DR Fundus Image Quality Classification And Lesion Identification Based On Residual-dense Network

Fundus image-assisted diagnosis is an intuitive and standard diagnostic technique which has developed rapidly in recent years.In hospital treatment and telemedicine,experts conduct pathological analysis and treat patients with diabetic retinopathy(DR)using retinal images that are captured by fundus cameras.However,the different experiences of fundus photographers result in different image qualities,and some images are marked as unreadable by the diagnostician which seriously affects the accuracy of fundus diseases diagnosis by ophthalmologists.Secondly,a large number of non-pathological fundus images occupy too much diagnostic time of ophthalmologists,delaying the timely treatment of patients and causing irreversible harm.Therefore,in order to ensure the quality of fundus images while reducing the time and effort of manual screening for DR images,it is necessary and urgent to evaluate the fundus image quality and diabetic retinopathy automatically and objectively during the acquisition process.To solve the problem of fundus image quality classification and DR differentiation,a modified residual dense block convolution neural network(MRDB-CNN)based on deep learning is proposed in this thesis.For the research of fundus image quality classification,the two categories of images,good quality and poor quality,are used as the training data set and testing data set to train the CNNs and verify the classification results.And the quality classification performance of the network is further verified by testing the ambiguous quality category.For the research of the DR differentiation,the training is carried out by using the fundus images with DR and those without DR,and the network structures are evaluated by the accuracy of the test set.In short,according to the two parts of experiments,the MRDB-CNN network avoids the complex preprocessing of traditional algorithms and has higher accuracy for fundus image quality classification and DR differentiation than other networks.In addition,compared with other network blocks,the MRDB can obtain more detailed features of fundus quality or different diabetic retinopathies,can obtain better generalization ability and higher classification accuracy network classification models.The accuracy of the two studies reached 99.90% and 94.90%,respectively,and it satisfies the needs of hospital treatment and telemedicine for the real-time quality judgement and DR differentiation for fundus images.