Development and external validation of a novel automatic segmentation model for detection and quantification of geographic atrophy from optical coherence tomography imaging

Abstract

Purpose :
Geographic atrophy (GA) is the defining atrophic lesion of advanced non-neovascular age-related macular degeneration (AMD). Detection and segmentation of GA from optical coherence tomography (OCT) imaging is necessary for diagnosis, monitoring, prognosis, and to inform therapy research for this orphan disease. The current standard of segmentation requires specialist manual effort, which is labour intensive and prone to inter-grader variability.
There is a need for validated and fully automated deep-learning approaches to qOCT detection and segmentation of GA that are applicable in clinical care of non-neovascular AMD patients with GA and potential to facilitate therapy research.

Methods :
Deep-learning GA segmentation models were developed for complete GA and its constituent features on 5049 manually segmented optical coherence tomography (OCT) B-scans from 399 eyes (200 patients) with GA secondary to AMD included in the FILLY2 trial (NCT02503332). Predictive performance was validated on an external clinical dataset comprising 884 OCT B-scans from 192 eyes (110 patients) examined at Moorfields Eye Hospital NHS Foundation Trust. The primary outcome was agreement (DSC [dice similarity coefficient] and ICC [intraclass correlation coefficient]) between model GA prediction and consensus of independent graders during external validation.

Results :
Our models accurately segmented GA (median DSC [Dice similarity coefficient] ± SD [standard deviation] 0.96 ± 0.15) and all its constituent features: retinal pigment epithelium (RPE)-loss (0.95 ± 0.21), photoreceptor degeneration (0.96 ± 0.21), hypertransmission (0.97 ± 0.15). Model performance was greater than agreement between specialist human graders (RPE-loss 0.93 ± 0.31, photoreceptor degeneration 0.89 ± 0.20; hypertransmission 0.81 ± 0.30;GA 0.80 ± 0.30). GA segmentation performance remained high in the presence of additional retinal pathologies, including nAMD, PED, and ERM.

Conclusions :
We report development and validation of a scalable deep-learning tool to automatically process OCT scans for detection and quantification of GA, with a predictive performance equivalent to human specialist graders that is robustly retained beyond the sample used for model development.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.