Modeling method and apparatus for diagnosing ophthalmic disease based on artificial intelligence, and storage medium

Patent metadata

Field	Value
Publication number	US-11636340-B2
Application number	US-201816967087-A
Country	US
Kind code	B2
Filing date	Apr 17, 2018
Priority date	Apr 17, 2018
Publication date	Apr 25, 2023
Grant date	Apr 25, 2023

Abstract

The present disclosure proposes a modeling method and apparatus for diagnosing an ophthalmic disease based on artificial intelligence, and a storage medium. The modeling method includes: establishing a data collection of ophthalmic images and a data collection of non-image ophthalmic disease diagnosis questionnaires; training a first neural network model by employing the data collection of the ophthalmic images to obtain a first classification model; training a second classification model by employing the data collection of non-image ophthalmic disease diagnosis questionnaires; and merging the first classification model and the second classification model to obtain a target classification network model, in which, a test result outputted by the target classification network model is used as a diagnosis result of the ophthalmic disease.

First claim

What is claimed is:

1. A modeling method for diagnosing an ophthalmic disease based on artificial intelligence, comprising:
establishing a data collection of ophthalmic images and a data collection of non-image ophthalmic disease diagnosis questionnaires;
training a first neural network model by employing the data collection of the ophthalmic images to obtain a first classification model;
training a second classification model by employing the data collection of non-image ophthalmic disease diagnosis questionnaires; and
merging the first classification model and the second classification model to obtain a target classification network model, wherein a test result outputted by the target classification network model is used as a diagnosis result of the ophthalmic disease,
wherein training the first neural network model by employing the data collection of the ophthalmic images to obtain the first classification model comprises at least one of:
training the first neural network model by employing the data collection of ophthalmic images, in which, the first neural network model comprises: at least two convolution layers, an activation function, at least two full connection layers, and a sigmoid activation function, and each convolution layer is connected to a pooling layer; and
migrating an existing neural network model trained based on massive public data collections to a supervised learning model, and taking the supervised learning model as the first classification model.

2. The method of claim 1 , wherein after training the first neural network model by employing the data collection of the ophthalmic images, the method further comprises:
performing fine adjustment on parameters of the convolution layers in the first neural network model, and training the first neural network model subjected to the fine adjustment by employing the data collection of the ophthalmic images; and
migrating the existing neural network model trained based on the massive public data collections to the supervised learning model comprises:
migrating the first neural network model subjected to the fine adjustment to the supervised learning model.

3. The method of claim 2 , wherein migrating the existing neural network model trained based on the massive public data collections to the supervised learning model comprises:
migrating the existing neural network model trained based on the massive public data collections to the supervised learning model by employing a multitasking learning mechanism.

4. The method of claim 1 , wherein migrating the existing neural network model trained based on the massive public data collections to the supervised learning model comprises:
migrating the existing neural network model trained based on the massive public data collections to the supervised learning model by employing a multitasking learning mechanism.

5. The method of claim 1 , wherein merging the first classification model and the second classification model to obtain the target classification network model comprises:
respectively intercepting a part of the convolution layers in the first classification model as a network feature and a feature of the second classification model;
merging the network feature and the feature of the second classification model by employing a dense connection network or a batch normalization technology to obtain a merged feature, and training and obtaining the target classification network model by employing the merged feature.

6. The method of claim 1 , wherein establishing the data collection of the ophthalmic images and the data collection of the non-image ophthalmic disease diagnosis questionnaires comprises:
enhancing initial ophthalmic image data by employing an image preprocessing algorithm, the image preprocessing algorithm comprising at least one of: an algorithm for randomly deforming an image, an algorithm for shearing an image, and an algorithm for compensating a color or brightness of an image;
generating image simulation data corresponding to enhanced ophthalmic image data as the data collection of the ophthalmic images; and
extracting the data collection of the non-image ophthalmic disease diagnosis questionnaires from an interrogation condition.

7. The method of claim 6 , wherein enhancing the initial ophthalmic image data by employing the image preprocessing algorithm further comprises:
segmenting a target feature in the initial ophthalmic image data to obtain the enhanced ophthalmic image data, wherein the target feature is a fundus vascular feature or a macular feature.

8. The method of claim 6 , wherein the data collection of the non-image ophthalmic disease diagnosis questionnaires comprises: symptom information of the ophthalmic diseases and personal information of patients, and training the second classification model by employing the data collection of the non-image ophthalmic disease diagnosis questionnaires comprises:
obtaining classification information of the ophthalmic diseases to obtain a plurality of categories of the ophthalmic diseases;
selecting a target category of ophthalmic diseases from the plurality of categories of the ophthalmic diseases by employing a decision tree algorithm; and
iteratively training the second classification model by employing the target category of the ophthalmic diseases, symptom information of an ophthalmic disease and personal information of a current patient,
in which, the second classification model is configured to analyze the symptom information of the ophthalmic disease and the personal information of the current patient based on match information of each category of ophthalmic diseases.

9. The method of claim 6 , wherein generating the image simulation data corresponding to the ophthalmic image data enhanced as the data collection of the ophthalmic images comprises:
generating the image simulation data corresponding to the enhanced ophthalmic image data as the data collection of the ophthalmic images by employing a generative adversarial network.

10. The method of claim 1 , wherein the data collection of the non-image ophthalmic disease diagnosis questionnaires comprises: symptom information of the ophthalmic diseases and personal information of patients, and training the second classification model by employing the data collection of the non-image ophthalmic disease diagnosis questionnaires comprises:
obtaining classification information of the ophthalmic diseases to obtain a plurality of categories of the ophthalmic diseases;
selecting a target category of ophthalmic diseases from the plurality of categories of the ophthalmic diseases by employing a decision tree algorithm; and
iteratively training the second classification model by employing the target category of the ophthalmic diseases, symptom information of an ophthalmic disease and personal information of a current patient,
in which, the second classification model is configured to analyze the symptom information of the ophthalmic disease and the personal information of the current patient based on match information of each category of ophthalmic diseases.

11. The method of claim 10 , wherein the symptom information of the ophthalmic disease comprises at least one of: red eye information, eye dryness information, eye pain information, eye itch information, foreign body sensation information, burning sensation information, photophobia information, and tears information.

12. The method of claim 10 , wherein the personal information of the patients comprises at least one of: age, gender, infection progress, and high-risk factors.

13. A modeling apparatus for diagnosing an ophthalmic disease based on artificial

Assignees

• Bgi Shenzhen
• Eye Ear Nose And Throat Hospital Of Fudan Univ

Inventors

• Liu Xiaoqing
• Hong Jiaxu
• Ni Yong
• Li Shuangshuang
• Wang Lili
• He Wei
• Guo Youwen
• Liu Yuxuan
• Liu Yong
• Wang Wei
• Xu Ruiqi
• Cheng Jingyi
• Tian Lijia
• Chen Wenbin
• Xu Xun

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N3/094

  Adversarial learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0475

  Generative networks · CPC title