Clinical decision support device, clinical decision support method, and storage medium

1 . A clinical decision support device comprising processing circuitry configured to:

acquire at least one of attribute information of a patient and first parameters regarding a first preference category for the patient by input from an external device capable of communicating via a network, or an input interface;

estimate second parameters of the patient with respect to a second preference category by inputting at least one of the attribute information and the first preference category to an estimation model;

determine preferable selections of the patient with respect to a predetermined preference category on the basis of the second parameters;

generate information that associates the preferable selections and the second parameters with each other and an image for inputting a change to the second parameters;

display the information and the image on a display;

generate vector information using an embedding technique to generate a patient attribute vector, a question vector, and an answer vector;

form a product vector of the question vector and the answer vector; and

use the patient attribute vector and the product vector to produce a rating value,

wherein the processing circuitry receives a change to the second parameters inputted through the input interface, changes the second parameters in response to the received change, and displays opinions of other patients for the same preference category on the display in association with the change to the second parameters,

wherein the estimation model is trained by Matrix Factorization over a matrix represented by preferences and attributes of the patient and one or more patient feature amounts,

wherein the processing circuitry learns a transformation function for the matrix factorization so as to minimize an error between the second parameters with respect to the second preference category and the first parameters of the same preference category included in at least one of the attribute information and the first preference category, and estimates the second parameters of the patient with respect to the second preference category using the learned transformation function, and

wherein, when the first parameters include a sentence in a natural language, the processing circuitry estimates the second parameters of the patient by performing language analysis on a patient attribute, question and answer.

2 . The clinical decision support device according to claim 1 , wherein the processing circuitry determines preferable selections of the patient with respect to a plurality of predetermined preference categories.

3 . The clinical decision support device according to claim 2 , wherein the processing circuitry further determines priorities for the preferable selections of the patient with respect to the plurality of predetermined preference categories.

4 . The clinical decision support device according to claim 3 , wherein the processing circuitry increases the priorities for the preferable selections of the patient as uncertainties for the second preference category included in the second parameters increase.

5 . The clinical decision support device according to claim 1 , wherein the second parameters are parameters including degrees of importance and uncertainties of the patient regarding the second preference category.

6 . The clinical decision support device according to claim 1 , wherein the processing circuitry displays a distribution image showing uncertainty regarding the predetermined preference category.

7 . The clinical decision support device according to claim 1 , wherein, when the first parameters include a sentence in a natural language, the processing circuitry estimates the second parameters of the patient using an embedding technique for mapping a feature amount of the sentence to a vector of another dimension.

8 . The clinical decision support device according to claim 1 , wherein the processing circuitry changes and displays opinions of other patients on the same preference category in response to change in degrees of importance for the predetermined preference category.

9 . The clinical decision support device according to claim 1 , wherein the processing circuitry

generates and displays a distribution indicating preferences of a target patient with respect to the second preference category,

generates a user-operable slide bar for the second preference category, and

generates a point on the distribution movable in association with moving the slide bar to indicate a position in the distribution representing uncertainty of the patient.

10 . A clinical decision support method, using a computer, comprising:

acquiring at least one of attribute information of a patient and first parameters regarding a first preference category for the patient by input from an external device capable of communicating via a network, or an input interface;

estimating second parameters of the patient with respect to a second preference category by inputting at least one of the attribute information and the first preference category to an estimation model;

determining preferable selections of the patient with respect to a predetermined preference category on the basis of the second parameters;

generating information that associates the preferable selections and the second parameters with each other and an image for inputting a change to the second parameters;

displaying the information and the image on a display;

receiving a change to the second parameters inputted through the input interface;

changing the second parameters in response to the received change;

displaying opinions of other patients for the same preference category on the display in association with the change to the second parameters;

generating vector information using an embedding technique to generate a patient attribute vector, a question vector, and an answer vector;

forming a product vector of the question vector and the answer vector; and

using the patient attribute vector and the product vector to produce a rating value,

wherein the estimation model is trained by Matrix Factorization over a matrix represented by preferences and attributes of the patient and one or more patient feature amounts,

wherein the method further comprises learning a transformation function for the matrix factorization so as to minimize an error between the second parameters with respect to the second preference category and the first parameters of the same preference category included in at least one of the attribute information and the first preference category, and estimating the second parameters of the patient with respect to the second preference category using the learned transformation function, and

wherein, when the first parameters include a sentence in a natural language, the estimating the second parameters of the patient comprises performing language analysis on a patient attribute, question and answer.

11 . A non-transitory computer-readable recording medium storing a program causing a computer to:

acquire at least one of attribute information of a patient and first parameters regarding a first preference category for the patient by input from an external device capable of communicating via a network, or an input interface;

estimate second parameters of the patient with respect to a second preference category by inputting at least one of the attribute information and the first preference category to an estimation model;

determine preferable selections of the patient with respect to a predetermined preference category on the basis of the second parameters;

generating information that associates the preferable selections and the second parameters with each other and an image for inputting a change to the second parameters;

display the information and the image on a display;

receive a change to the second parameters inputted through the input interface;

change the second parameters in response to the received change;

display opinions of other patients for the same preference category on the display in association with the change to the second parameters;

generate vector information using an embedding technique to generate a patient attribute vector, a question vector, and an answer vector;

form a product vector of the question vector and the answer vector; and

use the patient attribute vector and the product vector to produce a rating value,

wherein the estimation model is trained by Matrix Factorization over a matrix represented by preferences and attributes of the patient and one or more patient feature amounts,

wherein the program causes to the computer to learn a transformation function for the matrix factorization so as to minimize an error between the second parameters with respect to the second preference category and the first parameters of the same preference category included in at least one of the attribute information and the first preference category, and estimate the second parameters of the patient with respect to the second preference category using the learned transformation function, and

wherein, when the first parameters include a sentence in a natural language, the program causes the computer to estimate the second parameters of the patient by performing language analysis on a patient attribute, question and answer.

12 . A clinical decision support device comprising processing circuitry configured to:

acquire at least one of attribute information of a patient and first parameters regarding a first preference category for the patient by input from an external device capable of communicating via a network, or an input interface;

estimate second parameters of the patient with respect to a second preference category by inputting at least one of the attribute information and the first preference category to an estimation model;

determine preferable selections of the patient with respect to a predetermined preference category on the basis of the second parameters;

generate information that associates the preferable selections and the second parameters with each other and an image for inputting a change to the second parameters; and

display the information and the image on a display,

wherein the processing circuitry receives a change to the second parameters inputted through the input interface, changes the second parameters in response to the received change, and displays opinions of other patients for the same preference category on the display in association with the change to the second parameters,

wherein the estimation model is trained by Matrix Factorization over a matrix represented by preferences and attributes of the patient and one or more patient feature amounts, and

wherein the processing circuitry learns a transformation function for the matrix factorization so as to minimize an error between the second parameters with respect to the second preference category and the first parameters of the same preference category included in at least one of the attribute information and the first preference category, and estimates the second parameters of the patient with respect to the second preference category using the learned transformation function.