Abstract
Purpose :
In anti-VEGF therapy of neovascular AMD, inter-individual treatment requirements are vastly heterogeneous. Tools and biomarkers to predict these individual requirements represent an unmet medical and socioeconomic need. The aim of this study was to predict anti-VEGF injection requirements during the pro re nata (PRN) phase, using a set of OCT images acquired during the loading phase in treatment-naïve patients with neovascular AMD.
Methods :
Prospective clinical trial data of 288 evaluable subjects receiving PRN ranibizumab therapy according to protocol specified criteria in the HARBOR study after 3 initial monthly injections were included. SD-OCT images (512x128x1024 voxels, Cirrus, Zeiss) were analyzed at baseline, month 1 and month 2. Quantitative features based on automated segmentation of layers and fluid regions were extracted to describe the retinal microstructure. Fluid segmentations were based on deep learning and layer segmentations using a graph-theoretic approach. Features included inner retina, outer nuclear layer, photoreceptor outer segments with retinal pigment epithelium, and total retinal thickness (TRT) as well as intra- and subretinal fluid (IRF and SRF) volume and area (Figure 1). An ETDRS grid was used to compute regional features. Groups of low and high injection requirements were defined as ≤2 and ≥9 injections between month 3 and month 12, respectively. Random forest classification was used to predict the low and high treatment categories and was evaluated with a ten-fold cross validation.
Results :
The number of injections during the PRN phase until month 12 ranged from 0 to 10. 51/288 patients showed low (≤2) and 49/288 patients showed high (≥9) injection requirements. The classification results were evaluated as area under the ROC curve (AUC). Detection of low and high treatment frequency subgroups demonstrated an AUC of 0.67 and 0.70, respectively. Total retinal thickness at the fovea, as well as SRF in the central 3mm area at month 2 were found to be important OCT-derived prediction features.
Conclusions :
We proposed and evaluated a methodology to predict low and high anti-VEGF treatment needs from OCT scans taken early during treatment initiation. The results of this pilot study are a promising step toward image-guided prediction of treatment intervals in neovascular AMD therapy.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.