Purchase this article with an account.
Manuel Falcao, Carlos Figueiredo, Fernando Carvalho, Carlos Silva, Jorge Oliveira, Susana Penas, Amândio Sousa, Fernando Falcao-Reis; Automatic Segmentation and Quantification of Subretinal Fluid on Acute Central Serous Chorioretinopathy. Invest. Ophthalmol. Vis. Sci. 2017;58(8):676. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Identifying and quantifying subretinal fluid (SRF) in patients with acute central serous chorioretinopathy (CSC) is of paramount importance for the management of these patients. We aimed to develop an automatic algorithm that would be able to detect and quantify subretinal fluid on spectral domain OCT (Heidelberg Spectralis®) in patients with the acute form of the disease and to compare with manual segmentation.
Eighteen eyes of 18 patients with acute CSC were imaged on SD OCT using a macular cube centered on the fovea composed of 19 B-scans averaged 9 times. Each OCT B-Scan is composed by 509x19x432 voxels with resolution of 11.03x235x3.87 . In each B-Scan subretinal fluid was identified pixel by pixel and quantified manually using GIMP image analysis software (Fig 1). An algorithm was created to identify subretinal fluid in the scans. The proposed segmentation was based on multi-surface segmentation using graph models controlled by intensity gradient and hard and soft constraints. Additionally, on the last processing phase, sparse high order potentials and the detection of the horizontal extremities of pathology were used to improve the subretinal fluid pockets upper limit segmentation (Fig 2). A comparison between the two segmentation methods was performed. The algorithm performance was evaluated using true positive (sensitivity) and false positive rates for each OCT volume.
The mean number of pixels identified as subretinal fluid in SD-OCT volumes was 54039±70210 for the manual measurements and 69219±76539 for the automatic measurements (p<0.001). In all cases, the pixels identified by the algorithm were more than in the manual selection. The sensitivity of the algorithm was 0.96±0.04 and the false positive rate was 0.04±0.02.
Even though the automatic algorithm detected larger areas of subretinal fluid than the manual identification process, the proposed segmentation method developed has a high sensitivity for detecting subretinal fluid automatically and can be used clinically to detect the areas. Corrections in the algorithm can be made to improve the results. However, compared to manual segmentation this is a useful tool for the automatic quantification, management and follow-up of patients with acute CSC. Further improvement of the algorithm may lead to its widespread use in clinical practice.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
Manual Segmentation of SRF
Automatic Segmentation of SRF
This PDF is available to Subscribers Only