Method and device for serialized feature extraction and classification of corneal disease images based on deep neural network

Abstract

The invention discloses a method and device for extracting and classifying serialized features of corneal disease images based on a deep neural network. The method comprises the following steps: 1) the result of regional labeling of the corneal disease slit lamp image according to the natural domain of the ocular surface-cornea is used as a training data set, and a sliding window is used to sample the main lesion area in the corneal image to form a regional sub-block set; 2 ) For all the regional sub-blocks in each corneal image, extract its features through the DenseNet model to obtain the regional vectorized feature representation; 3) Sequentially link and combine the feature extraction results, retain the spatial structure relationship between the regional sub-blocks, and use LSTM (long-short-term memory model) to process it to form corneal image features and classify them. The invention applies the deep sequence learning model to the classification and diagnosis of corneal diseases. Compared with the general image classification algorithm, the present invention models the distinguishing key information in the diagnosis of corneal diseases, and effectively retains the characteristic space structure of corneal diseases.

Description

technical field [0001] The invention relates to the field of medical image aided diagnosis, in particular to a method for extracting serialized features that maintain the spatial constraint relationship of sub-blocks in a corneal disease lesion area, and completing the classification of corneal disease image types. Background technique [0002] Using computer vision to assist medical image feature analysis and disease diagnosis is a key technology with practical application significance, and it is also a key field for the application of computer vision technology. Corneal disease is a major ophthalmic disease with a high incidence and high blindness rate in the world, especially in developing countries. There are more than 10 million corneal disease patients in the country, of which 4 million are blind or cause severe visual impairment. The use of machine learning technology to analyze and diagnose disease pictures can assist clinicians to make rapid and accurate diagnoses o...

AI Technical Summary

Problems solved by technology

However, this classification method ignores the inherent spatial pattern of visual image information representing diseases, and in the process of image feature extraction and feature compression, it is easy to ignore important but subtle differences in the process of different disease types. local nuance

Therefore, it is difficult to obtain satisfactory classification accuracy and cannot provide a reasonable explanation for the classification results

Images