Granted Patent B2

US 12,597,521 · App. 18/833,163 · Granted Apr 7, 2026

Inventor: Jae Dong Lee (Seoul, KR)

Assignee: UNIVERSITY-INDUSTRY COOPERATION GROUP OF KYUNG HEE UNIVERSITY

G16H50/20G16H10/20

Abstract

Claims (70)

1 . A survey-based diagnosis method performed by at least one computing device comprising:

acquiring response information about a diagnosee with respect to a plurality of survey questions, wherein the plurality of survey questions include a first survey question for diagnosing an energy and blood generation function, a second survey question for diagnosing an energy and blood circulation function, and a third survey question for diagnosing an energy and blood balance adjustment function; and

performing a diagnose on the diagnosee based on the acquired response information,

wherein the performing of the diagnosis comprises Steps (i) or Steps (ii):

Steps (i):

acquiring a first autoencoder-based deep learning model trained using response information about a normal class and a patient class with respect to the plurality of survey questions, wherein the first autoencoder-based deep learning model includes a first encoder, a first decoder, and a first classifier for classifying the normal class and the patient class, the first encoder and the first decoder are trained based on a first reconstruction loss value, and the first encoder is further trained based on a first classification loss value of the first classifier for a first latent vector while the first decoder is frozen;

extracting the first latent vector for the acquired response information by the first encoder;

acquiring a first confidence score for each class for the extracted first latent vector by the first classifier; and

diagnosing the health condition of the diagnosee based on the acquired first confidence score,

Steps (ii):

acquiring a second autoencoder-based deep learning model trained using response information about the patient class with respect to the plurality of survey questions, wherein the patient class consists of patients in a first disease class and a second disease class, the first disease class is a different disease class from the second disease class, the second autoencoder-based deep learning model includes a second encoder, a second decoder, and a second classifier for classifying the first disease class and the second disease class, the second encoder and the second decoder are trained based on a second reconstruction loss value, and the second encoder is further trained based on a second classification loss value of the second classifier for a second latent vector while the second decoder is frozen;

extracting the second latent vector for the acquired response information by the second encoder;

acquiring a second confidence score for each class for the extracted second latent vector by the second classifier; and

diagnosing a disease type of the diagnosee based on the acquired second confidence score.

2 . The survey-based diagnosis method of claim 1 , wherein the performing of the diagnosis comprises

generating a state vector representing a health condition of the diagnosee based on the response information; and

diagnosing the health condition of the diagnosee based on similarity between the generated state vector and a disease vector.

3 . The survey-based diagnosis method of claim 2 , wherein the generating of the state vector comprises

generating a vector for each survey question based on scores of the first survey question to the third survey question; and

generating the state vector by synthesizing the generated vector based on a weight for each survey question,

wherein the weight varies depending on demographic characteristics and disease history of the diagnosee.

4 . The survey-based diagnosis method of claim 2 , wherein the generating of the state vector comprises

acquiring an autoencoder-based deep learning model trained using response information about a patient class with respect to the plurality of survey questions;

extracting a latent vector for the acquired response information using the acquired deep learning model; and

generating the state vector based on the extracted latent vector.

5 . The survey-based diagnosis method of claim 1 , wherein the performing of the diagnosis comprises

acquiring an autoencoder-based deep learning model trained using response information about a normal class with respect to the plurality of survey questions;

calculating a reconstruction loss value for the acquired response information using the acquired deep learning model; and

diagnosing the health condition of the diagnosee based on the calculated reconstruction loss value.

6 . The survey-based diagnosis method of claim 1 , wherein the acquired response information includes a missing value for a specific survey question, and

the performing of the diagnosis comprises

acquiring an autoencoder-based deep learning model trained using response information with respect to the plurality of survey questions;

allocating a specific value to the missing value;

calculating a reconstruction loss value for the response information allocated with the specific value using the acquired deep learning model;

correcting the missing value to the specific value in response to the determination that the calculated reconstruction loss value is less than or equal to a reference value; and

performing the correction of the diagnosee based on the response information with the corrected missing value.

7 . The survey-based diagnosis method of claim 6 , wherein the acquiring of the deep learning model comprises

acquiring a first deep learning model trained using response information about a first class and a second deep learning model trained using response information about a second class with respect to the plurality of survey questions, wherein the first class is a survey respondent class having different demographic characteristics from the second class; and

determining a deep learning model that matches the demographic characteristics of the diagnosee in the first deep learning model and the second deep learning model.

8 . A survey-based diagnosis system comprising:

one or more processors; and

a memory for storing one or more instructions,

wherein the one or more processors execute one or more of the stored instructions to perform

an operation of acquiring response information about a diagnosee with respect to a plurality of survey questions, wherein the plurality of survey questions include a first survey question for diagnosing an energy and blood generation function, a second survey question for diagnosing an energy and blood circulation function, and a third survey question for diagnosing an energy and blood balance adjustment function; and

an operation of performing a diagnosis for the diagnosee based on the acquired response information,

wherein the performing of the diagnosis comprises Steps (i) or Steps (ii):

Steps (i):

acquiring a first autoencoder-based deep learning model trained using response information about a normal class and a patient class with respect to the plurality of survey questions, wherein the first autoencoder-based deep learning model includes a first encoder, a first decoder, and a first classifier for classifying the normal class and the patient class, the first encoder and the first decoder are trained based on a first reconstruction loss value, and the first encoder is further trained based on a first classification loss value of the first classifier for a first latent vector while the first decoder is frozen;

extracting the first latent vector for the acquired response information by the first encoder;

acquiring a first confidence score for each class for the extracted first latent vector by the first classifier; and

diagnosing the health condition of the diagnosee based on the acquired first confidence score,

Steps (ii):

acquiring a second autoencoder-based deep learning model trained using response information about a patient class with respect to the plurality of survey questions, wherein the patient class consists of patients in a first disease class and a second disease class, the first disease class is a different disease class from the second disease class, the second autoencoder-based deep learning model includes a second encoder, a second decoder, and a second classifier for classifying the first disease class and the second disease class, the second encoder and the second decoder are trained based on a second reconstruction loss value, and the second encoder is further trained based on a second classification loss value of the second classifier for a second latent vector while the second decoder is frozen;

extracting the second latent vector for the acquired response information by the second encoder;

acquiring a second confidence score for each class for the extracted second latent vector by the second classifier; and

diagnosing a disease type of the diagnosee based on the acquired second confidence score.

9 . A computer program connected with a computing device and stored in a non-transitory computer-readable recording medium to execute steps of:

acquiring response information about a diagnosee with respect to a plurality of survey questions, wherein the plurality of survey questions include a first survey question for diagnosing an energy and blood generation function, a second survey question for diagnosing an energy and blood circulation function, and a third survey question for diagnosing an energy and blood balance adjustment function; and

performing a diagnosis for the diagnosee based on the acquired response information,

wherein the performing of the diagnosis comprises Steps (i) or Steps (ii):

Steps (i):

acquiring a first autoencoder-based deep learning model trained using response information about a normal class and a patient class with respect to the plurality of survey questions, wherein the first autoencoder-based deep learning model includes a first encoder, a first decoder, and a first classifier for classifying the normal class and the patient class, the first encoder and the first decoder are trained based on a first reconstruction loss value, and the first encoder is further trained based on a first classification loss value of the first classifier for a first latent vector while the first decoder is frozen;

extracting the first latent vector for the acquired response information by the first encoder;

acquiring a first confidence score for each class for the extracted first latent vector by the first classifier; and

diagnosing the health condition of the diagnosee based on the acquired first confidence score,

Steps (ii):

acquiring a second autoencoder-based deep learning model trained using response information about a patient class with respect to the plurality of survey questions, wherein the patient class consists of patients in a first disease class and a second disease class, the first disease class is a different disease class from the second disease class, the second autoencoder-based deep learning model includes a second encoder, a second decoder, and a second classifier for classifying the first disease class and the second disease class, the second encoder and the second decoder are trained based on a second reconstruction loss value, and the second encoder is further trained based on a second classification loss value of the second classifier for a second latent vector while the second decoder is frozen;

extracting the second latent vector for the acquired response information by the second encoder;

acquiring a second confidence score for each class for the extracted second latent vector by the second classifier; and

diagnosing a disease type of the diagnosee based on the acquired second confidence score.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (14)