Research On AMD Lesion Detection And Segmentation Based On SD-OCT Time Series Images

Age-related macular degeneration(AMD)is an aging change in the structure of the macular area,which is the leading cause of decreased visual acuity and even blindness in the elderly.Advanced age-related macular degeneration has two forms: "dry" and "wet".Geographic atrophy(GA)is "dry" AMD,and Choroidal neovascularization(CNV)is "wet" AMD.These two AMD lesions are retinopathy with a large number of people.Spectral domain optical coherence tomography(SD-OCT)has been widely used in the diagnosis of retinal diseases.Based on SD-OCT time series images,this paper studies the detection and segmentation of AMD lesions.The main content is divided into the following parts:(1)A GA detection method based on SD-OCT time series image is proposed.According to the SD-OCT time series image,the first moment image is taken as the training sample and the other moments as the testing sample.First,noise removal and layer segmentation are performed at each B-scan.In order to fully embody the outstanding features of GA,a sample construction method for limiting leveling is proposed for more effectively extracting histogram of oriented gradient(HOG)features to generate SVM models.The average of correlation coefficients and overlap ratio for GA projection area are 0.9881 and 82.62%,respectively.(2)This paper presents an automated CNV detection method based on object tracking strategy for time series SD-OCT volumetric images.In our proposed scheme,experts only need to manually calibrate CNV lesion area for the first moment of each patient,and then the CNV of the following moments will be automatically detected.In order to fully represent space consistency of CNV,a 3D-histogram of oriented gradient(3D-HOG)feature is constructed for the generation of random forest model.Finally,the similarity between training and testing samples is measured for model updating.The experiments on 258 SD-OCT cubes from 12 eyes in 12 patients with CNV demonstrate that our results have a high correlation with the manual segmentations.The average of correlation coefficients and overlap ratio for CNV projection area are 0.907 and 83.96%,respectively.(3)A segmentation method for SD-OCT time series CNV images based on 3D-Unet network is proposed.According to the SD-OCT time series image,the first moment image is taken as the training sample and the other moments as the testing sample.Firstly,noise removal and layer segmentation were performed on each B-scan.Then,3D training samples and corresponding tags were constructed by sliding Windows.Then,3D-U-Net neural network was trained to obtain CNV segmentation results.Experimental results demonstrate that this method can accurately segment CNV.(4)An AMD lesion detection and segmentation system based on SD-OCT time series images was established.The system includes SD-OCT image display,preprocessing,projection image generation,AMD lesion detection and segmentation,and result display.The overall structure design of the system is humanized,and the operation is simple and convenient.The algorithm also achieves a good precision.