wherein inputting comprises inputting to the machine-learned condition random field sequence classifier as a named entity recognition model using natural language processing.

7 . The method of claim 1 further comprising removing strings from the medical patient report with a search function prior to the inputting, the strings being a search term or occurring in relation to a search term.

8 . The method of claim 1 further comprising:
outputting, by the machine-learned condition random field sequence classifier, the medical patient reports with annotations identifying the protected health information; and
re-training the machine-learned condition random field sequence classifier based on the medical patient reports with the annotations.

9 . The method of claim 1 wherein inputting and outputting are performed on a first computer, and further comprising operating the machine-learned condition random field sequence classifier on a second computer different than the first computer.

10 . The method of claim 1 wherein inputting comprises inputting to the machine-learned condition random field sequence classifier at a cloud server and wherein outputting comprises outputting to a computer of the medical institution different than the cloud server.

11 . The method of claim 1 wherein inputting comprises inputting to the machine-learned condition random field sequence classifier and another machine-learned classifier, and wherein outputting comprises outputting an aggregation from the machine-learned condition random field sequence classifier and the other machine-learned classifier.

12 . The method of claim 11 wherein inputting to the other machine-learned classifier comprises inputting medical images and wherein outputting by the other machine-learned classifier comprises outputting anonym ized images.

13 . The method of claim 1 wherein inputting and outputting are performed as part of a single executable, and further comprising:
machine training the machine-learned condition random field sequence classifier as part of the single executable; and
machine training another classifier to extract diagnostic or prognostic information as part of the single executable.

14 . A method for machine-training to anonym ize medical patient data, the method comprising:
executing an anonym ization tool, the executed anonym ization tool including annotation, training, and application of the machine-trained model; annotating a plurality of first medical reports, the annotating identifying patient identifiers in the first medical reports as the annotation; machine learning the machine-trained model to anonymize the first medical reports from the annotation as the training; and applying the machine-trained model to second medical reports as the application, the application providing the second medical reports with the patient identifiers removed.

15 . The method of claim 14 wherein executing comprises executing with the executed anonymization tool including format conversion, and further comprising converting the first medical reports in heterogenous formats into a common format as the format conversion, wherein the annotating comprises annotating the first medical reports in the common format.

16 . The method of claim 14 wherein the first medical reports prior to annotating are in one or more first files, wherein the first medical reports after annotating are in one or more second files, wherein the machine-trained model is in a third file, and wherein the second reports with the patient identifiers removed are in one or more fourth files, and further comprising performing the annotating, machine learning, and applying at different computers based on exporting of the one or more first files, the one or more second files, the third file, and/or the one or more fourth files and based on instantiations of the anonymization tool on the different computers.

17 . The method of claim 14 further comprising, with the executed anonym ization tool, annotating third medical reports for identifications of prognosis or diagnosis, and machine training another machine-trained model to determine the prognosis or diagnosis from the identifications.

18 . The method of claim 14 wherein machine learning comprises training the machine-trained model as a condition random field sequence classifier.

19 . The method of claim 14 wherein executing comprises executing with the executed anonymization tool including user defined rules, and further comprising entering, by the user, text strings to remove and/or defining locations to be removed as the user defined rules.

20 . A system for keyword identification of medical reports for export, the system comprising:
a medical records database having stored therein a plurality of patient files in different formats; an interface configured by a machine training anonymization application to receive file identifiers for the plurality of patient files; and a processor configured to convert the plurality of patient files in the different formats into a common format and to machine learn keyword identification based on the patient files in the common format.

Assignees

• Siemens Healthcare Gmbh

Inventors

• Lo Ren-Yi
• Ullaskrishnan Poikavila

Classifications

• G16H15/00 Primary

  ICT specially adapted for medical reports, e.g. generation or transmission thereof · CPC title

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• G06F16/116

  Details of conversion of file system types or formats · CPC title

• G16H50/70

  for mining of medical data, e.g. analysing previous cases of other patients · CPC title

• G16H10/60

  for patient-specific data, e.g. for electronic patient records · CPC title