Research Of Computetr Aided Diagnosis Of Important Eye Diseases Based On Image Analysis

Purpose:Retinal diseases can lead to vision loss,is the leading cause of blindness.The traditional screening method of retinal diseases is the combination of fundus imaging and doctors' subjective observation,which cannot meet the current needs of big data and precision medicine.Computer-aided diagnosis(CADx),which takes full advantage of high-speed calculation and huge storage capabilities,is embedded in specialized image processing and artificial intelligence technology to assist ophthalmologists in clinical diagnosis.Retinal CADx,combining with the technology of image enhancement and deep learning,is promising to achieve intelligent diagnosis of retinal diseases.Methods:The retinal images were acquired through the portable fundus camera.The image quality of the portable fundus images and traditional desktop fundus images were compared.The fundus image quality assessment system was established based on the human visual system,and the distorted fundus images caused by refractive medium opacification were screened.A fundus image enhancement system helped doctors diagnose retinal diseases.The database of glaucoma fundus images was established to construct CADx system based on deep learning.The feasibility of the algorithm in clinical application was verified by the sensitivity and specificity of CADx system.Results:The new portable fundus camera can capture a clinically valuable fundus image as desktop fundus image in non-mydriatic conditions.For the comprehensive evaluation of fundus images,the area under the curve(AUC)of the three kinds of SVM classifiers were 93.10%,93.14%,and 87.83%,respectively.The overall sensitivity and specificity were 91.66%and 87.45%,respectively.AUC was 94.52%.Fundus image enhancement based on contrast limited adaptive histogram equalization was used to enhance the quality of degraded fundus images.After image enhancement,the AUC of glaucoma classification was 99.6%,Diabetic retinopathy(DR)classification was 98.9%.Age-related macular degeneration(AMD)classification was 97.5%.The other retinal diseases was 97.9%and the normal fundus was 97.6%.The computer-aided diagnosis of glaucoma based on deep learning had a sensitivity of 92.1%(95%confidence interval:90.5%-93.6%)and specificity of 89.6%(95%confidence interval:87.7%-91.1%).AUC was 98.6%.Conclusion:The new portable fundus camera combined with fundus image quality assessment system can provide fundus images with clinical diagnostic value.Through the fundus image enhancement and deep learning,intelligent diagnosis of eye diseases showed high sensitivity and specificity,which provided the basis for further intelligent diagnosis of eye diseases.