Detection And Study Of The Nerve Fiber Layer Of SD-OCT Retinal Image

Glaucoma is a second leading cause of blindness in human eyes which results in the increase of the cup-to-disk ratio and thinning the thickness of retinal nerve fiber layer(RNFL)around optic nerve head(ONH).So the measurement of the thickness of the retinal nerve fiber layer can assist in the diagnosis of glaucoma and other retinal diseases.Currently,there are many types of medical devices that can capture the retinal image.Especially spectral domain optical coherent tomography(SD-OCT)can clearly demonstrate the structure of the retinal layer.This paper uses machine learning technology to segment the RNFL of SD-OCT retinal images around optic nerve head which is crucial in the diagnosis of glaucoma.The main research of the paper is as follows:(1)Retinal layer structure is complex which is easy to be changed by some retinal diseases.For example,age-related macular degeneration(AMD)makes the retinal pigment epithelium up and glaucoma resulting in thinning the thickness of RNFL.The traditional segmentation algorithms cannot be directly applied to segment the retinal layers.In the third chapter,we put forward a method based on random forest to segment the RNFL which can be divided into three parts:The first step is the preprocessing,including image denoising and capture the region of interest(ROI).We finally choose BM3D filter for image denoising by comparing the effect of several image denoising algorithms.The ROI is limited between the Internal Limiting Membrane(ILM)and retinal pigment epithelium(RPE)because other regions are not reflected and look like black on retinal images.The second step is the feature extraction and feature training and the last step uses the training model from the second step to classify the other features to obtain the rough segmentation results.Finally we use polynomial fitting to smooth the results.(2)Glaucoma usually generated with multiple distinct characteristics,such as the RNFL missed or cup-to-disc ratio changed.Therefore,to evaluate the correlation of these features is of great significance to the study of the pathogenesis of glaucoma.In order to calculate the cup disc ratio and other parameters,we should segment the disk and cup accurately.In the fourth chapter,a new pitch search algorithm based on SVM is introduced in detail which can be divided into two parts:firstly,we use the relationship between RPE break points and disk to roughly locate the disk.Secondly,we develop a patch search method to find the most likely patch centered at the RPE break points and refine the segmentation results.In the last,we analyze the correlation between RNFL thickness and some parameters.(3)We develop a OCT retinal image segmentation and 3D reconstruction software which can segment five layers at the same time,and is robust to noise and diseases.