Research And Application Of Retinal Ellipsoid Zone Defect Detection Technology Based On OCT Image

Optical coherence tomography(OCT)is a kind of optical imaging technology,which has the advantages of non-contact and noninvasive.Since its invention,it has been widely used in the treatment of many diseases,especially in the retina tissue imaging.OCT is the only way to provide deep resolution information of retinal tissue.The technology can provide threedimensional images of the retinal layer and choroid layer in a few seconds and it has high resolution.With the invention of Fourier-domain OCT and it replaced the time-domain OCT,the imaging quality and the application range of OCT have been further improved.Based on that,Professor Yali Jia put forward the Split-spectrum amplitude-decorrelation angiography(SSADA)technology in 2012.The OCT images and blood flow images can be obtained simultaneously.Making it easier for doctors to contact the relationship between the retinal structure pathology and the vascular pathology.In order to get the high-quality retinopathy images,besides improving the scanning accuracy of OCT system,OCT and OCTA image postprocessing is also an important method.In the OCT image,the outer retina has four iconic high reflexes: External Limiting Membrane(ELM),Ellipsoid Zone(EZ),Interdigitation Zone(IZ)and Retinal Pigment Epithelium(RPE).The Ellipsoid Zone(EZ)is an important area of the photoreceptor,and it is the most suitable structure for evaluating the damage of photoreceptor.The integrity of the EZ is particularly useful in predicting the results of visual acuity in various retinal conditions.The detection and segmentation of the damage of the EZ is of great help to the doctors to judge the disease and predict the progress.However,because the doctor has a strong subjectivity,it is difficult to manually delineate the ellipsoid zone injury.It is urgent for the computer to automatically segment Ellipsoid Zone.Automatic segmentation of lesion areas can get more objective results and conduct a series of quantitative analysis,which can better assist doctors to make accurate judgments.Based on the urgent need of the segmentation in the Ellipsoid Zone,we have carried out three aspects of research and achieved some results.First,we used traditional fuzzy C means clustering technology to detect EZ loss in Diabetic Retinopathy(DR)disease.This method allocates the fuzzy probability values that belong to the health or damage category at the pixel level,the color chart shows the severity of the damage.Second,we used machine learning to detect ellipsoid zone in Choroideremia.In this work,we mainly use random forest classifier.According to disease characteristics and image features,we choose 12 features to learn.The result is highly accurate compared with manual results.Third,we used convolution neural network to detect EZ in the Inherited retinal degenerations(IRDS)disease.In this work,the area of interest is cut into small pieces and trained by CNN,and the good effect is gained.