Automated subretinal fluid volume quantification using multi-surface segmentation and sparse high order potentials

In patients suffering from acute central serous chorioretinopathy (CSC), quantification of subretinal fluid pockets (SFPs) is a clinically relevant finding to evaluate the pathology state and its progress over time. Automatic segmentation of SFPs is a challenging problem, given their diversity in terms of size, location and shape. In small SFPs, the fluid amount is reduced and consequently the contrast of their bottom and upper limits surfaces is lower, which compromises the accuracy of the segmentation. In this work, we propose an method for segmenting the limiting boundaries of SFPs, which is based on a multi-surface segmentation framework using graph models combined with sparse high order potentials (SHOPs). This algorithm was specifically developed to deal with the morphological variability caused by SFPs. The algorithm was evaluated in terms of volume using true and false positive rates and the overlap ratio in 18 SD-OCT volumes, obtaining average rates of 0.96 ± 0.05, 0.03 ± 0.03 and 0.68 ± 0.16, respectively. In most cases, the results of our method are comparable to manual segmentations, which suggests that it can be useful for clinical practice.