Background

Development and maintenance of the retinal vasculature is essential for proper ocular function and vision. Disease states where retinal vascularization is abnormal are common causes of blindness [1–3] and include diabetic retinopathy [4–7], retinopathy of prematurity (ROP) [8–12], and inherited retinal vascular diseases exemplified by familial exudative vitreoretinopathy (FEVR) [9,13–29]. In humans, the retinal vasculature is initiated by vascular precursors that lay down the primary vessels that project radially from the optic disc to the retinal periphery by term gestation to form the superficial vascular plexus. From this plexus, starting around 25–26 weeks of gestation, secondary vessels penetrate the retina to form two networks, the deep capillary plexus, on either side of the inner nuclear layer [27–35].

In rodents, as in humans, a primary vascular bed first extends from the optic nerve to the peripheral retina in a single plane along the superficial retina, and the superficial plexus then sends sprouts vertically into the deeper layers of the retina that will form two intraretinal networks. Unlike humans, who exhibit fully formed retinal vessels by term birth, the primary vascular plexus in rodents develops in the first postnatal week, and the formation of the deeper layers is completed by postnatal day 14 (P14). Rodent models offer a unique opportunity to study the evolution of developmental retinal vascular disorders [29,34,36] and, as such, are widely used to study mechanisms of human retinal disease. To do so, retinal vasculature is often determined by flat mounts of retinas followed by staining, visualization, and analysis. However, there is no standard approach that can quantify relevant parameters and capture the phenotypic diversity observed in the various genetic and environmental mouse models, comparing them and assessing responses to candidate treatments. An extensive number of parameters can be individually quantified (Tables 1 and 2). We devised and tested a computational protocol for determining clinically relevant retinal vascular properties ranging from size to network complexity.

Other computational tools have been published in the past, such as Angiogenesis Analyzer, REAVER, Vessel Analysis, AngioQuant, RAVE, SIVA, and AngioTool, among others [37–43]. However, they do not have a wealth of scientific literature regarding their application, they are inflexible regarding which parameters are quantified and how, they are not customizable, they do not have an active developer community, most are not open source, which prevents full transparency regarding the underlying algorithms, they present financial barriers, and they are not available on multiple operating systems. Our protocol thus responds to all of these limitations by using ImageJ, a free and fully open-source software with an active community, vast customizability, compatibility across iOS, Windows, and Linux operating systems, and extensive use in the literature [44].

Global vascular development of the mouse retina depends on the postnatal age and body weight and is measured as percent vascularized area (PVA) of the total retinal area and vessel density (VD). VD is affected by the number, length, and diameter of the vessels. Therefore, we include average branch length (BL), average branch thickness (BT), number of branches (BN), number of junctions (JN, a point where a branch divides), and triple points (TP, junctions where there are two branches emerging from one) in regions of interest (ROIs) (Table 1). Together, these parameters can clarify whether a large VD consists of many small, interconnected capillaries or fewer but larger vessels.

Table 1. Parameters and techniques used to quantify retinal vascular networks and their relevant regions of interest (ROIs)

Parameters such as VD, BL, BT, BN, JN, and TP rely on the assumption that defined vessels are present in a ROI, but this is not always the case. Retinas from a genetic mouse model of the childhood blinding disorder FEVR, the Fzd4-/- mouse model, may manifest large areas of vascular disorganization—termed meshes—with no defined vascular network. To capture these characteristics, each of these areas is outlined and measured, and its ratio to the total vascularized area is reported. We define meshes with a significant lack of defined vasculature as Grade 1 mesh, and areas with an almost complete lack of structure as Grade 2 mesh.

Table 2. Examples of fractal analysis methods

The term complexity can relate to branching, vessel overlap, tortuosity, density, symmetry, space-filling, and the heterogeneity of any of these factors. Quantifying this complexity in our vessel networks can be addressed by applying fractal geometry (Table 2). In combination with the other quantified structural characteristics of this vessel network (e.g., length, thickness, and number), we can glean information on both the development of vessels at the local level and how that impacts the network’s global complexity and vice versa. Each approach complements the other, as they capture information from different perspectives. Fractal analysis has been used to investigate retinal vascular development in many circumstances. The box-counting fractal dimension (D_B) method is the most widely utilized to analyze retinal vasculature complexity [53,60,68], thus being selected for our protocol. Box-counting fractal dimension quantifies how complex and space-filling a branching pattern is by counting how many grid boxes contain vessel segments as the box size decreases. In the context of the retinal vasculature, this measure reflects the geometric complexity and curvature of the vessel network.

Although D_B is a powerful tool, it can be nonspecific, as different patterns with similar D_B can remain qualitatively very different [71]. To mitigate this, lacunarity used in conjunction with fractal dimension characterizes the heterogeneity and symmetry of gaps surrounding an object and together form a more specific impression of the vascular network [71]. Lacunarity measures how evenly or unevenly a pattern fills space, capturing the size and variability of gaps within a structure. Low lacunarity indicates a uniform, tightly packed network with consistent spacing between elements, whereas high lacunarity reflects larger, more irregular gaps and a less cohesive pattern.

In this publication, we report the development of a standardized computational approach that captures and measures structural and pathological features in a mouse retina flat mount in a rapid and reproducible fashion in ImageJ tables, using the software and plugins in Table 3. More specifically, it quantifies the total retina size (μm²), the percent vascularized area (%), and the proportion of area of significant vascular disorganization (meshes) that have no definable network structure (%). Then, in chosen regions of interest at the periphery and midperiphery, we calculate VD (ratio), BL (μm), BT (μm), BN, JN, TP, number of cell clusters (CCs, microaneurysm-like structures), D_B (unitless), and lacunarity (Λ unitless).