Purchase this article with an account.
Ipek Oguz, Li Zhang, Kyungmoo Lee, Michael David Abramoff, Milan Sonka; 4D longitudinal choroidal thickness quantification improves reproducibility. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):5288.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Accurate and reproducible measurement of choroidal thickness is important for longitudinal studies of disease mechanisms and management. Joint segmentation of multiple 3D scans over time (termed 3D+time or 4D) has the potential to improve accuracy.
Topcon SS OCT scans were acquired every 2 weeks during anti-VEGF induction (7 scans/subject). Bruch’s membrane (BM) was segmented using the Iowa Reference Algorithms (www.iibi.uiowa.edu/octexplorer) to flatten the images. In the 3D approach, upper and lower surfaces of the choroid were segmented for each scan independently, using a graph-based algorithm. In the 4D approach, the 7 scans were segmented jointly by linking the individual graphs based on temporal registration. These links encourage temporal coherence of the segmentation while allowing up to 52μm surface position variation between each successive timepoint pair. Paired t-tests were used to compare choroidal thickness measurements obtained from the 3D and 4D approaches.
15 patients with exudative AMD were included. Compared to independent 3D segmentations, choroidal thickness was substantially more reproducible in 4D (standard deviation of thickness was 2.43μm in 4D vs. 7.05μm in 3D, p<0.001). The mean thickness (50.6 vs. 49.2μm) and temporal thinning rate (4.28 vs. 7.43μm/12 weeks) were not significantly different between the two methods (p=0.34, p=0.23), suggesting that the 4D approach is less affected by measurement noise and not exhibiting measurement bias, leading to increased temporal sensitivity.
We have developed the first automated 4D method for jointly quantifying choroidal thickness in longitudinal OCT studies. Our method is robust to image noise and produces more reliable choroidal thickness measurements than a sequence of independent 3D segmentations.
This PDF is available to Subscribers Only