Abstract
Purpose :
(1) To develop algorithms to automatically segment 12 retinal tissue structures in 3D optical coherence tomography (OCT) scans and 7 orbital structures in 3D T1 magnetic resonance images (MRI); (2) To establish trends between the eye (OCT) and orbit (MRI) in a general population and within glaucoma patients.
Methods :
This population study included 12,231 subjects (20,102 eyes) from the UKBiobank cohort (ID 76442). Of these, 223 subjects were diagnosed with glaucoma (384 eyes). First, deep learning algorithms were trained to identify major structures in OCTs and MRIs using 121 B-scans and 30 volumetric scans respectively. Performance was assessed (against manual labels) using the dice coefficient (DC). Second, algorithms to extract various parameters from the segmented scans were developed. MRI parameters included optic nerve tortuosity (NT), globe proptosis (GP), and axial length (AL) (Fig 1). Average retinal nerve fibre layer thickness (RNFL-T) in the superior, inferior, nasal, and temporal quadrants were measured from OCTs (Fig 2). The subject’s ages were also included. Finally, a multiple polynomial regression (MPG) (of order 2, no self-interaction) was performed to identify the relationships between MRI and OCT parameters.
Results :
Segmentation models achieved excellent performance (DC = 0.95) for both MRI and OCT. In the glaucoma population, RNFL-T in all quadrants was significantly associated with NT and with AL (both β>0, p<0.001) and also associated with GP (β<0, p<0.05). Interestingly, in the wider population, reverse trends were found. Interaction terms between AL, GP and NT were also significant predictors of RNFL-T, with comparable coefficient directions in the two populations. In particular, RNFL-T had positive association with AL×GP (β>0, p<0.001) in both the glaucoma and wider populations.
Conclusions :
The study revealed critical links between retinal and orbital structures. Specifically, RNFL thinning was associated with both low tortuosity and with high proptosis in glaucoma patients. In the glaucomatous eye, RNFL thinning could be due to less tortuous nerves exerting greater torsional forces during eye movements. Similarly, increased GP might lead to greater biomechanical stress on the nerve, which could contribute to glaucomatous damage and hence RNFL thinning. These findings underscore the potential importance of orbital structures, interactions between orbit parameters, and possibly orbit biomechanics in glaucoma.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.