Segmentation Of Retinal Cystoid Macular Edema With Macular Hole In OCT Images

Optical coherence tomography(OCT) is a non-invasive and high-resolution imaging of human retina. It has been widely used in the diagnosis of macular edema, macular hole(MH) and glaucoma. OCT is becoming a mainstay in ophthalmology currently. Macula is located at the center of the retina and it is the projection point of sight axis. It is responsible for accurate and sensitive vision. Acute maculopathy can cause the central vision loss and even lead to the blindness in which age-related macular degeneration(AMD), cystoid macular edema(CME) and MH are the leading causes to the loss of vision. It has been proved that the volume of CMEs in the retina can be an accurate predictor of visual acuity.In this thesis, we propose an automated method to segment and quantify the volume of CMEs for abnormal retina with MH in 3D OCT images. The proposed framework consists of two parts: preprocessing, CME segmentation.(1) Preprocessing includes SNR-balancing, denoising, ten intraretinal layers segmentation and flattening, MH and vessel silhouettes exclusion. The volume of interest(VOI) can be detected;(2) In the CME segmentation, a three-step strategy is applied. First, 57 features are extracted for the pixel in the VOI. An AdaBoost classifier is employed for training and testing to generate the initialized results. Second, the detected CME regions are processed by the morphologic operations(erosion and dilation) to generate the object and background seeds for the following graph cut algorithm. The parameters for graph cut are trained and the final results can be detected by the automatic shape constrained graph cut algorithm. Finally, cyst area information has been used to remove false positives(FPs). The dataset used in the experiment includes 19 macula-centered 3D OCT volumes from 19 eyes with a full-thickness MH surrounded by CMEs. The true positive volume fraction(TPVF), false positive volume fraction(FPVF), accuracy rate(ACC) and dice similarity coefficient(DSC) for CME volume segmentation are 81.0%, 0.8%, 99.7% and 80.9%, respectively. The creativity of the thesis is that it can fully segment the CMEs for the retinas with MH and provide significant metric for clinical diagnosis.