Automated Diagnosis And Segmentation Of Retinal Artery Occlusion In SD-OCT Images

Retinal artery occlusion is a common retinal pathology caused by closure of ocular vessels.In clinical,patients with RAO usually suffer from sudden visual loss or hypopsia,and their visual impairment and visual field defect can be partly recovered.Central retinal artery occlusion(CRAO)and branch retinal artery occlusion(BRAO)are two forms of RAO with highest incidences.According to the different occlusion degrees and disease phases,there are considerable variations in the treatments for patients with RAO.Therefore,early diagnosis of RAO is very important that RAO patients can get right treatment and avoid getting blind.In this paper,we mainly study the methodology to classify and segment retinal artery occlusion based on spectral-domain optical coherence tomography(SD-OCT)images.The main works and contributions are as follow:First,we propose an automated method for RAO classification based on AdaBoost classifier,which differentiates normal retina and four types of RAO,including CRAO acute phase,CRAO chronic phase,BRAO acute phase and BRAO chronic phase.The feature set consists of local binary pattern(LBP)texture feature,LBP shape feature and thickness model.Then we reduce the dimension of the feature set and the AdaBoost classifier is used for classification.The experiment result shows that the proposed method can effectively classify RAO with high accuracy.Then,we segment BRAO regions of acute phase based on Graph Search-Graph Cut method.First,we apply bayesian posterior probability for rough initialization.Then we detect BRAO regions based on the initialization results and generate BRAO object seeds and background seeds for segmentation.Last,the Graph Search-Graph method is applied for refine segmentation.Last,we propose an inner retina thickness model to segment BRAO regions of chronic phase.We build inner retinal thickness model based on normal retinas and the thickness model is used as thresholds for segmentation of BRAO regions in chronic phase.The experiment result shows that the proposed methods can effectively and quickly segment BRAO.