Search

Optical coherence tomography angiography (OCTA) is an emerging non-invasive technique for imaging the retinal vasculature. While there are many promising clinical applications for OCTA, determination of image quality remains a challenge. We developed a deep learning-based system using a ResNet152 neural network classifier, pretrained using ImageNet, to classify images of the superficial capillary plexus in 347 scans from 134 patients. Images were also manually graded by two independent graders as a ground truth for the supervised learning models. Because requirements for image quality may vary depending on the clinical or research setting, two models were trained—one to identify high-quality images and one to identify low-quality images. Our neural network models demonstrated outstanding area under the curve (AUC) metrics for both low quality image identification (AUC = 0.99, 95%CI 0.98–1.00, κ = 0.90) and high quality image identification (AUC = 0.97, 95%CI 0.96–0.99, κ = 0.81), significantly outperforming machine-reported signal strength (AUC = 0.82, 95%CI 0.77–0.86, κ = 0.52 and AUC = 0.78, 95%CI 0.73–0.83, κ = 0.27 respectively). Our study demonstrates that techniques from machine learning may be used to develop flexible and robust methods for quality control of OCTA images. [ABSTRACT FROM AUTHOR]

When subjected to loads, granular materials tend to yield and exhibit some localized particle reorganizations. Due to the complex disordered structure of granular materials, it is challenging to identify the key preexisting defects in the static, unloaded structure that eventually promotes dynamical particle rearrangements once a load is applied. Here, based on discrete element simulations of an archetypal frictional granular material model, we introduce a machine learning framework that pinpoints such structural defects with unprecedented accuracy. We show that the optimal structural fingerprint of plastic flow defects depends on strain, wherein the plastic flow is governed by short-range defects at low strain but become dominated by medium-range defects at high strain. [ABSTRACT FROM AUTHOR]

Upon sustained loading, colloidal gels tend to feature delayed viscoplastic creep deformations. However, the relationship, if any, between the structure and creep dynamics of gels remains elusive. Here, based on accelerated molecular dynamics simulations and the recently developed softness approach (i.e., classification-based machine learning), we reveal that the propensity of a gel to exhibit long-time creep is encoded in its static, unloaded structure. By taking the example of a calcium–silicate–hydrate gel (the binding phase of concrete), we extract a local, non-intuitive structural descriptor (a revised version of the "softness" metric proposed by the pioneering work from Cubuk et al.) that is strongly correlated with the dynamics of the particles. Notably, the macroscopic creep rate exhibits an exponential dependence on the average softness. We find that creep results in a decrease in softness in the gel structure, which, in turn, explains the gradual decay of the creep rate over time. Finally, we demonstrate that the softness metric is strongly correlated with the average energy barrier that is accessible to the particles. [Display omitted] [ABSTRACT FROM AUTHOR]