Lucy Cobbs, MD
Ophthalmology Resident
Wills Eye Hospital
Dr. Ursula M. Schmidt-Erfurth from the Medical University of Vienna presented on the burgeoning role of artificial-intelligence (AI) guided OCT imaging in the management of geographic atrophy (GA) in age-related macular degeneration (AMD).
AI-guided OCT imaging uses 3D-to-2D en face semantic segmentation to perform automated detection of the retinal pigment epithelium (RPE). Photoreceptors are represented by applying an ensemble of u-shaped fully convolutional neural networks to retrieve a 2D B-scan semantic segmentation. This methodology provides a distinct photoreceptor thickness map indicating loss, junctional zone alteration, and photoreceptor integrity.
Feasibility of using AI-guided OCT for managing GA has recently been demonstrated in the phase III clinical trials, OAKS and DERBY, which evaluated the efficacy of intravitreal pegcetacoplan, a C3 inhibitor, in treating GA. Dr. Schmidt-Erfurth described five major applications for AI-guided OCT imaging in management of GA, including early detection, monitoring disease progression, quantification of the therapeutic response, and predicting disease activity and response for each individual patient.
In terms of early detection, the ideal approach to AMD management is to predict and detect conversion to GA before it becomes clinically manifest. Using AI-guided OCT imaging, focal outer nuclear layer and photoreceptor thinning can be identified more than a year earlier than by manual grading of OCT images. Notably, this thinning has been shown to be unrelated to drusen dynamics.
Second, monitoring disease activity and progression of GA lesion growth is highly variable. A fundamental rule of GA growth is that photoreceptor loss exceeds and precedes RPE loss. Accordingly, in a real-world natural history study of 74 eyes, “early converters” to GA started with a high photoreceptor-to-RPE loss ratio measured by AI-guided OCT imaging. Progression of GA was marked by decreasing photoreceptor-to-RPE loss ratios until end-stage GA in which photoreceptor loss equaled RPE loss. AI-guided OCT characterization of photoreceptor and RPE loss may play a crucial role in understanding GA growth dynamics in future clinical trials.
Third, AI-guided OCT has been successfully employed to quantify therapeutic efficacy. In the phase II FILLY trial, pegcetacoplan treatment inhibited RPE loss compared to sham injections. Post hoc AI analysis performed on 246 FILLY study patients using more than 30,000 B-scans of OCT volumes importantly demonstrated that pegcetacoplan had an even greater degree of inhibition on photoreceptor loss.
Fourth, AI analysis has been shown to reliably predict progression of GA. AI-guided OCT analysis was performed on the FILLY study participants for more than 31,000 local GA margin locations and successfully predicted individual progression of GA on baseline OCT’s, depicted as heatmaps. Topographic analysis also captured focal treatment effects such as slower progression toward the fovea. The photoreceptor-to-RPE loss ratio was the most relevant predictor of therapeutic benefit for pegcetacoplan. Eyes with the higher ratios of photoreceptor-to-RPE loss also had the greatest benefit from treatment. AI analysis offers a robust method for differentiating treatment responders from non-responders.
Finally, AI-guided OCT has confirmed and expanded on the outcomes of the phase III trials DERBY and OAKS. AI analysis demonstrated a 70-77% decrease in photoreceptor loss and 20-36% decrease in RPE loss in patients treated with monthly pegcetacoplan injections in DERBY and OAKS at month 12.
Dr. Schmidt-Erfurth concluded that OCT-based AI analysis can reliably measure photoreceptor loss, which can predict early GA conversion, disease activity and therapeutic response. AI-guided OCT imaging may represent a paradigm shift in GA clinical trials and real-world management of AMD.
