Role of Deep Learning–Quantified Hyperreflective Foci for the Prediction of Geographic Atrophy Progression

Abstract

To quantitatively measure hyperreflective foci (HRF) during the progression of geographic atrophy (GA) secondary to age-related macular degeneration (AMD) using deep learning (DL) and investigate the association with local and global growth of GA.METHODS: Eyes with GA were prospectively included. Spectral-domain optical coherence tomography (SDOCT) and fundus autofluorescence images were acquired every 6 months. A 500-mm-wide junctional zone adjacent to the GA border was delineated and HRF were quantified using a validated DL algorithm. HRF concentrations in progressing and nonprogressing areas, as well as correlations between HRF quantifications and global and local GA progression, were assessed.RESULTS: A total of 491 SDOCT volumes from 87 eyes of 54 patients were assessed with a median followup of 28 months. Two-thirds of HRF were localized within a millimeter adjacent to the GA border. HRF concentration was positively correlated with GA progression in unifocal and multifocal GA (all P < .001) and de novo GA development (P [ .037). Local progression speed correlated positively with local increase of HRF (P value range <.001-.004). Global progression speed, however, did not correlate with HRF concentrations (P > .05). Changes in HRF over time did not have an impact on the growth in GA (P > .05).CONCLUSION: Advanced artificial intelligence (AI) methods in high-resolution retinal imaging allows to identify, localize, and quantify biomarkers such as HRF. Increased HRF concentrations in the junctional zone and future macular atrophy may represent progressive migration and loss of retinal pigment epithelium. AIbased biomarker monitoring may pave the way into the era of individualized risk assessment and objective decision-making processes. NOTE: Publication of this article is sponsored by the American Ophthalmological Society.

Artificial Intelligence Algorithms for Analysis of Geographic Atrophy: A Review and Evaluation

“…Three publications used AI to predict future spatial GA progression: Niu et al, Pfau et al, and Schmidt-Erfurth et al 45 – 47 Niu et al and Schmidt-Erfurth et al both combined segmentation with progression modeling. Niu et al 45 utilized their previously published Chan-Vese model and added a random forest with 100 trees to build its prediction model using 19 extracted features.…”

“…Schmidt-Erfurth and colleagues used a residual U-Net for their segmentation and a linear regression for their progression modeling. 47 Results from the segmentation were not available. They found that hyper-reflective foci (HRF) concentration was positively correlated with GA progression in unifocal and multifocal GA (all P < 0.001) and de-novo GA development ( P = 0.037).…”

“…Sensitivity, specificity, positive predictive value, and negative predictive value ranges were all from Niu et al 45 One study by Schmidt-Erfurth et al is not presented, as the results were presented as various correlation P values. 47 c These results represent a total of 4 publications that have assessed CFP, including CFP only and CFP and FAF. …”

“…Schmidt-Erfurth et al utilized a segmentation process, and then characterized GA progression using linear regression. 47 A custom-built algorithm based on the residual U-Net was used for the semantic segmentation of HRF voxels (volumetric pixels) to investigate the growth of GA, using SD-OCT and FAF images. The authors used the Spearman correlation coefficient to investigate the associations between HRF concentrations and GA growth.…”

Optical Coherence Tomography-Angiography in Geographic Atrophy

“…To date, a plethora of progression risk factors have been described in the setting of GA [78]. This includes, amongst other characteristics (e.g., the described vascular alterations), the fundus autofluorescence phenotype [6], baseline lesion size [79], lesion shape [28, 80], and structural OCT characteristics, such as incomplete outer retinal atrophy [81, 82], lesion boundary configuration [83], reticular drusen [84], and hyperreflective foci [85]. Based on clinical experience and preliminary data, these imaging features provide correlated/redundant information.…”