Prescribed Safety Performance Imitation Learning From a Single Expert Dataset.

Existing safe imitation learning (safe IL) methods mainly focus on learning safe policies that are similar to expert ones, but may fail in applications requiring different safety constraints. In this paper, we propose the Lagrangian Generative Adversarial Imitation Learning (LGAIL) algorithm, which can adaptively learn safe policies from a single expert dataset under diverse prescribed safety constraints. To achieve this, we augment GAIL with safety constraints and then relax it as an unconstrained optimization problem by utilizing a Lagrange multiplier. The Lagrange multiplier enables explicit consideration of the safety and is dynamically adjusted to balance the imitation and safety performance during training. Then, we apply a two-stage optimization framework to solve LGAIL: (1) a discriminator is optimized to measure the similarity between the agent-generated data and the expert ones; (2) forward reinforcement learning is employed to improve the similarity while considering safety concerns enabled by a Lagrange multiplier. Furthermore, theoretical analyses on the convergence and safety of LGAIL demonstrate its capability of adaptively learning a safe policy given prescribed safety constraints. At last, extensive experiments in OpenAI Safety Gym conclude the effectiveness of our approach.

Reliability of Vessel Density Measurements in the Peripapillary Retina and Correlation with Retinal Nerve Fiber Layer Thickness in Healthy Subjects Using Optical Coherence Tomography Angiography.

PURPOSE: To evaluate the reliability of vessel density measurements in the peripapillary retina using optical coherence tomography angiography (OCT-A) and to analyze the correlation with retinal nerve fiber layer (RNFL) thickness in healthy subjects. METHOD: Thirty-five healthy volunteers were recruited in the study. The optic disc region was scanned three times with spectral-domain OCT (SD-OCT) and split-spectrum amplitude decorrelation angiography by two skilled examiners. Vessel density of the peripapillary retina was automatically calculated by the software RTVue-XR (version 2015.1.1.98). The RNFL thickness on the optic nerve head was measured by SD-OCT. The coefficient of variation (CV), coefficient of repeatability, and intraclass correlation coefficients (ICC) were calculated for intraobserver repeatability. The Bland-Altman analysis was used to determine interobserver reproducibility. Correlations between peripapillary retinal vessel density and RNFL thickness were analyzed using a multivariable linear regression. RESULTS: The mean age of the volunteers was 47.0 ± 29.7 years. The intraobserver repeatability in different sectors of the peripapillary retina was good with a high coefficient of repeatability, low CV (< 0.2%), and high ICC (0.847-0.952). The interobserver reproducibility was also good in different sectors, but should be interpreted with caution due to the difference bias caused by different observers in some quadrants. There was a significant positive correlation between vessel density and RNFL thickness; optic disc rim area and disc area were negatively related to vessel density (p = 0.008 and p = 0.001, respectively). CONCLUSION: Vessel density measurements showed good repeatability and reproducibility by OCT-A in the peripapillary retina, the vessel density was positively related to RNFL thickness and negatively related to optic disc area and rim area.

Repeatability and Reproducibility of Foveal Avascular Zone Area Measurements Using AngioPlex Spectral Domain Optical Coherence Tomography Angiography in Healthy Subjects.

PURPOSE: The aim of this study was to evaluate the repeatability and reproducibility of foveal avascular zone (FAZ) area measurements using AngioPlex spectral domain optical coherence tomography (OCT) angiography in normal subjects. METHODS: Twenty-two healthy subjects (25 eyes) underwent FAZ area measurements with AngioPlex OCT. Each volunteer was separately examined 3 consecutive times by the 2 experienced observers. The FAZ area was measured using ImageJ software. Intraobserver repeatability was evaluated by calculating the coefficient of variation (CoV) and intraclass correlation coefficient (ICC). Interobserver reproducibility was also assessed using the Bland-Altman test and concordance correlation coefficient (CCC). RESULTS: The FAZ areas were measured as 0.373 ± 0.109 and 0.377 ± 0.112 mm2 by observers 1 and 2, respectively. The repeatability assessment of the FAZ area measurements yielded CoV values of 0.029 and 0.034 and ICC values of 0.997 and 0.996 by observers 1 and 2, respectively. The mean difference between the 2 observers was 0.004. CCC values ranged from 0.9705 to 0.9844. CONCLUSIONS: The FAZ area measurements obtained using AngioPlex OCT showed a good repeatability and reproducibility in healthy subjects. Excellent reliability makes AngioPlex OCT a valid device for measuring the FAZ area.

Antiangiogenic Effects of Doxazosin on Experimental Choroidal Neovascularization in Mice.

PURPOSE: The present study was designed to evaluate the effects of doxazosin on experimental choroidal neovascularization (CNV) in mice. METHODS: Six- to 8-week-old male C57BL/6 mice were divided into a control group and a doxazosin-treated group (5 mg/kg, i.p., daily). Experimental CNV was induced by laser photocoagulation. Seven and 14 days after laser induction, fluorescein angiography, choroidal flat mounts, and histological studies were performed to evaluate the fluorescence leakage, area, and thickness of CNV lesions, respectively. In addition, western blot analysis was carried out to assess the inhibitory effects of doxazosin on the PI3K/Akt/mTOR signaling pathway and the expression levels of hypoxia-inducible factor 1α (HIF-1α) and vascular endothelial growth factor (VEGF), which are involved in CNV model. RESULTS: Compared with the control group, the doxazosin-treated group demonstrated significantly less fluorescence leakage on day 7 and 14 after laser induction. Both the area and the thickness of CNV lesions in the doxazosin-treated group were significantly decreased. Mechanistically, PI3K/Akt/mTOR signaling pathway activation was significantly suppressed in the doxazosin-treated group. The expression of HIF-1α and VEGF was also notably reduced by systemic doxazosin treatment. CONCLUSIONS: Doxazosin exerts antiangiogenic actions in an experimental mouse model of CNV and may be a potential adjunctive therapy for neovascular age-related macular degeneration in humans.