Prognostic impact and correlation with disease activity of three‐dimensional oct biomarkers in vogt‐koyanagi‐harada patients

Abstract

Listen

Translate article

Aims/Purpose: Vogt‐Koyanagi‐Harada syndrome (VKH) is an autoimmune disease characterized by a chronic bilateral granulomatous panuveitis accompanied by extraocular manifestations, namely of the nervous, auditory, and integumentary systems. Retinal changes documented by optical coherence tomography (OCT) were found to have prognostic value, including detachments between retinal layers with fluid accumulation. The aim of this study was to estimate the total detachment volume (TDV) between retinal layers with fluid accumulation that characterizes the Vogt‐Koyanagi‐Harada syndrome.Methods: An OCT analysis framework was developed in Python which, after a parameter tuning process, automatically segments the 3D detachments between retinal layers and calculates the TDV. The pipeline processes each slice individually following a chain of image processing steps: image enhancement; binarization; size filter; and contour detection. To validate this strategy, this method was applied to one 3D OCT image of a patient with VKH, and the resulting masks were compared with the ground truth obtained after manual segmentation by an expert.Results: Our pipeline has revealed an Intersection of Union (IoU) score of 84%, demonstrating its ability to efficiently segment retinal detachments after a parameter tuning process. As a consequence of such capability, the relative deviation between the experimental and ground truth TDV was 8.3%.Conclusions: This semi‐automated approach offers a promising tool for clinicians to detect retinal changes in VKH patients. By enabling early and precise detection of fluid accumulation and retinal detachments, this method can enhance prognostic evaluations and potentially improve patient management and treatment outcomes. Future work should focus on validating this framework across a larger dataset and refining the processing pipeline to further improve segmentation accuracy and robustness.