Clinical validation for automated geographic atrophy monitoring on OCT under complement inhibitory treatment

Abstract

Geographic atrophy (GA) represents a late stage of age-related macular degeneration, which leads to irreversible vision loss. With the first successful therapeutic approach, namely complement inhibition, huge numbers of patients will have to be monitored regularly. Given these perspectives, a strong need for automated GA segmentation has evolved. The main purpose of this study was the clinical validation of an artificial intelligence (AI)-based algorithm to segment a topographic 2D GA area on a 3D optical coherence tomography (OCT) volume, and to evaluate its potential for AI-based monitoring of GA progression under complement-targeted treatment. 100 GA patients from routine clinical care at the Medical University of Vienna for internal validation and 113 patients from the FILLY phase 2 clinical trial for external validation were included. Mean Dice Similarity Coefficient (DSC) was 0.86 ± 0.12 and 0.91 ± 0.05 for total GA area on the internal and external validation, respectively. Mean DSC for the GA growth area at month 12 on the external test set was 0.46 ± 0.16. Importantly, the automated segmentation by the algorithm corresponded to the outcome of the original FILLY trial measured manually on fundus autofluorescence. The proposed AI approach can reliably segment GA area on OCT with high accuracy. The availability of such tools represents an important step towards AI-based monitoring of GA progression under treatment on OCT for clinical management as well as regulatory trials.

Figure 1

Limits of agreement for inter-grader agreement (A) and model-grader (B,C) agreements of the total geographic atrophy lesion size. The mean difference is plotted in blue and the limits of agreement are plotted in orange (mean difference + 1.96 standard deviation of the difference) and green (mean difference − 1.96 standard deviation of the difference). Sqrt square root transformation. Grader 1 slightly underestimated GA lesion size compared to grader 2. The model performance was closer to grader 2 than to grader 1. The limits of agreement were wider for the inter-grader agreement than for the model-grader agreements, showing that the AI model operated within inter-grader variability.

Figure 2

Examples of en-face segmentation of geographic atrophy on OCT of a small (first column), medium (second column) and large (third column) baseline lesion size, marked in blue. (A) represents the manually annotated baseline area, (B) represents the automatically segmented baseline area, (C) represents the manually segmented growth area at month 12, marked in red and (D) represents the automatically segmented growth area at month 12, marked in red.

Figure 3

Boxplots for manual vs. automated segmentation of geographic atrophy growth on OCT at month 12 for the different treatment groups. Asterisks denote statistically significant difference in growth rates between SM and AM treatment group by automated segmentation (p = 0.030) and manual segmentation (p = 0.028). Note that there was no statistically significant difference between manual and automated segmented growth rates for all treatment groups. SM sham, AEOM every other month, AM monthly treated group, Sqrt square root transformation, GA geographic atrophy.

Figure 4

Correlation scatterplot for manual and predicted GA growth rates at month 12 for the different treatment groups. The correlation coefficient across treatment groups was r = 0.81 with an R² = 0.62. AEOM every other month, AM monthly treated group, SM sham group, GA geographic atrophy, GT ground truth.