System and method for analyzing corneal lesion using anterior ocular segment image, and computer-readable recording medium

Patent metadata

Field	Value
Publication number	US-2025195011-A1
Application number	US-202519064892-A
Country	US
Kind code	A1
Filing date	Feb 27, 2025
Priority date	Dec 10, 2019
Publication date	Jun 19, 2025
Grant date	—

Abstract

A system for and a method of analyzing a corneal lesion using an anterior segment image according to the present invention. The system includes: an image acquisition unit configured to acquire an anterior segment image from the eyeball of a subject, a feature extractor configured to extract feature information on a position and a cause of a lesion in the cornea from the anterior segment image by applying a convolution layer to the anterior segment image through machine learning on the basis of a database in which clinical information pre-acquired by analyzing positions and causes of lesions in the corneas of subjects is stored; and a result determination unit configured to identify a position of the cornea from anterior segment image using the feature information and to analyze and determine the position and the cause of the lesion in the cornea from the position of the cornea.

First claim

1 . A system for analyzing a corneal lesion using an anterior segment image, the system comprising:
an image acquisition unit configured to acquire the anterior segment image from an eyeball of a subject; a feature extractor configured to extract feature information on a position and a cause of a lesion in a cornea of the eyeball from the anterior segment image by applying a convolution layer to the anterior segment image through machine learning on the basis of a database in which clinical information pre-acquired by analyzing positions and causes of lesions in the corneas of subjects is stored; and a result determination unit configured to identify a position of the cornea from the anterior segment image using the feature information and to analyze and determine the position and the cause of the lesion in the cornea from the position of the cornea.

2 . The system of claim 1 , wherein the feature extractor comprises:
a residual network (ResNet) obtained by stacking at least one network of a plurality of networks each including a convolution layer, a pooling layer, and an activation function or a rectified linear unit (ReLU) function and the ResNet extracts the feature information, including a multiple-channel feature map for extracting a suspicious region, from the anterior segment image.

3 . The system of claim 2 , wherein the feature extractor comprises:
a lesion guiding module configured to extract a lesion region more precise than the suspicious region through convolution of the feature information and positional data of a lesion in the clinical information, the positional data representing a position of the lesion.

4 . The system of claim 1 , wherein the feature extractor comprises:
a slit lamp mask adjustment module adjusting a slit beam image present in a region of the anterior segment image.

5 . The system of claim 4 , wherein the slit lamp mask adjustment module excludes a slit lamp portion from the anterior segment image and causes the position of the cornea or the position of the lesion in the cornea, which is included in the anterior segment image, to be learned.

6 . The system of claim 1 , wherein the result determination unit re-inputs a 3rd label feature vector output from the feature extractor and a prediction vector corresponding to a 2 nd label feature vector computed by applying a fully connected layer to the 3rd label feature vector, into the fully connected layer and thus identifies the cause of the lesion in the cornea.

7 . A method of analyzing a corneal lesion using an anterior segment image, the method comprising:
a step of acquiring, by an image acquisition unit, the anterior segment image from an eyeball of a subject; a step of extracting, by a feature extractor, feature information on a position and a cause of a lesion in a cornea of the eyeball from the anterior segment image by applying a convolution layer to the anterior segment image and a database in which clinical information pre-acquired by analyzing positions and causes of lesions in the corneas of subjects is stored; and a step of analyzing and determining, by a result determination unit, the position and the cause of the lesion in the cornea after identifying a position of the cornea from the anterior segment image using the feature information.

8 . The method of claim 7 , wherein the step of extracting, by the feature extractor, the feature information comprises:
a step of extracting the feature information including a multiple-channel feature map for extracting a suspicious region, from the anterior segment image, the feature information being destined for a residual network (ResNet) obtained by stacking at least one network of a plurality of networks each including a convolution layer, a pooling layer, and an activation function or a rectified linear unit (ReLU) function.

9 . The method of claim 8 , wherein the step of extracting the feature information comprises:
a step of extracting a lesion region more precise than the suspicious region through convolution of the feature information and positional data of a lesion in the clinical information, the positional data representing a position of the lesion, by a lesion attention module.

10 . The method of claim 7 , wherein the step of extracting, by the feature extractor, the feature information comprises, a step of adjusting, by a slit lamp mask adjustment module of the feature extractor, a slit beam image present in a region of the anterior segment image.

11 . The method of claim 7 , wherein the step of determining, by the result determination unit, the position and the cause of the lesion in the cornea comprises:
a step of outputting, by the feature extractor, a 3rd label feature vector; a step of outputting a prediction vector corresponding to a 2 nd label feature vector computed by applying a fully connected layer to the 3rd label feature vector; and a step of re-inputting, by the result determination unit, the feature vector and the prediction vector to the fully connected layer.

12 . A computer-readable storage device on which a program that performs the method according to claim 7 is recorded.

Assignees

• Samsung Life Public Welfare Foundation

Inventors

• Lim Dong Hui
• Han Gyu Le
• Kim Young Jun
• Ro Yong Man
• Lee Hye Bin

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06T2207/30096

  Tumor; Lesion · CPC title

• G06T2207/30041

  Eye; Retina; Ophthalmic · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title