Premature infant retinopathy automatic partition recognition method based on attention mechanism and deep supervision strategy

Abstract

The embodiment of the invention discloses a premature infant retinopathy automatic partition recognition method based on an attention mechanism and a deep supervision strategy. The premature infant retinopathy automatic partition recognition method comprises the steps: preprocessing an image, sampling a two-dimensional retina fundus color photo image to 256 * 256 through bilinear interpolation, carrying out mean value reduction processing, carrying out online data amplification operation on the data, constructing a network structure: setting a space channel attention module SACAB in a DenseNet121 convolutional neural network, and introducing a deep supervision strategy to construct the network structure, training and testing the model, taking the DenseNet121 convolutional neural network pre-trained on the ImageNet as a pre-training model through transfer learning, training a network structure through data in a training set, and testing the performance of the network structure through atest set after the network structure is trained. According to the invention, automatic classification and identification of the I region / II region / III region in the retinal fundus color photograph image of the premature infant are realized, and a foundation is laid for subsequent ROP automatic diagnosis.

Description

technical field [0001] This application relates to the technical field of retinal image classification methods, in particular to an automatic partition recognition method for retinopathy of prematurity based on attention mechanism and deep supervision strategy. Background technique [0002] Retinopathy of Prematurity (ROP) is a retinal vascular proliferative disease, one of the most dangerous and serious eye complications of premature infants, and the main cause of blindness in children worldwide. In most cases, ROP is curable if diagnosed in time and treated properly. International classification of ROP (Internationalclassification of ROP, ICROP) defines three clinical examination parameters: area, stage and additional disease (Plus). According to the location of the lesion, ICROP defines the clinical ROP divisions as zones I, II, and III. The closer the lesion is to the posterior pole, the more severe the condition is. That is, lesions in zone I are the most severe and l...

AI Technical Summary

Problems solved by technology

However, the macular development of newborns is not complete, and the macular structure is not obvious in the corresponding fundus color photos, which may lead to a low recognition accuracy of the macula, which in turn affects the recognition accuracy of the ROP I area

Finally, the algorithm only realizes the automatic recognition of the I zone, and does not involve the research on the automatic recognition of the II zone and the III zone

Currently, no studies have been reported on the automatic identification of the three regions of ROP, which is important for assessing the severity of ROP

Images