Visualization of erythrocyte stasis in the living human eye in health and disease.

Blood cells trapped in stasis have been reported within the microcirculation, but their relevance to health and disease has not been established. In this study, we introduce an in vivo imaging approach that reveals the presence of a previously-unknown pool of erythrocytes in stasis, located within capillary segments of the CNS, and present in 100% of subjects imaged. These results provide a key insight that blood cells pause as they travel through the choroidal microvasculature, a vascular structure that boasts the highest blood flow of any tissue in the body. Demonstration of clinical utility using deep learning reveals that erythrocyte stasis is altered in glaucoma, indicating the possibility of more widespread changes in choroidal microvascular than previously realized. The ability to monitor the choroidal microvasculature at the single cell level may lead to novel strategies for tracking microvascular health in glaucoma, age-related macular degeneration, and other neurodegenerative diseases.

A Pilot Study Assessing Retinal Blood Flow Dysregulation in Glaucoma Using Erythrocyte Mediated Velocimetry.

Purpose: The purpose of this study was to compare autoregulation of retinal arteriolar and venular blood flow in patients with glaucoma, glaucoma suspect participants, and control participants using erythrocyte mediated velocimetry. Methods: This prospective cohort pilot study included 7 eyes of 5 participants with glaucoma, 15 eyes of 8 glaucoma suspect participants, and 11 eyes of 6 control participants. Mean erythrocyte velocity in retinal arterioles and venules was measured using erythrocyte mediated velocimetry at room air and after oxygen supplementation. Change in erythrocyte velocity was compared among all groups using generalized estimating equations. Results: In total, 64 vessels (18 with glaucoma, 31 that were glaucoma suspect, and 15 controls) of 33 eyes of 19 participants were analyzed. There was no significant difference in baseline velocities in arterioles or venules among the three groups. With induction of hyperoxia, mean arterial erythrocyte velocity decreased in glaucoma (-7.2 ± 13.7%), which differed from controls and glaucoma suspects where erythrocyte velocity increased with hyperoxia by 4.6 ± 13.3% (P = 0.002) and 7.2 ± 21.7% (P = 0.03), respectively. A higher baseline arteriolar velocity (ß = -3.9% per mm/s, P = 0.002), glaucoma diagnosis (ß = -21.1%, P = 0.03), and White race (ß = -20.0%, P = 0.01) were associated with decreased velocity in response to arterial hyperoxia. Conclusions: Hyperoxia increased erythrocyte velocity in control and glaucoma suspect participants, but decreased erythrocyte velocity in glaucoma participants, possibly due to impaired autoregulation. Baseline velocity, glaucoma diagnosis, and White race were associated with a decrease in velocity with induction of hyperoxia. Translational Relevance: The European Medicines Agency (EMA) permits precision measurements of blood flow which may aid in the development of biomarkers of glaucoma-related dysregulation of blood flow.

Hard Attention Net for Automatic Retinal Vessel Segmentation.

Automated retinal vessel segmentation is among the most significant application and research topics in ophthalmologic image analysis. Deep learning based retinal vessel segmentation models have attracted much attention in the recent years. However, current deep network designs tend to predominantly focus on vessels which are easy to segment, while overlooking vessels which are more difficult to segment, such as thin vessels or those with uncertain boundaries. To address this critical gap, we propose a new end-to-end deep learning architecture for retinal vessel segmentation: hard attention net (HAnet). Our design is composed of three decoder networks: the first of which dynamically locates which image regions are "hard" or "easy" to analyze, while the other two aim to segment retinal vessels in these "hard" and "easy" regions independently. We introduce attention mechanisms in the network to reinforce focus on image features in the "hard" regions. Finally, a final vessel segmentation map is generated by fusing all decoder outputs. To quantify the network's performance, we evaluate our model on four public fundus photography datasets (DRIVE, STARE, CHASE_DB1, HRF), two recent published color scanning laser ophthalmoscopy image datasets (IOSTAR, RC-SLO), and a self-collected indocyanine green angiography dataset. Compared to existing state-of-the-art models, the proposed architecture achieves better/comparable performances in segmentation accuracy, area under the receiver operating characteristic curve (AUC), and f1-score. To further gauge the ability to generalize our model, cross-dataset and cross-modality evaluations are conducted, and demonstrate promising extendibility of our proposed network architecture.