Abstract
Purpose :
Accurate and reproducible segmentation of fluid-filled regions from retinal OCT images is an important step allowing quantification of the disease status, longitudinal disease progression, and response to therapy in wet-pathology retinal diseases. However, segmentation of fluid-filled regions from OCT images is a challenging task due to their inhomogeneous appearance, the unpredictability of their number, size, location, and image appearance.
Methods :
We developed a knowledge-based approach that leverages a carefully designed cost function and graph-based segmentation techniques to provide a vendor-independent solution to this problem. The cost function reflects the known properties of the SEAD’s (Symptomatic Exudate Associated Derangements) in a layer-specific manner. Retinal tissue characteristics are determined directly from the image, rather than enforcing a priori intensity models. We evaluate this approach on two publicly available datasets. The OPTIMA Cyst Segmentation Challenge training data consists of 15 AMD patients acquired with devices from 4 different vendors (Spectralis, Cirrus, Topcon, Nidek), with 49 to 128 B-scans per OCT volume. The DME dataset (http://people.duke.edu/~sf59/Chiu_BOE_2014_dataset.htm) consists of 10 Spectralis scans with 61 B-scans each.
Results :
Compared to a previous machine-learning based approach, the absolute volume similarity error in the DME dataset was dramatically reduced from 81.3 ± 56.4% to 43.0 ± 37.4% (paired t-test, p << 0.01). Comparison to the first (second) manual rater resulted in 87% vs. 46.4% (75.2% vs. 39.5%) volume dissimilarity for the previous and new methods. Figs. 1 and 2 present qualitative results.
Conclusions :
The preliminary results are highly promising for robust vendor-independent segmentation of fluid-filled regions from OCT scans in exudative retinal diseases.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.