Generation of medical records based on doctor-patient dialogue

Patent metadata

Field	Value
Publication number	US-11676735-B2
Application number	US-201916569818-A
Country	US
Kind code	B2
Filing date	Sep 13, 2019
Priority date	Sep 13, 2019
Publication date	Jun 13, 2023
Grant date	Jun 13, 2023

Abstract

Methods and apparatus for generating medical records from a doctor-patient dialogue are provided. A main content portion of a written doctor-patient conversation is identified. The main content portion of the conversation is extracted from the conversation. The main content of the conversation is divided into sections according to a pre-defined set of sections, and, based on the sections and their respective content, a medical record is generated according to a pre-defined template. The pre-defined template is one of a hard medical record format or a soft medical record format.

First claim

What is claimed is:

1. A method comprising:
identifying a main content portion of a written record of a doctor-patient conversation;
extracting the main content portion from the doctor-patient conversation;
dividing the main content portion into sections according to a pre-defined set of sections;
identifying a section type, of a pre-defined set of section types, to assign at least a first section of the main content portion to, comprising:
generating at least one word vector based on the first section of the main content portion;
generating at least one sentence vector based at least in part on the at least one word vector and a sentence embedding;
generating an entity vector based on at least one medical entity identified in the first section of the main content portion and a word embedding; and
assigning the section type to the first section of the main content portion by processing the at least one sentence vector and the entity vector using a trained neural network, wherein:
the trained neural network comprises a bidirectional long short-term (Bi-LSTM) model; and
the trained neural network is trained to categorize the generated sentence vector and the generated entity vector into a corresponding section type, of the plurality of pre-defined section types; and

based on respective contents of the sections, generating a medical record according to a pre-defined template, wherein:
the pre-defined template comprises a plurality of fields,
each respective field of the plurality of fields comprises a respective name that matches a respective section type of the pre-defined set of sections,
a first field of the plurality of fields matches the section type assigned to the first section of the main content portion, and
the first field is filled with one or more values of the at least one medical entity identified in the first section.

2. The method of claim 1 , further comprising:
receiving the doctor-patient conversation in oral form, and using speech recognition, transforming the oral conversation into a written conversation.

3. The method of claim 1 , wherein the pre-defined set of sections of the main portion includes a chief complaint section, a patient information section, and a medical advice section.

4. The method of claim 1 , wherein the pre-defined template is one of a hard medical record format or a soft medical record format.

5. The method of claim 4 , wherein the pre-defined template is a hard medical record format, and further comprising identifying values for one or more pre-defined medical entities in one or more of the sections.

6. The method of claim 5 , wherein the medical record includes values for at least one of the pre-defined medical entities.

7. The method of claim 6 , wherein the pre-defined medical entities include symptom, diagnosis, drug and patient occupation.

8. The method of claim 1 , wherein the doctor-patient conversation includes a plurality of question and answer (QA) pairs.

9. The method of claim 8 , wherein dividing the main content portion into sections further comprises assigning each QA pair or other statement to one of the pre-defined set of sections.

10. The method of claim 9 , wherein assigning each QA pair or portion of a QA pair to one of the pre-defined sections further comprises:
parsing each QA pair to obtain sentence vectors and entity vectors, and
processing the sentence vectors and entity vectors to identify their section type.

11. A system, comprising:
a processor; and
a memory storage device, including processor-executable instructions that when performed by the processor perform an operation comprising:
identifying a main portion of a written doctor-patient conversation;
extracting the main portion of the written doctor-patient conversation;
identifying a section type, of a pre-defined set of section types, to assign at least a first section of the main content portion to, comprising:
generating at least one word vector based on the first section of the main content portion;
generating at least one sentence vector based at least in part on the at least one word vector and a sentence embedding;
generating an entity vector based on at least one medical entity identified in the first section of the main content portion and a word embedding; and
assigning the section type to the first section of the main content portion by processing the at least one sentence vector and the entity vector using a trained neural network, wherein:
the trained neural network comprises a bidirectional long short-term (Bi-LSTM) model; and

the trained neural network is trained to categorize the generated sentence vector and the generated entity vector into a corresponding section type, of the plurality of pre-defined section types; and

based on respective contents of the sections, generating a medical record according to a pre-defined template, wherein:
the pre-defined template comprises a plurality of fields,
each respective field of the plurality of fields comprises a respective name that matches a respective section type of the pre-defined set of sections,
a first field of the plurality of fields matches the section type assigned to the first section of the main content portion, and
the first field is filled with one or more values of the at least one medical entity identified in the first section.

12. The system of claim 11 , wherein the operation further comprises:
receiving a recording of a doctor-patient conversation, and transforming it into a written conversation between the doctor and the patient.

13. The system of claim 11 , wherein the pre-defined template is one of a hard medical record format or a soft medical record format.

14. The system of claim 13 , wherein the pre-defined template is a hard medical record format, and further comprising a named entity recognition engine configured to identify values for one or more pre-defined medical entities in one or more of the sections.

15. The system of claim 11 , wherein the doctor-patient conversation includes a plurality of QA pairs, and wherein the operation further comprises assigning each QA pair or other statement to one of a pre-defined set of sections.

16. A computer program product for generating medical records from a doctor-patient dialogue, the computer program product comprising:
a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to:
identify a main content portion of a written record of a doctor-patient conversation;
extract the main content portion from the doctor-patient conversation;
divide the main content portion into sections according to a pre-defined set of sections;
identify a section type, of a pre-defined set of section types, to assign at least a first section of the main content portion to, comprising:
generate at least one word vector based on the first section of the main content portion;
generate at least one sentence vector based at least in part on the at least one word vector and a sentence embedding;
generate an entity vector based on at least one medical entity identified in the first section of the main content portion and a word embedding; and
assign the section type to the first section of the main content portion by processing the at least one sentence vector and the entity vector using a trained neural network, wherein:
the trained neural network comprises a bidirectional long short-term (Bi-LSTM) model; and
the trained neural network is trained to categorize the generated sentence vector a

Assignees

• IBM

Inventors

• Wang Ke
• Jiang Jian Min
• Wang Jian
• Li Jing
• Huang Songfang

Classifications

• G06N20/00

  Machine learning · CPC title

• G16H15/00

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

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G16H80/00 Primary

  ICT specially adapted for facilitating communication between medical practitioners or patients, e.g. for collaborative diagnosis, therapy or health monitoring · CPC title

• G06N3/045

  Combinations of networks · CPC title