Knowledge Bank

Glaucoma, the primary cause of permanent vision loss in the world, is currently diagnosed by first examining patients' visual field, followed by quantification of retinal nerve fiber layer (RNFL) thickness. However, these methods only detect glaucoma when damage is already present. Recent studies have suggested that the changes to the peripapillary sclera structure and biomechanical properties may be an indicator of early-stage glaucoma pathogenesis. As such, better characterization of this structure may enable improved disease diagnosis and staging. Ultrasound at 20 MHz is able to achieve sufficient penetration to image the posterior eye in vivo. Thus, this project first develops a standard protocol for in vivo ultrasound imaging of the posterior eye using a clinical 20 MHz ultrasound system. Then, to demonstrate feasibility of automated segmentation of these in vivo images, deep learning (DL) models are trained on ex vivo data to automatically segment the peripapillary sclera for subsequent biomechanical analyses. As B-mode images collected using linear stages do not all contain the optic nerve head (ONH), volumetric data is first interpolated into radial slices to homogenize the training set and then manually segmented by experienced graders to generate ground truths. This work will facilitate further research to develop predictive models for glaucoma development from biomechanical changes in the peripapillary sclera, potentially enabling improved disease staging and prevention planning for at risk patients.