Automated Fluid/Cyst Segmentation: A Quantitative Assessment of Diabetic Macular Edema

Purpose : Diabetic macular edema (DME), manifested by fluid cysts within the retina and retinal thickening, is the most common cause of moderate vision loss in diabetes. Retinal thickness, as measured by optical coherence tomography (OCT), indirectly measures the cystic spaces. However, there may be utility to measuring the actual cystic volume instead. This is difficult to test because commercially available algorithms do not measure this and manual segmentation of fluid/cyst regions is time consuming. We propose a fully-automated algorithm to segment fluid/cyst regions in OCT images of DME.

Methods : Our algorithm uses a novel neutrosophic transformation and a graph-based shortest path method. In the neutrosophic domain, an image g is transformed into three sets: T, I and F. First, a new method is introduced to compute the indeterminacy set I, and a new lambda correction operation is introduced to compute the set T in the neutrosophic domain. Second, a graph shortest-path method is applied in the neutrosophic domain to segment the inner limiting membrane (ILM) and retinal pigment epithelium (RPE) to bound the region of interest (ROI), and the outer plexiform layer (OPL) and inner segment myeloid (ISM) using a novel edge weight definition. Third, a new cost function for cluster-based fluid/cyst segmentation in the ROI is presented. Finally, the final fluid regions are identifed by ignoring the regions between middle layers. The results of the different proposed steps are shown in Fig. 1.

Results : Segmentation results of the proposed method are compared with manual segmentation results by two ophthalmologists in two publicly available datasets: Duke, Optima and a local dataset from the UMN clinic. The proposed algorithm achieves sensitivity of 67.3%, 88.8% and 76.7%, precision of 64.98%, 73.89% and 74.91%, and dice coefficients of 57.51%, 70.52% and 69.40% for the Duke, Optima, and the UMN datasets, respectively.

Conclusions : Our segmentation method automatically and highly accurately segments fluid/cyst abnormal regions in retinal of DME subjects. Accurate segmentation of DME cysts will allow investigation of new biomarkers for guiding patient care and predicting outcomes.