Diabetic retina image automatic classification method

An automatic classification and diabetes-based technology, which is applied in the direction of instruments, character and pattern recognition, and recognition of medical/anatomical patterns, etc., can solve problems such as poor classification performance, few retinal images, and difficult extraction of retinal image features to achieve accurate classification results , the effect of good robustness

CN108009592AInactive Publication Date: 2018-05-08YUNNAN UNIV
9 Cites 11 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2018-05-08
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a diabetic retina image automatic classification method. According to the method, first, denoising, normalization and other preprocessing are performed on images; second, underthe condition that sample data is insufficient, a migration learning strategy and a data enhancement strategy are adopted to extract depth features for classification; and finally the extracted features are input into a classifier to classify retina pathological images into five types. Through the method, the classification accurate rate can reach 93%, and the method has good robustness and generalization performance.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention is an automatic classification method for diabetic retinal images, which is applicable to the technical fields of machine learning, pattern recognition and medical image processing. Background technique

[0002] The automatic classification of diabetic retinal pathological images has important clinical application value. In the classification of retinal pathological images, extracting representative and discriminative features is the key factor to achieve a good classification effect. The current classification method based on artificial pathological images , mainly has the following limitations: (1) Image quality. The quality of the collected retinal images is easily affected by many other factors such as illumination, lens, machine equipment, and image acquisition personnel's experience; (2) Doctor's personal experience .Doctors usually assess and determine the degree of retinal lesions by visually inspecting retinal images, but the fea...

Examples

Embodiment Construction

[0019] See figure 1 , figure 2 , image 3 , a diabetic retinal image automatic classification method, the present invention is characterized in that:

[0020] 1) The data comes from the Diabetic Retinopathy Detection competition in the data modeling and data analysis competition platform (kaggle). The retinal images in this data set are all high-resolution RGB images, and the retinal images are divided into normal and mild lesions according to the degree of lesions , moderate lesions, severe lesions, and proliferative lesions;

[0021] 2) Perform preprocessing such as denoising, histogram equalization, normalization, black border removal, data enhancement and feature analysis on the retinal image data;

[0022] 3) In order to avoid problems such as slow convergence caused by changes in data distribution during the training process of the model, a BNnet is obtained by introducing a batch normalization layer before each convolutional layer and fully connected layer on the ba...