Linking AI-based biomarker analysis to visual acuity changes in central serous chorioretinopathy (CSCR)

Abstract

Purpose : Multiple biomarkers are established routine in the diagnosis and monitoring of central serous chorioretinopathy (CSCR). However, their significance in disease progression and their impact on visual acuity (VA) are little understood. We used AI-based biomarker segmentation and analysis on a large dataset to evaluate the correlation to VA.

Methods : Retrospective analysis of 1685 examinations of 317 patients with CSCR, including healthy partner eyes. With the help of artificial intelligence, 1685 autofluorescence images were annotated for hyperfluorescence (HF) and1368 optical coherence tomography (OCT) exams were annotated for subretinal fluid (SRF) and length of photoreceptors (LPR). The imaging modalities were then automatically registered onto each other. For correlation to the VA, only biomarkers within 250 μm distance to the fovea were analyzed. Groups were compared based on biomarker presence; furthermore, multiple linear regression analysis was performed.

Results : It could be shown that both patients with foveal SRF and elongated photoreceptors had a significantly reduced VA (p < 1E^-12 for SRF, p < 1E-14 for elongated photoreceptors), whereas the presence of HF was not found to deteriorate VA significantly (p=0.13). In the subgroup of patients with present SRF, multiple linear regression was performed. This revealed that elongated photoreceptors were highly significantly correlated with reduced VA (p = 0.001), whereas HF (p = 0.83) and height of SRF (p = 0.56) were no significant covariates.

Conclusions : AI-based biomarker detection and quantification can help to understand the disease. Elongated photoreceptors seem to be the main cause of VA loss, whereas the height of the SRF seems to be less relevant. CSR may be heterogenous. Currently, we investigate the therapeutic and prognostic importance of these biomarkers.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

Representative Example of AI-based segmentation in CSCR
Left: The hyperfluorescence (HF) in autofluorescence is segmented in yellow. The ETDRS grid is also depicted for orientation. The red line shows the location of the OCT scan.
Middle: Segmented retina (red) with subretinal fluid (blue)
Right: Same OCT scan with segmentation of the outer photoreceptors (green)