Fundus disease visual analysis method based on computer vision and deep learning

A technology for computer vision and fundus diseases, applied in neural learning methods, calculations, image analysis, etc., to achieve wide applicability, improve reliability and intuition

Pending Publication Date: 2022-03-08
BEIJING UNIV OF TECH
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] However, simple visualization processing can only roughly find out the degree of attention of the model to certain areas in the image from the category activation map. If you need to further obtain key information in the image or summarize the rules in it, you need to use it. mathematical method

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Fundus disease visual analysis method based on computer vision and deep learning
  • Fundus disease visual analysis method based on computer vision and deep learning
  • Fundus disease visual analysis method based on computer vision and deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach

[0046] The flow chart of the implementation is as figure 1 shown, including the following steps:

[0047] Step S10, establishing a fundus lesion labeled image database;

[0048]Step S20, visualizing the region of interest of the model;

[0049] Step S30, the visualization area is superimposed;

[0050] Step S40, regionalized image feature screening and statistics;

[0051] Step S50, statistical data association analysis.

[0052] The model attention area visualization processing step S20 of the embodiment also includes the following steps:

[0053] Step S200, select a deep learning model that can be used for fundus disease image analysis to train the fundus disease images, and obtain the output weight of the model.

[0054] Step S210, loading the obtained model output weights, and performing multi-level visualization processing on the model by using the corresponding labeled images of fundus lesions to obtain category activation maps (heat maps) of different levels.

[0...

Embodiment

[0068] figure 1 A schematic flow chart of a method for visualizing and analyzing fundus diseases based on computer vision and deep learning is shown, taking diabetic retinopathy as an example. It is mainly divided into two parts: model focus area visualization and data statistical analysis.

[0069] The embodiment adopts a DenseNet-based image classification model and fundus images of patients in a certain hospital, and the size of the image file used for classification is generally around 300K.

[0070] First, according to step 1, use the labelimg software to manually label and combine the lesions in the medical fundus disease images in units of categories, and the obtained lesion label images are as follows: figure 2 shown.

[0071] According to step 2, use the CAM-based improved algorithm to visualize the output of the 4 DenseBlock structures of DenseNet to obtain multi-level convolutional layer output weights, and obtain the output category activation maps of each level...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a fundus disease visual analysis method based on computer vision and deep learning, and the method comprises the steps: carrying out the visual processing of a result of an existing fundus disease model, and marking a key region which supports the model to make a final decision; superposing a category activation image generated by a visualization algorithm with the original fundus image so as to screen image information in the model attention area; further analyzing the features in the fundus image by using a statistical algorithm according to the screening result, and associating the focus features with the fundus disease analysis result obtained by the model. Compared with an existing method, the fundus disease visualization method has the advantages that deep and shallow output of the network is considered, so that the decision process of the model can be analyzed more comprehensively, and meanwhile, a new analysis method is provided for mechanism research of fundus diseases.

Description

technical field [0001] The invention belongs to deep learning model visualization technology, in particular to a visual analysis method for fundus diseases based on computer vision and deep learning. The decision-making process of the model is explained and analyzed by using the method of class activation map mapping combined with statistical algorithms. Background technique [0002] Clinical studies have shown that specific changes in the fundus will inevitably affect people's lives, and severe retinopathy can even lead to blindness. Therefore, regular detection and diagnosis of fundus lesions are of great significance for timely detection and treatment of patients. With the development of the times, the emergence of computer vision and deep learning technology provides new technical support for the diagnosis and treatment of fundus diseases. [0003] However, due to the lack of interpretability of the decision-making basis of the deep learning model, in order to better an...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/90G06T7/62G06T7/136G06T7/11G06T7/00G06N3/08G06N3/02
CPCG06T7/0012G06T7/90G06T7/11G06T7/136G06T7/62G06N3/02G06N3/08G06T2207/30041
Inventor 李煜朱美龙孙光民陈佳阳李侨宇汤长新
Owner BEIJING UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap