AUTOMATED QUANTIFICATION OF VITREOUS HYPERREFLECTIVE FOCI AND VITREOUS HAZE USING OPTICAL COHERENCE TOMOGRAPHY IN PATIENTS WITH UVEITIS
Hyungwoo Lee, Seungmin Kim, Hyewon Chung, Hyung Chan Kim
Abstract
PURPOSE: Development of an automated method to quantify the count of vitreous hyperreflective foci (vHF) and intensity of vitreous haze in patients with uveitis by optical coherence tomography. METHODS: A method based on deep learning to automatically segment the vHF, vitreous, and retinal pigment epithelium (RPE) in optical coherence tomography was developed using 1,058 scans from 88 optical coherence tomography volumes of 33 patients with intermediate, posterior or panuveitis. Based on segmented images, the vHF count and the relative intensity of vitreous to RPE (VIT/RPE-relative intensity) were quantified. Dice coefficient and intraclass correlation coefficient were calculated between ground truth and the trained network. RESULTS: The segmented area of vHF, vitreous, and RPE by the deep learning-based model showed good agreement with the clinicians' results, yielding a Dice coefficient of 0.69, 0.99, and 0.88, respectively. The intraclass correlation coefficient of the vHF count and the VIT/RPE-relative intensity per scan was 0.99 and 1.00, respectively. In eyes of test set, changes in vHF and VIT/RPE-relative intensity during treatment did not show similar patterns. CONCLUSION: Automated segmentation of the vHF, vitreous, and RPE in optical coherence tomography images of patients with uveitis was accomplished by a deep learning approach. The vHF count and VIT/RPE-relative intensity could be quantified with high reliability.