System and method for clinical decision support system with inquiry based on reinforcement learning

Patent metadata

Field	Value
Publication number	US-11763944-B2
Application number	US-201916409293-A
Country	US
Kind code	B2
Filing date	May 10, 2019
Priority date	May 10, 2019
Publication date	Sep 19, 2023
Grant date	Sep 19, 2023

Abstract

A method and apparatus include receiving, by a device, medical information associated with a user. Inquiry information is determined based on the medical information associated with the user and a reinforcement learning model. The inquiry information is provided to permit response information to be received. The response information is received based on providing the inquiry information. Diagnosis information is determined based on the medical information and the response information using a machine learning model. The diagnosis information is provided to a set of devices via a network.

First claim

What is claimed is:

1. A method comprising:
receiving, by a device, medical information associated with a user;
training, by the device, a reinforcement learning model using at least one of electronic medical record (EMR) data and electronic health record (EHR) data to maximize a reward function that scores importance of a plurality of potential questions;
determining, by the device, questions to ask the user based on the medical information associated with the user and the trained reinforcement learning model;
providing, by the device, the questions to a doctor to permit response information to be received;
receiving, by the device, the response information based on the questions;
determining, by the device, diagnosis information based on the medical information and the response information using a machine learning model; and
providing, by the device, the diagnosis information in a standardized format to a set of devices via a network in real time,
wherein the determining questions comprises determining, by the device, whether an importance score of a potential question of the plurality of potential questions is greater than a threshold score, wherein the importance score is determined using at least a risk of death associated with question uncertainty, reinforcement learning, and at least a risk of death associated with question uncertainty, reinforcement learning, and at least one of: how relevant a potential question is, or how probative a potential question is; and
when the importance score is greater than the threshold score, determining the potential question as one of the questions to ask the user.

2. The method of claim 1 , further comprising:
performing a named entity recognition technique using the medical information; and
wherein determining the diagnosis information comprises determining the diagnosis information based on the named entity recognition technique.

3. The method of claim 1 , further comprising:
performing a semantic role labelling technique using the medical information; and
wherein determining the diagnosis information comprises determining the diagnosis information based on the semantic role labelling technique.

4. The method of claim 1 , wherein the reinforcement learning model has been trained by:
determining, based on the medical information, a set of potential questions;
determining respective importance scores associated with the set of potential questions that indicate the importance of the associated potential questions;
inputting the respective scores and the set of potential questions into the reinforcement learning model,
wherein the reward function maximizes the importance score.

5. The method of claim 1 , wherein determining the diagnosis information comprises determining the diagnosis information using at least one of a recurrent neural network (RNN), a convolutional neural network (CNN), and support vector machine (SVM).

6. The method of claim 1 , wherein the determining the questions to ask the user comprises:
determining a set of importance scores corresponding to the plurality of potential questions based on the medical information associated with the user and the trained reinforcement learning model that has been trained using at least one of electronic medical record (EMR) data and electronic health record (EHR) data to maximize the reward function that scores the importance of the plurality of potential questions.

7. A device comprising:
at least one memory configured to store program code;
at least one processor configured to read the program code and operate as instructed by the program code, the program code comprising:
first receiving code configured to cause the at least one processor to receive medical information associated with a user;
first training code configured to cause the at least one processor to train a reinforcement learning model using at least one of electronic medical record (EMR) data and electronic health record (EHR) data to maximize a reward function that scores importance of a plurality of potential questions;
first determining code configured to cause the at least one processor to determine questions to ask the user based on the medical information associated with the user and the trained reinforcement learning model;
providing code configured to cause the at least one processor to provide the questions to a doctor to permit response information to be received;
second receiving code configured to cause the at least one processor to receive the response information based on the questions;
second determining code configured to cause the at least one processor to determine diagnosis information based on the medical information and the response information using a machine learning model; and
providing code configured to cause the at least one processor to provide the diagnosis information in a standardized format to a set of devices via a network in real time,
wherein the first determining code is configured to cause the at least one processor to determine whether an importance score of a potential question of the plurality of potential questions is greater than a threshold score, wherein the importance score is determined using at least a risk of death associated with question uncertainty, reinforcement learning, and at least one of: how relevant a potential question is, or how probative a potential question is; and
when the importance score is greater than the threshold score, determine the potential question as one of the questions to ask the user.

8. The device of claim 7 , further comprising:
performing code configured to cause the at least one processor to perform a named entity recognition technique using the medical information; and
wherein the second determining code is configured to cause the at least one processor to determine the diagnosis information based on performing the named entity recognition technique.

9. The device of claim 7 , further comprising:
performing code configured to cause the at least one processor to perform a semantic role labelling technique using the medical information; and
wherein the second determining code is configured to cause the at least one processor determine the diagnosis information based on the semantic role labelling technique.

10. The device of claim 7 , wherein the reinforcement learning model has been trained by:
determining, based on the medical information, a set of potential questions;
determining respective importance scores associated with the set of potential questions that indicate the importance of the associated potential questions;
inputting the respective scores and the set of potential questions into the reinforcement learning model,
wherein the reward function maximizes the importance score.

11. The device of claim 7 , wherein the second determining code is configured to cause the at least one processor to determine the diagnosis information comprises using at least one of a recurrent neural network (RNN), a convolutional neural network (CNN), and support vector machine (SVM).

12. The device of claim 7 , wherein to determine the questions to ask the user, the first determining code is configured to cause the at least one processor to at least:
determine a set of importance scores corresponding to the plurality of potential questions based on the medical information associated with the user and the trained reinforcement learning model that has been trained using at least one of electronic medical record (EMR) data and electronic health record (EHR) data to maximize the reward function that scores the importance of the plurality of potential questions.

13. A non-transitory

Assignees

• Tencent America LLC

Inventors

• Chen Xiaozhong
• Wang Kun
• Du Nan
• Tu min
• Zhang Shangqing
• Xie Yusheng
• Fan Wei

Classifications

• G16H50/20 Primary

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

• G06N20/00

  Machine learning · CPC title

• G16H50/70

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

• G06N5/041

  Abduction · CPC title

• G06N3/006

  based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO] · CPC title