Bedggood Phillip, Ding Yifu, Dierickx David, Dubra Alfredo, Metha Andrew
Department of Optometry and Vision Sciences, University of Melbourne, 3010, Australia.
Department of Ophthalmology, Stanford University, Palo Alto, CA 94303, USA.
Biomed Opt Express. 2025 Jan 7;16(2):473-498. doi: 10.1364/BOE.547734. eCollection 2025 Feb 1.
Cells and other microscopic phase objects can be visualized in the living retina, non-invasively, using non-confocal light detection schemes in adaptive optics scanning light ophthalmoscopes (AOSLOs). There is not yet widespread agreement regarding the origin of image contrast, nor the best way to render multichannel images. Here, we present data to support the interpretation that variations in the intensity of non-confocal images approximate a direct linear mapping of the prismatic deflection of the scanned beam. We advance a simple geometric framework in which local 2D image gradients are used to estimate the spherocylindrical refractive power for each element of the tissue. This framework combines all available information from the non-confocal image channels simultaneously, reducing noise and directional bias. We show that image derivatives can be computed with a scalable, separable gradient operator that minimizes directional errors; this further mitigates noise and directional bias as compared with previous filtering approaches. Strategies to render the output of split-detector gradient operations have been recently described for the visualization of immune cells, blood flow, and photoreceptors; our framework encompasses these methods as rendering astigmatic refractive power. In addition to astigmatic power, we advocate the use of the mean spherical equivalent power, which appears to minimize artifacts even for highly directional micro-structures such as immune cell processes. We highlight examples of positive, negative, and astigmatic power that match expectations according to the known refractive indices and geometries of the relevant structures (for example, a blood vessel filled with plasma acts as a negatively powered cylindrical lens). The examples highlight the benefits of the proposed scheme for the visualization of diverse phase objects including rod and cone inner segments, immune cells near the inner limiting membrane, flowing blood cells, the intravascular cell-free layer, and anatomical details of the vessel wall.
使用自适应光学扫描激光检眼镜（AOSLO）中的非共焦光检测方案，可以在不进行侵入性操作的情况下，在活体视网膜中可视化细胞和其他微观相位物体。关于图像对比度的来源以及呈现多通道图像的最佳方法，目前尚未达成广泛共识。在此，我们提供数据以支持这样一种解释，即非共焦图像强度的变化近似于扫描光束棱镜偏转的直接线性映射。我们提出了一个简单的几何框架，其中局部二维图像梯度用于估计组织每个元素的球柱面屈光力。该框架同时结合了来自非共焦图像通道的所有可用信息，减少了噪声和方向偏差。我们表明，可以使用可扩展的、可分离的梯度算子来计算图像导数，该算子可将方向误差降至最低；与先前的滤波方法相比，这进一步减轻了噪声和方向偏差。最近已经描述了用于免疫细胞、血流和光感受器可视化的分裂探测器梯度运算输出的渲染策略；我们的框架将这些方法作为像散屈光力的渲染方法包含在内。除了像散屈光力之外，我们提倡使用平均球镜等效屈光力，即使对于免疫细胞突起等高度定向的微观结构，它似乎也能将伪像降至最低。我们突出显示了正屈光力、负屈光力和像散屈光力的示例，这些示例与根据相关结构的已知折射率和几何形状所预期的结果相符（例如充满血浆的血管充当负屈光力的柱面透镜）。这些示例突出了所提出方案在可视化各种相位物体方面的优势，这些物体包括视杆和视锥细胞内节、内界膜附近的免疫细胞、流动的血细胞、血管内无细胞层以及血管壁的解剖细节。
