Method and system for simultaneous classification and regression of clinical data

Patent metadata

Field	Value
Publication number	US-12462380-B2
Application number	US-202217725051-A
Country	US
Kind code	B2
Filing date	Apr 20, 2022
Priority date	Apr 23, 2021
Publication date	Nov 4, 2025
Grant date	Nov 4, 2025

Abstract

This disclosure discloses a method for analyzing clinical data. The Method includes extracting a first feature information by applying a neural network to the clinical data; predicting a disease status related parameter by applying a regression model to the extracted first feature information; generating a second feature information based on the extracted first feature information and the disease status related parameter; and predicting a disease status classification result by applying a classification model to the second feature information. The method can improve the prediction accuracy and the diagnosis efficiency of doctors.

First claim

What is claimed is:

1 . A computer-implemented method for analyzing clinical data, comprising:
extracting, by a processor, a first feature information by applying a neural network to a medical image containing a vessel in the clinical data; predicting, by the processor, a disease status related parameter by applying a regression model to the extracted first feature information; generating, by the processor, a second feature information based on the extracted first feature information and the disease status related parameter; and predicting, by the processor, a disease status classification result by applying a classification model to the second feature information; wherein the disease status related parameter includes an estimation score of a Fractional Flow Reserve (FFR) of the vessel or a plaque vulnerability risk score of the vessel, and the disease status classification result includes a stenosis level or a plaque vulnerability level of the vessel.

2 . The computer-implemented method of claim 1 , wherein,
the stenosis level of the vessel includes a first level indicative of no stenosis, a second level indicative of a non-significant stenosis, and a third level indicative of a significant stenosis; or the plaque vulnerability level of the vessel includes a first level indicative of non-vulnerable, a second level indicative of vulnerability at a low risk, and a third level indicative of vulnerability at a high risk.

3 . The computer-implemented method of claim 1 , wherein:
the regression model and the classification model are trained jointly by using a joint loss function, which includes a regression loss term and a penalty term, wherein the penalty term is designed to penalize the regression model for predicting a value belonging to a different disease status classification result from that of a regression ground truth.

4 . The computer-implemented method of claim 3 , wherein:
in case that the value predicted by the regression model and the regression ground truth are distributed at the same side of a preset threshold for disease status classification, the penalty term is set to zero, and wherein in case that the value predicted by the regression model and the regression ground truth are distributed at different sides of the preset threshold, the penalty term is designed to increase as the deviation between the regression predicted value and the regression ground truth increases.

5 . The computer-implemented method of claim 4 , wherein the penalty term is a threshold regularization loss and is represented by:

{

P
-
t

GT
R

-
t

>
0

,

L
=
0

P
-
t

GT
R

-
t

≤
0

,

L
=

exp
⁡
(
X
)

,

X
=

(

P
-

GT
R

)

2

wherein, P represents the regression predicted value, t represents the preset threshold, GTR represents the regression ground truth, and L represents the penalty term.

6 . The computer-implemented method of claim 1 , wherein generating a second feature information based on the extracted first feature information and the disease status related parameter further comprises:
extending the extracted first feature information into a one-dimensional vector and concatenating the one-dimensional vector with the disease status related parameter, to obtain the second feature information.

7 . A computer-implemented method for analyzing clinical data, comprising:
extracting, by a processor, a first feature information by applying a neural network to a medical image containing a vessel the clinical data; predicting, by the processor, a disease status classification result by applying a classification model to the extracted first feature information; transforming, by the processor, the predicted disease status classification result into a one-hot representation and fusing the one-hot representation with the first feature information, to generate a second feature information; and predicting, by the processor, a disease status related parameter by applying a regression model to the second feature information; wherein the disease status related parameter includes an estimation score of an FFR of the vessel or a plaque vulnerability risk score of the vessel, and the disease status classification result includes a stenosis level or a plaque vulnerability level of the vessel.

8 . The computer-implemented method of claim 7 , wherein,
the stenosis level of the vessel includes a first level indicative of no stenosis, a second level indicative of a non-significant stenosis, and a third level indicative of a significant stenosis; or the plaque vulnerability level of the vessel includes a first level indicative of non-vulnerable, a second level indicative of vulnerability at a low risk, and a third level indicative of vulnerability at a high risk.

9 . The computer-implemented method of claim 7 , wherein,
the regression model and the classification model are trained jointly by using a joint loss function, which includes a regression loss term weighted by a penalty weight, wherein the penalty weight is designed to penalize the regression model for predicting a value belonging to a different disease status classification result from that of a regression ground truth.

10 . The computer-implemented method of claim 9 , wherein
the penalty weight in

Assignees

• Shenzhen Keya Medical Tech Corporation
• Shenzhen Keya Medical Tech

Inventors

• Wang Xin
• Yin Youbing
• Kong Bin
• Lu Yi
• Guo Xinyu
• Yang Hao-Yu
• Song Qi

Classifications

• G16H50/20 Primary

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

• G06V2201/03

  Recognition of patterns in medical or anatomical images · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/30101

  Blood vessel; Artery; Vein; Vascular · CPC title