Granted Patent B2

US 12,475,562 · App. 19/049,194 · Granted Nov 18, 2025

Inventors: Kyubo Shin (Ulsan, KR); Jongchan Kim (Ulsan, KR); Jaemin Park (Ulsan, KR)

Assignee: THYROSCOPE INC.

G06T7/0012G06V10/24G06V10/26G06V10/774G06V10/82G06T2207/20081G06T2207/20084G06T2207/30041G06T2207/30201G06V2201/03

Abstract

Claims (27)

1 . A method for estimating eye protrusion value of a subject, performed by one or more processors, comprising:

receiving, by a communication unit, a frontal facial image representing at least both eyes of the subject, a nose and eyebrows of the subject,

wherein the frontal facial image comprises a plurality of pixels, the each of the plurality of pixels comprises at least one of pixel values, wherein the pixel values are not depth value which indicates calculated distance information;

pre-processing the frontal facial image to an input data, wherein the pre-processing comprises at least segmenting a region related to at least one eye of the subject; and

estimating an eye protrusion value, which represents a degree of anterior protrusion of an eyeball, for the subject by applying the input data to a pre-trained eye protrusion value estimation model, wherein the pre-trained eye protrusion value estimation model is trained using a set of training input data and label data, wherein the training input data and label data are obtained on at least one clinical site by medical staff, wherein the training input data comprises a clinical frontal facial image is obtained by capturing an image of at least one of patient and wherein the label data is obtained by measuring a distance in an anterior direction between a corneal apex and a lateral orbital rim of an eye of the at least one of patient.

2 . The method of claim 1 , wherein the pre-trained eye protrusion value estimation model comprises at least one of Linear Regression model, Polynomial Regression model, Ridge Regression model, Lasso Regression model, Support Vector Machines (SVM) model, Decision Tree Regression model, Random Forest Regression model, K-Nearest Neighbors (KNN) model, Feed forward Neural Networks model, Convolutional Neural Networks (CNNs) model, Recurrent Neural Networks (RNNs) model, Long Short-Term Memory (LSTM) networks model, Gated Recurrent Units (GRUs) model, Gradient Boosting model, LightGBM model, CatBoost model and Adaboost model.

3 . The method of claim 1 , wherein the pre-processing of the frontal facial image comprises aligning a horizontal orientation of the frontal facial image.

4 . The method of claim 3 , wherein the aligning the horizontal orientation of the frontal facial image comprises:

determining central positions of pupils of the both eyes of the subject in the frontal facial image; and

adjusting an orientation of the frontal facial image so that a straight line connecting the central positions of the pupils of the both eyes becomes horizontal.

5 . The method of claim 1 , wherein the pre-processing of the frontal facial image comprises obtaining a plurality of radial MPLD (Mid-Pupil Lid Distance) values for at least one of the both eyes of the subject in the frontal facial image, wherein the plurality of radial MPLD values comprises a first radial MPLD value corresponding to a first angle and a second radial MPLD value corresponding to a second angle.

6 . The method of claim 5 , wherein the input data comprises a summation value of the first radial MPLD value and the second radial MPLD value.

7 . The method of claim 6 , wherein the input data further comprises a horizontal length of the at least one of the both eyes,

wherein the first angle corresponds to 0 degree,

wherein the second angle corresponds to 180 degree, and

wherein the 0 degree is defined as a direction from a central position of a pupil of the at least one of the both eyes to a lacrimal caruncle in the frontal facial image.

8 . The method of claim 7 , wherein the input data further comprises a length between the central position of the pupil of the at least one of the both eyes and a straight line connecting a leftmost and right most points of the at least one of the both eyes.

9 . The method of claim 6 , wherein the input data further comprises a vertical length of the at least one of the both eyes,

wherein the first angle corresponds to 90 degree,

wherein the second angle corresponds to 270 degree, and

wherein a 0 degree is defined as a direction from a central position of a pupil of the at least one of the both eyes to a lacrimal caruncle in the frontal facial image.

10 . The method of claim 9 , wherein the input data further comprises a length between the central position of the pupil of the at least one of the both eyes and a straight line connecting an uppermost and lowermost points of the at least one of the both eyes.

11 . The method of claim 1 , wherein the input data comprises a segmentation image representing the region related to the both eyes of the subject.

12 . The method of claim 11 , wherein the region related to the both eyes of the subject comprises eyeball areas and pupil areas of the both eyes of the subject.

13 . The method of claim 1 , wherein the pre-processing of the frontal facial image comprises obtaining a plurality of 3D facial landmark coordinate values, wherein the plurality of 3D facial landmark coordinate values comprises a first landmark coordinate value and a second landmark coordinate value.

14 . The method of claim 13 , wherein the input data comprises z-axis length value between the first landmark coordinate value and the second landmark coordinate value.

15 . The method of claim 14 , wherein the first landmark coordinate value represents a location of an outer edge of a pupil of at least one of the both eyes, and wherein the second landmark coordinate value represents a location of an outer corner of the at least one of the both eyes.

Assignments (1)

Priority Claims (1)

Continuity (2)

References Cited (21)