Multiscale 3D Graph Cut Based Neurosensory Retinal Detachment Segmentation

In order to improve the accuracy and efficiency of NRD segmentation,we propose a fully automatic segmentation method based on multiscale 3D graph cut algorithm.Firstly,we accurately locate the target and background regions based on changes in retinal thickness,and the grayscale distribution of the target and background provides prior information for the graph cut algorithm.Then,we obtain the coarse segmentation result by performing 3D graph cut algorithm on the downsampling SD-OCT image.Finally,on the original resolution image,we use 3D graph cut algorithm on the narrow-band regions on both sides of the coarse segmentation result to obtain the final segmentation result.The results of segmentation experiments on 18 sets of Cirrus SD-OCT datasets show that the Dice coefficient of the proposed algorithm is 95.07%,and the average time for segmenting a set of data is 57 seconds.Both the accuracy and efficiency of the proposed algorithm are better than existing algorithms.