Faculty Bibliography

The direct analysis of 3D Optical Coherence Tomography (OCT) volumes enables deep learning models (DL) to learn spatial structural information and discover new bio-markers that are relevant to glaucoma. Downsampling 3D input volumes is the state-of-art solution to accommodate for the limited number of training volumes as well as the available computing resources. However, this limits the network's ability to learn from small retinal structures in OCT volumes. In this paper, our goal is to improve the performance by providing guidance to DL model during training in order to learn from finer ocular structures in 3D OCT volumes. Therefore, we propose an end-to-end attention guided 3D DL model for glaucoma detection and estimating visual function from retinal structures. The model consists of three pathways with the same network architecture but different inputs. One input is the original 3D-OCT cube and the other two are computed during training guided by the 3D gradient class activation heatmaps. Each pathway outputs the class-label and the whole model is trained concurrently to minimize the sum of losses from three pathways. The final output is obtained by fusing the predictions of the three pathways. Also, to explore the robustness and generalizability of the proposed model, we apply the model on a classification task for glaucoma detection as well as a regression task to estimate visual field index (VFI) (a value between 0 and 100). A 5-fold cross-validation with a total of 3782 and 10,370 OCT scans is used to train and evaluate the classification and regression models, respectively. The glaucoma detection model achieved an area under the curve (AUC) of 93.8% compared with 86.8% for a baseline model without the attention-guided component. The model also outperformed six different feature based machine learning approaches that use scanner computed measurements for training. Further, we also assessed the contribution of different retinal layers that are relevant to glaucoma. The VFI estimation model achieved a Pearson correlation and median absolute error of 0.75 and 3.6%, respectively, for a test set of size 3100 cubes.

Purpose/UNASSIGNED:The prevailing theory about the function of lamina cribrosa (LC) connective tissues is that they provide structural support to adjacent neural tissues. Missing connective tissues would compromise this support and therefore are regarded as "LC defects", despite scarce actual evidence of their role. We examined how so-called LC defects alter IOP-related mechanical insult to the LC neural tissues. Methods/UNASSIGNED:We built numerical models incorporating LC microstructure from polarized light microscopy images. To simulate LC defects of varying sizes, individual beams were progressively removed. We then compared intraocular pressure (IOP)-induced neural tissue deformations between models with and without defects. To better understand the consequences of defect development, we also compared neural tissue deformations between models with partial and complete loss of a beam. Results/UNASSIGNED:The maximum stretch of neural tissues decreased non-monotonically with defect size. Maximum stretch in the model with the largest defect decreased by 40% in comparison to the model with no defects. Partial loss of a beam increased the maximum stretch of neural tissues in its adjacent pores by 162%, compared with 63% in the model with complete loss of a beam. Conclusions/UNASSIGNED:Missing LC connective tissues can mitigate IOP-induced neural tissue insult, suggesting that the role of the LC connective tissues is more complex than simply fortifying against IOP. The numerical models further predict that partial loss of a beam is biomechanically considerably worse than complete loss of a beam, perhaps explaining why defects have been reported clinically but partial beams have not.

Collagen fibers organized circumferentially around the canal in the peripapillary sclera are thought to provide biomechanical support to the sensitive tissues within the optic nerve head (ONH). Recent studies have demonstrated the existence of a family of fibers in the innermost sclera organized radially from the scleral canal. Our goal was to determine the role of these radial fibers in the sensitivity of scleral canal biomechanics to acute increases in intraocular pressure (IOP). Following the same general approach of previous parametric sensitivity studies, we created nonlinear generic finite element models of a posterior pole with various combinations of radial and circumferential fibers at an IOP of 0 mmHg. We then simulated the effects of normal and elevated IOP levels (15 and 30 mmHg). We monitored four IOP-induced geometric changes: peripapillary sclera stretch, scleral canal displacement, lamina cribrosa displacement, and scleral canal expansion. In addition, we examined the radial (maximum tension) and through-thickness (maximum compression) strains within the ONH tissues. Our models predicted that: 1) radial fibers reduced the posterior displacement of the lamina, especially at elevated IOP; 2) radial fibers reduced IOP-induced radial strain within the peripapillary sclera and retinal tissue; and 3) a combination of radial and circumferential fibers maintained strains within the ONH at a level similar to those conferred by circumferential fibers alone. In conclusion, radial fibers provide support for the posterior globe, additional to that provided by circumferential fibers. Most importantly, a combination of both fiber families can better protect ONH tissues from excessive IOP-induced deformation than either alone.

Glaucoma is a neurodegenerative disease of the visual system and is the leading cause of irreversible blindness worldwide. To date, its pathophysiological mechanisms remain unclear. This study evaluated the feasibility of advanced diffusion magnetic resonance imaging techniques for examining the microstructural environment of the visual pathway in glaucoma. While conventional diffusion tensor imaging (DTI) showed lower fractional anisotropy and higher directional diffusivities in the optic tracts of glaucoma patients than healthy controls, diffusion kurtosis imaging (DKI) and the extended white matter tract integrity (WMTI) model indicated lower radial kurtosis, higher axial and radial diffusivities in the extra-axonal space, lower axonal water fraction, and lower tortuosity in the same regions in glaucoma patients. These findings suggest glial involvements apart from compromised axonal integrity in glaucoma. In addition, DKI and WMTI but not DTI parameters significantly correlated with clinical ophthalmic measures via optical coherence tomography and visual field perimetry testing. Taken together, DKI and WMTI provided sensitive and comprehensive imaging biomarkers for quantifying glaucomatous damage in the white matter tract across clinical severity complementary to DTI.

The trabecular meshwork (TM) is an ocular tissue that maintains intraocular pressure (IOP) within a physiologic range. Glaucoma patients have reduced TM cellularity and, frequently, elevated IOP. To establish a stem cell-based approach to restoring TM function and normalizing IOP, human adipose-derived stem cells (ADSCs) were induced to differentiate to TM cells in vitro. These ADSC-TM cells displayed a TM cell-like genotypic profile, became phagocytic, and responded to dexamethasone stimulation, characteristic of TM cells. After transplantation into naive mouse eyes, ADSCs and ADSC-TM cells integrated into the TM tissue, expressed TM cell markers, and maintained normal IOP, outflow facility, and extracellular matrix. Cell migration and affinity results indicated that the chemokine pair CXCR4/SDF1 may play an important role in ADSC-TM cell homing. Our study demonstrates the possibility of applying autologous or allogeneic ADSCs and ADSC-TM cells as a potential treatment to restore TM structure and function in glaucoma.

In 2013, the senior author delivered the American Academy of Ophthalmology Robert N. Shaffer Lecture entitled "Glaucoma Changes-Reality Bites." This talk focused on describing the longitudinal structure-function relationships in glaucoma progression. The study was based on a 10-year longitudinal dataset created by calibrated measurements across multiple OCT generations with corresponding visual fields (VFs). The prior held observation was that functional damage follows structural damage. The lecture posited that structure and function change at similar times, but that current measurement technology limits our ability to detect functional abnormalities and change early in glaucoma, as well as to measure structural change late in the disease. The Shaffer lecture provided evidence that structure and function change concordantly and that any apparent discordance in the relationship was due to technologic limitations to measure glaucomatous change. Furthermore, we observed 5 longitudinal relationships of concordance and discordance that can exist with structure-function interactions. Concordance: (1) structure-structure progression, (2) structure-function tipping point, (3) structural floor tipping point. Discordance: (4) functional progression in a "stable" VF with structure-function correlation, (5) functional progression with "normal" structure. In this review article, we will review longitudinal glaucoma progression studies with long-term follow-up and discuss the clinical relevance of relationships of concordance and discordance that can exist with structure-function interactions.