Automatic Linear Lesion Segmentation And Detection In High Myopia Fundus Images

The increasing prevalence of high myopia has attracted global attention recently.Linear lesions are the main signs in high myopic eyes,and can be clearly revealed by indocyanine green angiography(ICGA).Automatic linear lesion segmentation in ICGA images can provide vital information for diagnosis,following-up examination and quantitative analysis for patients with high myopia.Since the injection of indocyanine green fluorescent dye may cause the adverse reactions to a part of patients,the automatic detection of linear lesions in retinal optical coherence tomography(OCT)image is quite meaningful to the clinical diagnosis of linear lesions.To achieve accurate segmentation of linear lesions in ICGA images,an improved conditional generative adversarial network(cGAN)based method is proposed.A new partial densely connected network is adopted as the generator of cGAN to encourage the reuse of features and make the network time-saving.Dice loss and weighted binary cross-entropy loss are added to solve the data imbalance problem.The problem is formulated as a three-class segmentation task,including linear lesions,retinal vessels and backgrounds,so that the network can be trained to learn the differences between linear lesions and retinal vessels.Experiments on the dataset from Shanghai General Hospital,including 152 ICGA images,indicated that the proposed network achieved good performance on linear lesion segmentation with Dice similarity coefficient of 69%.A novel deep learning network based method is proposed to solve the linear lesion detection problem in retinal OCT images.The U-Net with partial dense connections is adopted to segment the Bruch membrane in OCT images,which is then used to emerge the projection of Bruch membrane.Residual network is trained to classify the projection of Bruch membrane.Experiments on the dataset from Shanghai General Hospital,including 68 OCT images,showed the proposed method can achieve good performance on linear lesion detection in OCT images with accuracy of 85%.