Introduction

Organ-on-a-chip (OoC) is a highly advanced approach combining microfluidics with cell culture, leading to significant advancements in simulating human microphysiological systems and functions [1,2,3]. The diffusion limit of nutrients, such as oxygen, ranges from 100 to 200 μm in the human body. Once the size of an organ or tissue surpasses this range, efficient substance exchange is achieved through the development of microvasculature [4]. Organoids cultured in vitro usually have diameters ranging from 50 to 1000 μm and rely solely on passive diffusion to receive nutrients and oxygen and to remove waste products [5]. When the size of an organoid exceeds the passive diffusion limit in hydrogel (~ 400 μm), its growth and function are significantly affected [6, 7]. Recently, numerous investigators have endeavored to construct microvascular networks in vitro using microfluidic chips, resulting in the vascularization of organoids and enhancing the authenticity of the OoC system. Vascularized organ-on-a-chip (VOoC) employs specialized microfluidic chips with specific structures to co-culture target organoids or spheroids with vascular endothelium and its supporting cell lines, leading to the formation of a perfusable vascular network (Fig. 1). Incorporating additional stimuli, such as flowing mechanical force and growth factor, can significantly improve the success rate of vascular perfusion.

Fig. 1
figure 1

Schematic diagram of characterization pipeline of VOoC. Created with BioRender.com

Full size image

The continuous advancement of VOoC has generated a growing need for advanced imaging techniques that can efficiently and accurately capture the relevant biological processes and the interaction between the organoid and its microenvironment. In this regard, confocal/multiphoton fluorescence microscopy (CFM/MPM) has gained prominence in VOoC imaging due to its exceptional imaging quality [8, 9]. Furthermore, with the progress of imaging technology, emerging microscopic techniques such as light-sheet fluorescence microscopy (LSFM) and fluorescence micro-optical sectioning tomography (fMOST) have generated attention and demonstrated results in achieving comprehensive 3D imaging of organoid structures [10, 11].

The enhanced substance exchange in the deep layer of organoids allows for further increases in their size. Concurrently, the co-culture of multiple cell types related to microvasculature leads to a significant increase in cellular complexity within organoid microenvironment. However, these factors, along with the heterogeneity of the chip material, can introduce challenges, such as light scattering and refractive index (RI) mismatches. Consequently, the CFM imaging depth is often limited to ~ 100 μm or even lower [8]. Given these limitations, obtaining comprehensive information about VOoC becomes challenging, hindering the analysis of the physiological activities within microvasculature and organoids. Recently, tissue optical clearing (TOC) methods, such as CLARITY and CUBIC, have been introduced to VOoC for obtaining the 3D structure of organoids [12,13,14]. Furthermore, the combination of TOC and LSFM enables rapid acquisition of the 3D images [15, 16], significantly improving imaging efficiency. While the efficient acquisition of comprehensive 3D data for the entire VOoC is essential, it is equally critical to process this data quickly, accurately, and automatically. Deep learning techniques have rapidly advanced in recent years [17] and have played a crucial role in cell identification, location tracking [18,19,20], and the analysis of microvascular networks [21, 22], greatly enhancing the automation level of VOoC systems (Fig. 1).

In this review, we will initially summarize the chip designs and vascularization culturing strategies employed in VOoC. Subsequently, we will discuss the imaging methods that apply to VOoC and briefly outline the advancements in TOC techniques within the context of the chip. Furthermore, we will present a comprehensive overview of the vascular analysis methods applicable to VOoC. Finally, we will delve into the future development prospects of VOoC and the ongoing advancements in vascular analysis methods.

Vascularized organoid-on-a-chip

Chip design

Microfluidic chips have emerged as promising platform for in vitro modeling of microvasculature due to several advantages. Firstly, microfluidic chips can construct perfusable microvessels with a diameter of 20–50 μm by controlling interstitial flow within the extracellular matrix (ECM) [23], closely resembling microvasculature’s function and size in the human body. Secondly, microfluidic technology enables the creation of hierarchical vessels with similar complexity to physiological vessels. This capability allows for a more realistic representation of the vascular network. Thirdly, microfluidic chips with relatively simple structural designs can be mass produced through techniques, such as laser cutting or injection molding [24, 25], which is conducive to high-throughput culturing and analysis of vascularized organoid models. Moreover, microfluidic chips offer precise control over various parameters, such as growth factor concentration, shear force, and other environmental conditions. This precise regulation facilitates optimal cell growth and angiogenesis by providing a controlled and tailored microenvironment [26]. One of key points in VOoC design involves incorporating hydrogel channels on the chip to facilitate substance exchange and provide structural support for vascular growth. Here, we categorized the design of VOoC into mainly two types, designs for self-assembly of endothelial cells (ECs) to form microvessels and designs for pre-patterned vessel lumens.

Designs for self-assembly of ECs

The designs for self-assembly of ECs on chip predominantly exploit the intrinsic hydrophobic properties of the chip material, along with the surface tension and capillary effects of the hydrogel liquid. These effects collectively contribute to effectively confining the hydrogel within the channels. Upon curing of the hydrogel under specific conditions, a durable solid–liquid interface is established. This interface facilitates the adhesion and proliferation of ECs, or alternatively, it offers structural support for the self-assembly of vascular ECs into microvascular network within a 3D hydrogel environment. Currently, there are several microstructures that can be employed, including micropillar, microridge, and capillary burst valve.

As depicted in Fig. 2a, in the micropillar design, multiple pillars are distributed at specific intervals along the sides of the hydrogel channels. The stability of the hydrogel channels can be improved by optimizing the hydrophobicity of the chip material and selecting suitable micropillar spacing (Rx) and channel height (Rz). These factors influence the maximum pressure perturbations a hydrogel channel can withstand.

$$\Delta P=\gamma \left(\frac{1}{{R}_{x}}+\frac{1}{{R}_{z}}\right),$$
(1)

where \(\Delta P\) is the pressure differential sustained by surface tension and \(\gamma\) is the surface tension coefficient. A trapezoidal micropillar structure can be utilized to enhance stability and promote a more suitable air–hydrogel interface during hydrogel injection. This structure effectively reduces the perturbation and ensures a more controlled hydrogel filling process [27]. Typically, the chip channel heights range from 100 to 200 μm [28], while lowering the channel height theoretically improves stability, channel heights below 100 μm may result in 3D vascular networks that lack sufficient depth. Micropillar-based OoC is commonly employed in various applications, including the construction of blood–brain barrier (BBB) chips [29], kidney chips [30], and VOoC [14]. However, it is essential to acknowledge that the hydrogel channel formed using micropillar structures can be susceptible to depression and local defects due to the inherent differences in properties between the chip material and the hydrogel, which may hinder the formation of a well-defined and functional vascular barrier.

Fig. 2
figure 2

Chip designs of VOoC. a Micropillars with specific spacing (Rx) and height (Rz) can bind the hydrogel in the central channel to provide the scaffold for microvasculature growth [27]. b Microridges can form a large area of flat hydrogel interface via the meniscus-pinning effect [75], making it suitable for vascular EC adhesion to form the vascular barrier. c Chambers are generally connected with the medium channel by the capillary burst valves to provide nutrients for vascular growth and regulate the interstitial flow [42]. d Microtubes construct the hydrogel lumen utilizing an artificially created vascular mold. Subsequently, a suspension containing vascular ECs is infused into this hydrogel lumen to create microvascular tubes [54]. e Mix ECs with hydrogel, inject into the channel, and place the device with gravity. Add low-viscosity culture medium at the inlet and collect at the outlet. This washes away high-viscosity hydrogel, forming symmetrical three-dimensional hydrogel lumens [51]. f The fugitive support material Pluronic F-127 is 3D printed on the chip’s hydrogel to create vascular molds. After removing Pluronic F-127, ECs are added to the resulting hollow vascular channels. Once the vessels form, organoids can be seeded into the hydrogel for further study [52]. Created with BioRender.com

Full size image

The “microridge,” also known as “phaseguide™,” is a design that utilizes the meniscus-pinning effect, which is generated by a material line or a change in geometry, to form the liquid-hydrogel boundary (Fig. 2b) [31]. Generally, a higher and wider microridge structure contributes to increased hydrogel stability. However, it also leads to a reduced contact area between the hydrogel and the medium channels on both sides. To strike a balance, the height of the geometric change in microridge structures is typically set to be around 1/4 of the channel height, while the width is typically kept similar to the height [32,33,34,35]. The structural characteristics of microridge enable the formation of extensive and flat liquid–hydrogel interface, promoting tight adhesion of ECs for effective barrier functionality [36, 37]. Hence, this structure is highly suitable for constructing barrier-type organoids that demand high hydrogel interface integrity, as well as for observing the local formation of tip/stalk ECs [32, 34, 38]. Simultaneously, the higher surface area-to-volume ratio facilitates substance transport studies [39,40,41]. However, the hydrogel channel formed by microridge is prone to damage and leakage due to significant shear stress caused by fluid flow, particularly during the injection of the culture medium. Therefore, when using microridge chips, it may be advisable to reduce the height of the hydrogel channels, raise the height and width of the microridge, and increase the concentration of relevant proteins in the hydrogel to improve its stability during perfusion [31].

As shown in Fig. 2c, the capillary burst valve design is primarily employed in tissue chamber chips, serving to connect the central chamber with the culture medium channels on both sides [42]. Usually, the size of the capillary burst valve is 50 μm and the chamber is 1 × 2 × 0.1–0.12 mm (length × width × height). Due to surface tension and capillary effects, the hydrogel injected into the central chamber will not leak into the side channels. Upon the gelation of the hydrogel, the culture medium is introduced into the channels on both sides, with the medium channels typically connected to culture medium reservoirs. By adjusting the liquid level in the reservoirs, interstitial flow can be generated in the chip, thereby stimulating cell growth inside the tissue chambers [42,43,44]. The relatively simple structure of the tissue chamber chips makes them compatible with a 96-well plate [45], enabling high-throughput construction of VOoC. Furthermore, using multi-layer microfluidic chips connected by vertical perfusion channels holds significant potential for facilitating co-culture between different organoids [46]. However, it is important to note that the tissue chamber chip does impose certain limitations on the size (smaller than chamber channel’s height, typical 120 μm) and seeding sequence of co-cultured organoids. If researchers intend to seed larger organoids following the formation of microvasculature, alternative chip designs may need to be considered.

The microstructures mentioned above share a common characteristic in the formation of hydrogel channels, which are guided by capillary forces. They are primarily employed in VOoC to separate hydrogel and culture medium channels. One or more channels are used for injecting the hydrogel mixed with cell suspension, while the remaining channels serve for culturing, administering growth factors, or staining reagents. Interstitial flow can be achieved by creating a hydrostatic pressure difference, facilitating the delivery of nutrients and removal of cellular waste products [47].

Designs for pre-patterned lumens

The lumen design constructs microtube structures in hydrogels by artificially designing a series of tubular molds, facilitating the adhesion and growth of ECs to form blood vessels. Microneedle (typical diameter: 150–550 μm) is a commonly used vascular mold for simple structure [48, 49]. More intricate vascular structures can be created using techniques, such as viscous finger patterning (VFP) [50, 51] or 3D bioprinting [52, 53].

Figure 2d illustrates an example for constructing specified 3D vascular structures in vitro using a microneedle [54]. In specific, the process involves placing microneedle on the chip, then introducing the hydrogel into the chip to embed the mold. Once the hydrogel polymerizes, the mold can be removed, establishing a microtube with a controlled lumen size. The ECs suspension is then introduced into the hydrogel lumen and incubated for 15–30 min to promote adhesion. To prevent damage to the hydrogel structure during the molding process, the surface of the vascular mold is typically coated with bovine serum albumin [55]. What’s more, by utilizing various sizes of microneedles and multiple demolding steps, it is possible to achieve the construction of blood vessels with a multi-layered perivascular cell structure, which structurally appears more realistic compared to microvasculature composed of a single layer of ECs. However, this method is constrained by the diameter of the microneedles, making it challenging to construct capillaries that closely resemble real sizes [56]. In vascularization applications, vascularized tumor models can be achieved by introducing tumor organoids into parallel microtube or co-culturing tumor cells or organoids within hydrogel [54]. Moreover, the BBB function can be reconstructed by introducing cerebral vascular ECs and applying fluid flow stimulation [57,58,59]. It is evident that the generation of tens of micrometers level blood vessels using microneedle poses a considerable challenge, and the corresponding demolding process can be quite complex and damaging. Fortunately, VFP and 3D bioprinting show great potential in tackling this issue [52, 60,61,62].

VFP leverages the property of a less viscous fluid generating finger-like protrusions in more viscous fluids, providing an alternative to microneedle molding method (Fig. 2e). The specific methodology involves infusing a high-viscosity fluid (e.g., collagen I) into microchannels and prior to solidification, reinfusing the channels with a low-viscosity fluid (e.g., cell culture medium). At this juncture, the high-viscosity fluid within the channels gradually is replaced by the low-viscosity fluid, yet a layer of high-viscosity fluid persists on the inner walls of the channels, giving rise to finger-like microtubes [63, 64]. Some studies indicate that the perfusion pressure and flow angle of the low-viscosity fluid can influence the thickness and morphology of the resulting microtubes [50, 62]. Specifically, when the chip is vertically infused with low-viscosity fluid, a more uniformly thick lumen wall is established and an increase in pressure during the infusion of low-viscosity fluid results in a larger diameter of the formed lumens [51]. However, the resultant vessels commonly exhibit widths in the order of hundreds of microns, posing challenges in downscaling to the dimensions of typical capillaries. Therefore, Chen et al. utilized VFP to construct larger vessels on the side channels and employed EC self-assembly to create a microvascular network in the central hydrogel channel, achieving a chip with hierarchical microvascular structures [50].

3D bioprinting plays a crucial role in the field of tissue engineering, allowing for the precise placement of cells in 3D space to achieve the desired organ functions, particularly in the vascular construction (Fig. 2f). Currently, there are two main approaches: directly printing cells to form vessels with specific structures [53] and printing vascular mold to allow ECs to attach and grow into vascular tubes [52]. Both these two vascular fabrication approaches are essentially angiogenic remodeling strategies [65]. To achieve the construction of a capillary network, some researchers have adopted cartilage-derived ECM microfibers bioink [66] or techniques, such as laser bioprinting [67,68,69].

Other designs

In recent years, some intriguing designs have significantly enhanced the diversity and functionality of VOoC. For example, Rajasekar et al. developed a microdevice called IFlowPlate based on a 384-well plate, applying it to vascularization in colon organoids and greatly improving the analytical efficiency of VOoC [70]. Additionally, certain studies have utilized leaf venation networks as molds to fabricate VOoC, allowing perfusion with enhanced physiological relevance [71, 72]. Furthermore, these designs often offer dedicated organoid compartments, streamlining the loading and co-culturing processes [24, 73, 74].

Vascularization culturing strategies on the chip

Vascularization has been successfully implemented in various organoids, including the brain, heart, kidney, liver, and intestine [5]. Additionally, a growing body of research focuses on the tumor microenvironment, which aims to establish microvasculature within tumor spheroid to simulate interactions between cells and the ECM [76, 77]. The initial formation of organoid/spheroid is crucial in most vascularization processes and dramatically influences their overall success and researchers often implement structural modifications to the chip to facilitate the cultivation and vascularization of organoids better.

Currently, the xenograft and VOoC models are two primary strategies for generating vascularized organoids. The xenograft model is presently considered the most physiologically relevant method for vascularization. In this approach, organoids are implanted into experimental animals, such as immunodeficiency mice, through the surgical process [78,79,80,81,82]. However, this organoid vascularization strategy necessitates better integration and cost reduction. Additionally, it is subjected to various factors, including animal-individual differences and ethical issues. Microfluidic systems, serving as carriers for VOoC, offer greater control than xenograft  models, providing a valuable in vitro platform for studying the interaction between blood vessels and organoids [61, 83]. Based on the morphogenic processes of microvasculature, the VOoC model is specifically classified into vasculogenesis and angiogenesis.

Organoid/spheroid formation

Stem cells (SCs) can differentiate into both organoids and various supporting cells necessary for vascular generation by exogenous signals regulation (Fig. 3a) [84]. Consequently, regardless of whether they are employed in xenograft or VOoC models, SCs-differentiated organoids exhibit a relatively high successful rate in terms of vasculation. Among various SCs, human-induced pluripotent stem cells are predominantly utilized and have exhibited remarkable potential in establishing stable vascularized organoids, including those of the brain [78], kidney [79],intestine [81], skeletal muscle [85], and other tissues [70, 86].

Fig. 3
figure 3

Vascularization culturing strategies. a SCs can differentiate into organoid and vascular formation supporting cells (such as ECs and pericytes), significantly improving the success rate of vascularization. b Three typical spheroid formation strategies for vascularization. (c-f) VOoC model based on vasculogenesis and angiogenesis. Created with BioRender.com

Full size image

Additionally, researchers commonly employ tumor cells or fibroblasts to establish mono-culture spheroids for studying on-chip vascularization (Fig. 3b) [87]. However, some studies have revealed the challenges associated with constructing a perfusable vascular network within a spheroid consisting solely of a single tumor cell line [88, 89], so this kind of spheroid is primarily employed for investigating tumor migration and invasion in the microvasculature [74, 90, 91]. Fibroblasts spheroids can secrete angiogenesis growth factors, such as basic fibroblast growth factor, thereby inducing the development of a perfusable vascular network and providing a convenient platform to observe substance exchange within spheroids [24, 74].

In contrast to the conventional mono-culture system of spheroids, numerous studies have implemented a proportional ratio of ECs, fibroblasts, and other vascular-supporting cells to enhance the success rate of vascularization and perfusibility. Fibroblasts play a crucial role in promoting microvascular sprouting by secreting related growth factors and providing basic physical cell–cell interactions [5, 92], while ECs contribute to the pre-formation of vascular structures within the spheroids (pre-vascularized) [87]. Typically, cell suspensions of experimental cells and vascular-supporting cells are combined in specific proportions and cultured in a low-adhesion U-shaped well plate. A low volume ratio of Matrigel (e.g., 1% v/v) or Methocel® A4M and various growth factors are added [14, 74, 88, 93] or the cells are cultured in gelled droplets to form spheroids [83]. The specific mixing ratio may vary depending on the cell lines utilized. Due to the higher aggregation ability of fibroblasts compared to other cells, they tend to gather in the center of the co-cultured spheroids [89], which hinders the diffusion of growth factors produced by fibroblasts. To address this, Wan et al. introduced a fibroblast suspension into tumor spheroids and observed that fibroblasts adhered to the surface of the spheroids, resulting in a more significant effect on enhanced vascularization compared to direct co-culture [94]. Additionally, a typical tri-culture approach involves culturing experiment cells, fibroblasts, and ECs in a specific proportion. Nashimoto et al. demonstrated that tri-culture spheroids with an MCF7:HLF:HUVEC ratio of 3:1:1 exhibited the highest success rate in vascularization and perfusibility, effectively reducing the necrosis within the inner core of spheroids [89].

Modification of chip for seeding organoid/spheroid

For the convenience of seeding organoids/spheroids into the chip, researchers typically make modifications to VOoC.

For VOoC based on self-assembly design, there are generally two modification approaches. The first involves creating an external seeding port at the central position of the chip’s hydrogel channel. The diameter of this seeding port typically ranges from 1 to 2 mm (Fig. 3c) [14, 38, 74, 83, 88, 89]. In theory, the larger the organoid/spheroid to be seeded, the larger the seeding port needed. To match the diameter of the seeding port, researchers typically trim the tips of 10-μL or 200-μL pipettes with a surgical knife. During the seeding process, hydrogel and organoid/spheroid should be slowly mixed to avoid the formation of additional bubbles. This method provides better control over the seeding location of organoid/spheroid but carries the risk of increased exposure to the external environment and may, to some extent, affect the fluid distribution within the chip. The second approach involves increasing the channel height of the chip’s hydrogel channel to allow organoid/spheroid to enter the chip through the inlet [91, 94,95,96,97,98]. This method ensures the integrity of the chip to the greatest extent, and the interstitial flow within the chip is not affected. However, the drawback is that the distribution of organoid/spheroid is relatively random, making it less conducive to subsequent imaging observations. Researchers can try to limit the movement of organoid/spheroid by designing micro pit structures inside the chip channels.

For VOoC based on lumen or other designs, it usually has a larger space (diameter: ~ 200 μm) and will be relatively convenient to seed organoid/spheroid into the lumen or directly into the hydrogel, thereby observing the subsequent vascularization effects (Fig. 2d–f) [52, 54, 61, 70, 86, 99].

Vasculogenesis

Vasculogenesis describes the de novo formation of new blood vessels and occurs mainly during the embryonic stage and during wound healing in adults [100, 101]. In microfluidic chips, vasculogenesis often refers to the process where ECs uniformly distributed in the ECM, under external stimuli, self-assemble into tube-like networks to form newly generated vessels [102]. Correspondingly, the vasculogenesis-based vascularization strategy involves co-culturing mature organoids/spheroids directly in hydrogels that contain ECs and supporting cell lines, such as fibroblasts and pericytes. Depending on the experimental designs, organoids/spheroids can be seeded into the chip together with ECs (Fig. 3d) [14, 90, 91, 94]. Alternatively, ECs can be initially seeded in the hydrogel, and the organoids/spheroids are placed on the chip once the ECs have self-assembled into a vascular bed. This approach allows for observing microvasculature invasion into the organoid/spheroid (Fig. 3e) [83]. While the latter method necessitates a longer culturing duration, it offers a temporal window for independent observation and evaluation of the microvascular network. This enables convenient pre- and post-comparison and validation of the effects of the organoids/spheroids on the microvascular network [24, 83, 103].

Angiogenesis

Different from vasculogenesis, angiogenesis characterizes the formation of new blood vessels from pre-existing ones, encompassing sprouting and intussusceptive angiogenesis. In sprouting angiogenesis, new vessels emerge as branches from existing ones, whereas intussusceptive angiogenesis involves the division of pre-existing blood vessels [104]. In the VOoC model based on angiogenesis, ECs are seeded into the medium channels on both sides of the hydrogel, allowing them to adhere to the hydrogel interface and form a “pre-existing blood vessel.” Under the stimulation of mechanical flow force and growth factors, the pre-existing blood vessels on either side grow and sprout, invading the hydrogels to establish microvasculature. Once the vascular bed is formed, the organoids/spheroids can be placed on the central hydrogel channel (Fig. 3f). The microvasculature on both sides gradually invades and vascularizes the organoids/spheroids under the influence of growth factors and flow stimulation [38]. Similar to vasculogenesis, ECs can be seeded into the medium channels simultaneously with the seeding of organoids/spheroids into the central hydrogel channel. Skipping the step of vascular bed formation can reduce the time required for the vascularization of organoids/spheroids significantly [88, 89]. Moreover, the specific seeding sequence can be adjusted based on different cell lines or experimental objectives. For instance, when using organoids and ECs from SCs, the organoids are first seeded into the chip under appropriate differentiation conditions. After organoid differentiation is complete, ECs are further seeded [86, 105].

Key factors for VOoC culturing

Regardless of whether the vascularization approach is based on vasculogenesis or angiogenesis, the processes of vascular growth and vascularization are influenced by several key factors, including interstitial flow, growth factors, the ECM environment, and supporting cell types.

It has been reported that interstitial flow can not only promote vasculogenesis and spheroid vascularization by regulating matrix metalloproteinase-2 [106] but also increase the sprout count and length of angiogenesis by providing proper shear stress to ECs, enhancing the success rate of anastomosis and perfusability in microvascular networks [47]. Generally, interstitial flow can be achieved on the chip by setting a hydrostatic pressure difference, and a range of 2–25 mmH2O pressure difference can typically generate interstitial flow rates of 0–10 μm/s within the ECM [102]. Researchers can calculate the distribution of interstitial flow rates in the chip by combining the channel size and ECM properties (e.g., permeability and porosity) of their VOoC, constructing hydrostatic pressure differences that conform to the physiological flow rate range (1.7–11 μm/s) [24]. Additionally, as shown in Table 1, 3D rocker and syringe pump can also generate controllable interstitial flow in the VOoC.

Vascular endothelial growth factor (VEGF) has played a crucial role in promoting vascular growth. It has been reported that a concentration of VEGF between 20 and 100 ng/mL can effectively promote vascular growth on the chip [102], with a concentration of 50 ng/mL being commonly used [24, 47, 104]. Furthermore, the combined use of various growth factors (such as bFGF, S1P, and PMA) can enhance the proliferative effect to some extent [34]. As shown in Table 1, the mostly used ECM in VOoC is primarily composed of fibrinogen and collagen I, with concentrations ranging from 1 to 10 mg/mL. Wan et al. found a negative correlation between the density and diameter of microvascular network on the chip and the concentration of the fibrinogen solution [107].

Table 1 Summary of designs and applications for VOoC
Full size table

In addition, Ferrari et al. reported that the stiffness of ECM hydrogel controls the size of the patterned vasculature and the density of sprouting angiogenesis [104]. Recently, some studies have demonstrated the spontaneous formation of microvascular networks using synthetically ECM, such as soft gelatin methacrylate and a mixture of calcium alginate and decellularized ECM [98, 108,109,110].

Multicellular interactions should be involved in the complex vascular growth process. As mentioned earlier, fibroblasts are the commonly used supporting cells in VOoC. Fibroblasts can help to prevent early network regression, prolong the lifespan of the vascular network, and enhance open lumens formation [92]. Wan et al. reported that as the proportion of co-cultured fibroblasts with ECs increased in the chip, the diameter and perfusability of the formed microvasculature decreased, but the complexity of overall network increased [107]. Additionally, depending on the specific VOoC being constructed, the corresponding supporting cells may be adjusted. For example, when constructing neurovascular unit (NVU) chip, astrocytes, oligodendrocytes, and neuronal cells may be selected as co-cultured supporting cells [29, 36].

Optical imaging methods applicable to VOoC

To characterize the structural and functional information of the vascularization, advanced optical imaging methods are needed that span different spatial and temporal scales [111]. In the routine culture of VOoC, researchers commonly rely on simple inverted fluorescence microscopy (IFM) to assess the growth status of the microvasculature. However, for more accurate imaging and analysis of 3D settings, other imaging techniques, such as CFM/MPM, fMOST, and LSFM, have become mainstream choices. The quality and accuracy of the acquired image data significantly influence the subsequent data analysis.

CFM/MPM

CFM and MPM techniques are favored in practical applications for their excellent imaging quality and compatibility with VOoC systems (Fig. 4a). CFM not only allows the acquisition of 3D vascular data in VOoC but also enables the tracking of fluorescent particles/cells (Fig. 4b-II) and the assessment of microvascular structure and function through perfusion of FITC-Dextran solutions (Fig. 4b-V) [14, 83]. However, traditional CFM has a limited imaging depth of approximately 100 µm [112]. In contrast, MPM overcomes this limitation by enabling deeper and higher-quality imaging using longer-wavelength excitation lasers, leveraging non-linear optical effects, reducing scattering, and minimizing fluorescence excitation outside the region of interest. Two-photon fluorescence microscopy has been widely employed for imaging deep layers of organoids, including the brain [113], intestine [114], and pancreas [115]. And three-photon microscopy has achieved intact imaging of label-free and non-transparent tissue in cerebral organoids at a depth of approximately 2 mm, providing a valuable option for tracking organoid growth [116].

Fig. 4
figure 4

Optical imaging methods applicable to VOoC. a Schematic representation of CFM for imaging VOoC. b-I Co-culture of HepG2 organoid and vascular network on-a-chip. Reproduced with permission [14]. Copyright 2022, Elsevier. b-II Tumor cells are arrested in two microvasculatures in VOoC. Reproduced under CC-BY-NC-ND license [107]. Copyright 2022, The Authors, published by Wiley–VCH GmbH. b-III Vascularized kidney organoids on-a-chip. Reproduced under exclusive license to Springer Nature America, Inc [136]. Copyright 2019, The Authors, published by Springer Nature. b-IV Immunofluorescence staining for CD31 of VOoC. Reproduced under exclusive license to Springer Nature America, Inc. [33]. Copyright 2022, The Authors, published by Springer Nature. b-V Perfusion testing with 70-kDa FITC-Dextran solution in VOoC. Reproduced with permission [24]. Copyright 2022, Royal Society of Chemistry. c Schematic representation of fMOST applied to cerebral organoids. d Z-projection of cerebral organoids showing GFAP + cells (d-I) and PI-labeled cytoarchitecture (d-II). Reproduced under terms of the CC-BY license [11]. Copyright 2022, The Authors, published by MDPI, Basel, Switzerland. e Schematic representation of open-top LSFM applied to VOoC. f-I Microvasculature images captured by LSFM and its 3D reconstruction. Reproduced with permission [131]. Copyright 2020, Elsevier. f-II 3D digital reconstruction of immunofluorescence on CLARITY-processed bioprinted tissue stained for CD31. Reproduced under terms of the CC-BY-NC-ND license [128]. Copyright 2018, The Authors, published by Elsevier. f-III Tracing the process of microvascular sprouting. Reproduced under the terms of the Optica Open Access Publishing Agreement [122]. Copyright 2022, The Authors, published by Optica Publishing Group. f-IV Quantifying the absolute displacement near the surface of the vascular sprout. Reproduced under terms of the CC-BY license [133]. The Authors, published by Public Library of Science. f-V 3D reconstruction of tumor spheroid. Reproduced under the terms of the Optica Open Access Publishing Agreement [129]. Copyright 2023, The Authors, published by Optica Publishing Group. Created with BioRender.com

Full size image

Classic confocal systems, utilizing point scanning and photomultiplier tube detectors, are relatively time-consuming and can result in photobleaching [117]. Although techniques like MPM or spinning disk confocal microscopy can enhance imaging depth and speed, there is still a demand for further advancements in efficient and accurate 3D imaging acquisition of entire organoids. To address this challenge, a combination of optical sectioning and mechanical slicing can be employed for large samples, such as sequential two-photon tomography and fMOST.

fMOST

fMOST is a technique that combines fluorescence microscopy imaging with a microtome to achieve high-resolution full-size 3D imaging. It is widely used for reconstructing large-scale samples. The process involves embedding the entire sample in resin, followed by fluorescence imaging using a laser scanning microscope to provide high sensitivity. To ensure imaging stability and speed, a non-mechanical acousto-optical deflector is employed as the scanner, with the scan mode set to rapid frequency swap. Additionally, the sample can be immersed in a fluorescent dye solution, facilitating real-time staining (Fig. 4c) [118,119,120]. Ma et al. carried out high-resolution 3D imaging of cerebral organoids using fMOST, achieving 3D visualization of GFAP + signal and continuous spatial distribution of neurons. The authors revealed regional nuclear size and density difference within cerebral organoids (Fig. 4d) [11]. While the majority of VOoC substrates are presently composed of glass, limiting direct application to fMOST, the vascularized organoids can be retrieved from the VOoC for subsequent fixing and embedding and subjected to fMOST for imaging and analysis. Furthermore, the chip’s substrate can be substituted with a flexible material, such as PDMS, to facilitate sectioning [121]. In summary, fMOST provides a powerful imaging platform that offers a new option for visualizing VOoC in future.

LSFM

The LSFM generates a thin light sheet to selectively excite samples at a specific plane, while the emitted fluorescence is detected orthogonally. This enables 3D imaging with higher acquisition efficiency and reduced photobleaching and phototoxicity compared to other imaging techniques [122]. LSFM has been extensively used for 3D imaging of various real tissues and organs [123,124,125], cerebral organoids [126, 127], colon organoids [10], tumor spheroids [128, 129], pancreas organoids [130], etc.

In the context of organoids, LSFM and similar imaging techniques play a crucial role in characterizing the angiogenesis and vasculogenesis processes [122]. For example, Atlas et al. employed digital scanned light-sheet microscopy to observe the formation process of a dense microvascular network in a collagen hydrogel and quantified relevant generation parameters (Fig. 4f-I) [131]. Sebastian et al. used LSFM to visualize a pre-vascularized biological scaffold derived from rat intestines with intact arterio-venous capillary loops, allowing analysis of the distribution of typical endothelial markers CD31 and vWF factor (Fig. 4f-II) [132]. Bissardon et al. utilized LSFM to observe and track the angiogenesis process on a microfluidic chip, providing critical analytical parameters for understanding microvasculature formation (Fig. 4f-III) [122]. Steuwe et al. applied LSFM to quantitatively measure the subcellular level displacement of angiogenic sprouts near the surface (Fig. 4f-IV) [133].

LSFM systems with conventional orthogonal setups require special sample mounting methods to accommodate their orthogonal dual-objective configurations. Therefore, they are not well suited for imaging samples contained within a flat microchip. Additionally, in traditional microfluidic chips using materials like PDMS, the RI of PDMS itself differs significantly from that of the samples in the hydrogel channels, which is unfavorable for light-sheet imaging systems. To address this issue, Zhu et al. developed an open-top optical system and combined it with a novel chip material, Bio 133, allowing better imaging results in tumor spheroids and enabling high-throughput screening of anti-tumor drugs (Fig. 4f-V) [129].

LSFM is a powerful technique for observing the dynamic process of microvasculature growth in VOoC with speed and accuracy. This capability is crucial for studying and understanding the interaction between tissues and blood vessels in vitro [134]. Additionally, when combined with optical clearing (see section below), LSFM provides an essential means to obtain the 3D structure of tissue samples [135].

Tissue optical clearing on the chip

With the advancement of microfluidic and OoC technologies, tissue on chips increasingly resembles the structures and functions of real organs. However, these 3D cultures face several imaging challenges, such as adjusting the sample focal distance, handling tissue refraction variability, and addressing reduced light penetration caused by tissue opacity [121].

In general, light attenuation in biological tissue primarily arises from the absorption and scattering of light by its internal components. For instance, heme, abundant in residual blood within tissues, exhibits high light absorption within the visible light range (400–600 nm) and is a primary contributor to light intensity attenuation in tissues. High scattering is the leading cause of tissue turbidity, mainly caused by the difference in the RI between tissue components. On a macro scale, most biological tissues consist of water with a low RI, along with lipids, proteins, and other substances with a high RI. The primary objective of emerging TOC technology is to reduce tissue absorption and scattering (Fig. 5a) [15]. This can be achieved through dehydration, lipid removal, decalcification, decolorization, RI matching, and other approaches while also aiming to preserve the fluorescence signal and target protein molecules within the tissue to the greatest extent possible [137, 138].

Fig. 5
figure 5

TOC on the chip. a TOC on the chip greatly improves the imaging depth and quality of vascularized organoids by reducing scattering through refractive index matching. b Schematic diagram of the XY- and Z-views of vascularized HepG2 spheroids before and after clearing. Reproduced with permission [14]. Copyright 2022, Elsevier. c Schematic of on-chip clearing, as well as the tumor spheroids before and after clearing. Reproduced with permission [8]. Copyright 2016, National Academy of Science. d The TOC process utilizes mLSI chips and depicts the cleared hACS spheroids in bright-field imaging. Reproduced with permission [13]. Copyright 2017, Royal Society of Chemistry

Full size image

Currently, existing TOC methods, can be categorized into ex vivo clearing methods [139] and in vivo clearing methods [140,141,142]. However, most of these methods involve relatively complex reagent exchange steps, which are labor intensive and can easily result in damage or loss of microtissues. Microfluidic systems, on the other hand, possess excellent fluid control capabilities, enabling efficient and rapid liquid exchange during the clearing process. Therefore, microfluidic systems are advantageous for achieving fully automatic clearing. However, it is worth noting that microfluidic systems also have certain limitations. For instance, the size of the chip limits the number of access ports, making it challenging to achieve complex clearing processes. Additionally, some reagents used in solvent-based clearing methods may have corrosive properties, potentially causing damage to the tubes connecting the microfluidic system and the chip itself.

With the advancement of organoid and OoC technology, obtaining deep tissue information from organoids has become a prominent area of research. Ahn et al. improved the CUBIC method, allowing for observing vascular marker signal CD31 at the depth of the HepG2 tumor spheroid (Fig. 5b) [14]. Chen et al. integrated spheroid formation, CLARITY clearing, and imaging into a spiral-structured chip for quick labeling, clearing, and analysis and showed significantly enhanced imaging depth for spheroids (Fig. 5c) and successful visualization of Langerhans’ vascular structure from islets [8]. Santisteban et al. developed a microfluidic large-scale integration (mLSI) chip platform for high-throughput, integrated CLARITY clearing (Fig. 5d). The extraction time for lipids from adipose-derived stem cell (hASC) spheroids was reduced by utilizing pH change osmotic pumping [13]. Briana et al. optimized the Scale clearing method in combination with vibratome sectioning to visualize thick neural layers in the spinal cord chips [121]. Grist et al. utilized SeeDB, ClearT2, and ScaleSQ with microfluidic chips for high-throughput and fast optical clearing of breast cancer spheroids [9]. Zhang et al. employed a simplified CLARITY for passive tissue clearing and lens-free microscopy for low-cost and efficient imaging [12]. Oh et al. developed a transfer-free 3D micro-tumor culture and visualization system, effectively improving imaging depth and facilitating high-content tumor phenotypic analysis [143]. Further, Yu et al. proposed an on-chip clearing method for imaging live 3D cell cultures, offering a new strategy for dynamically monitoring deep tissue in VOoC [144].

Vascular analysis methods of VOoC

Classic analysis tools

Vascular analysis of VOoC involves quantifying key parameters, such as vascular area, number of branches, and diameter; moreover, it involves tracking the movement trajectory of fluorescent particles to assess vascular perfusibility and connectivity. The standard pipeline for investigating microvascular networks comprises image preprocessing, image binarization, skeletonization of the vascular network, and analysis of characteristic parameters (Fig. 6a). The vascular data in VOoC often deviate significantly from real vessels in terms of morphology and structure. Researchers often employ analysis tools with intuitive user interfaces and real-time feedback on results, such as Fiji, Imaris, and AngioTool, which are widely used for biological image analysis. These tools allow researchers to adjust parameters based on the results to ensure analysis reliability. Moreover, some researchers develop analysis algorithms specifically designed for the characteristics of developing VOoC models, thereby improving the efficiency and accuracy of the analysis for microvasculature. In the following, we present several examples showcasing the application of these classical tools for the analysis of VOoC.

Fig. 6
figure 6

Classic vascular analysis methods of VOoC. a Classic vascular network analysis pipeline. b Fiji (ImageJ) software was employed to investigate vasculature growth and demonstrate the improved connectivity of microvascular networks cultured under flow stimulation. Reproduced with permission [47]. Copyright 2021, Royal Society of Chemistry. c Imaris was utilized for 3D reconstruction of blood vessels ((c-I) Reproduced with permission [151]. Copyright 2021, American Chemical Society and (c-II) Reproduced under the CC-BY license [149]. Copyright 2022, The Authors, published by Elsevier), as well as particle tracking of microbeads or fluorescent particles ((c-III) Reproduced with permission [150]. Copyright 2014, Society for Laboratory Automation and Screening, and (c-IV) Reproduced with permission [14]. Copyright 2022, Elsevier). d Schematic representation of AngioTool analysis, wherein the yellow line denotes the outline of the blood vessel, the red line indicates the centerline of the blood vessel, and the blue point represents the branch point of the blood vessel. Reproduced under the Creative Commons CC0 public domain dedication [152]. Copyright 2011, The Authors, published by Public Library of Science. e Schematic representation of angiogenic sprout tracking on the chip, using disk fitting (j disks) to model the angiogenic process with the disk radius representing the blood vessel radius. Reproduced with permission [154]. Copyright 2013, Elsevier

Full size image

ImageJ/Fiji

Fiji, a distribution of the open-source software ImageJ specifically designed for biological image analysis, offers rapid algorithm design and application through its powerful software library and a broad range of scripting languages [145]. The software’s robust image processing capabilities and plug-in libraries make it convenient for analyzing microvascular networks formed on the chip. For instance, Winkelman et al. utilized ImageJ to project the maximum value of confocal imaging data, binarize the result to analyze the vascular area, and skeletonize the binary image to assess parameters, such as the number of branches, branch length, and diameter. This analysis demonstrated the promotion of angiogenesis under flow stimulation (Fig. 6b) [47]. However, the traditional binarization method used after maximum projection can introduce errors along the Z-axis. To address this, Lee et al. manually divided the image data into upper and lower parts before projection and transformed it into 2D binary mask images to measure vascular morphology (diameter, area, branch) and evaluate differences in vascular growth in diverse culture environments [146]. Additionally, Fiji plug-ins such as Analyze Particles, Analyze Skeleton (2D/3D), and Automated Sprout Analysis significantly enhance the accuracy and efficiency of vascular analysis [147, 148].

Imaris

Imaris is a powerful software tool designed for the 3D reconstruction and analysis of biomedical images. It provides a convenient image preprocessing tool for vascular analysis in scientific research, including deconvolution, signal enhancement, denoising, and more. It also enables the reconstruction of the 3D structure of blood vessels (Fig. 6c-I, II) to analyze spatial relationships between vessels and can be used for the analysis of volume changes in vessels or organoids. With the powerful 3D reconstruction capability, Imaris facilitates intuitive visualization of fluorescent particles within microvascular lumen based on the z-stack images (Fig. 6c-III) [149,150,151]. Ahn et al. used Imaris software to 3D render and visualize the vascularized tumor spheroid structures. The ‘surface’ tool in Imaris was used to reconstruct the blood vessels and the spheroids, while the ‘spot’ tool was employed to reconstruct the flow path of fluorescent beads (Fig. 6c-IV) [14].

AngioTool

AngioTool is a lightweight and user-friendly software developed by Zutaire et al. It is specifically designed to quantitatively analyze vascular networks with repeatability and ease. The software employs a series of algorithms to process the input image. First, the image is convolved with the fast recursive Gaussian kernel and tube-like structures to segment the blood vessel. Then, based on a combination of the eigenvalues of the 2D Hessian matrix, the software computes the vascular structures. AngioTool automatically calculates and outputs various parameters, such as vascular area, length, and density, according to user-defined vascular diameter. It also provides desktop applications, is particularly suitable for entry-level researchers as it displays real-time segmentation and recognition results on the image, allowing users to adjust parameters easily (Fig. 6d) [152]. This software has demonstrated impressive performance in the quantitative analysis of VOoC, including colon organoids [70], co-cultured embryoid bodies—tumor spheroids organoids [153], and NVU organoids [86].

Other methods

Unlike the analysis of microvascular networks, the angiogenesis process is highly dynamic and relatively complex, requiring high-performance algorithms. Common processing steps include image stitching, registration, vascular segmentation, vascular skeletonization, and branch tracking. The Bayesian filtering framework is particularly useful in determining the correlation between ECs in the sprouting process and treating the tip/stalk cell, lumen, and filopodia as separate parts to predict sprout motion (Fig. 6e) [154]. Some algorithms combine convolutional neural network (CNN) and support vector machine (SVM) to improve vessel segmentation and skeletonization [155, 156]. To enhance the algorithm’s robustness, some research groups have explored using the total cell cytosolic integral and the total nuclear integral invading the gel region above the monolayer to quantify the sprouting activity and directed migration. This approach has shown low sensitivity to signal noise, outliers, and diverse shapes and sizes of sprouts [27]. Furthermore, utilizing differential equations or mathematical models to simulate the process of angiogenesis regulated by tumor or growth factors can help strengthen experimental parameter control and provide new insights into the angiogenesis process [157, 158].

In summary, classical vascular analysis methods predominantly utilize software or traditional image processing and analysis algorithms, sometimes in combination. Manual debugging and proofreading can yield relatively accurate results but are limited in achieving fully automatic analysis. Although there have been efforts to incorporate deep learning into vascular analysis, the current focus has primarily been on tasks such as image segmentation and parameter classification. Therefore, a more effective combination of deep learning and VOoC analysis methods will likely be the future development trend of this field.

Deep learning-assisted vascular analysis

With the increasing popularity and development of deep learning, powerful toolkits such as TensorFlow and PyTorch have emerged. Furthermore, in recent years, MATLAB has provided a comprehensive deep learning network toolbox, which has alleviated the environmental set-up difficulties and simplified the usage of deep learning. In general, the application of a deep learning network involves the following key steps: (1) data acquisition, (2) data preprocessing, (3) design requirements, (4) network selection and training model, (5) model evaluation, (6) hyperparameter optimization, and (7) model deploying (Fig. 7). In different scenarios, each step may require adjustment. Notably, the data acquisition phase often requires considerable time and effort. In deep learning tasks, careful attention must be given to data processing, network model selection, and training. During hyperparameter optimization, the choice of an appropriate loss function and adjusting other influential hyperparameters, such as learning rate and batch size, can notably enhance the performance of the model [17, 21].

Fig. 7
figure 7

The pipeline of deep learning for analysis of VOoC. Created with BioRender.com

Full size image

Currently, many studies have successfully integrated deep learning with vascular analysis [159], particularly focusing on the vasculature of real organs [160,161,162]. Diverse approaches were developed, including VesSAP, DeepVess, and TubeMAP, for segmenting and quantifying brain blood vessels in different disease models [125, 163, 164]; Singapore I vessel assessment system for measuring the diameter of human retinal blood vessels [161]; and topology and width aware generative adversarial network for retinal artery/vein classification [165]. Moreover, deep learning networks were used to differentiate vessel invasion and non-invasion in various cancers and identify tumor angiogenesis [166,167,168].

Deep learning relies on a substantial amount of data to achieve optimal performance, and OoC technology represents a cost-effective and high-throughput approach to generating the necessary training samples for deep learning networks. This combination between deep learning and OoC enables automated and intelligent data analysis [21, 169, 170]. The requisite data volume for deep learning on OoC varies based on the specific task and the type of network employed. Common image processing tasks and CNNs typically utilize datasets ranging from hundreds to thousands of images [171,172,173,174]. In the case of OoC generated from precious samples, where obtaining large datasets may be challenging, subsequent enhancements in network performance can be achieved through data augmentation methods (e.g., flipping, rotation, scaling, and noise injection). In recent years, there have been notable advancements in the field of integrating deep learning with OoC [175], with a primary focus on applications, such as recognition, classification [176, 177], target tracking [173, 178], aid imaging [179, 180], drug analysis, and disease diagnosis [172, 181].

Recently, deep learning techniques have been used in the vascular analysis of VOoC. For instance, Urban et al. developed a deep learning model capable of effectively classifying image data before and after drug treatment on the chip, enabling the screening of compounds that can disrupt the vascular network [171]. Irisa et al. developed a deep learning-based visual methodology in conjunction with a vasculature-on-a-chip model to recognize and classify developmental phases of vascular networks [182]. Tronolone et al. employed a random forest model to predict the linear relationship between oxygen transport and microvascular network characteristic parameters in VOoC [183]. For tracking angiogenesis, Wang et al. developed an automated tracking system for blood vessels that can extract quantitative information about the blood vessels and track their formation. They utilized machine learning techniques to enhance the effectiveness of blood vessel skeletonization, making the system achieved a high precision of 97.3% and a recall of 93.9% compared to the ground truth [156]. Subsequently, Wang et al. further optimized the efficiency of EC tracking using CNN and combined it with a Kalman filter to rapidly track and accurately analyze the expansion and migration of ECs during angiogenesis [155].

In summary, deep learning has emerged as a valuable tool in vascular analysis across various fields. However, its application specifically in the context of VOoC is still limited. Influenced by various factors, including cell types, seeding density, and hydrogel strength, microvascular data on VOoC tends to exhibit relative instability. Moreover, the throughput of current VOoC microphysiological models is relatively low, making it challenging to acquire a large volume of data suitable for training purposes. Establishing a versatile deep learning dataset and analysis network applicable to diverse situations poses a challenge. To tackle this issue, a potential solution involves designing a deep learning tool for VOoC analysis that accommodates both training and analysis, which should allow users to integrate new data during usage, facilitating the update of analysis network weights. This approach may enhance the tool’s adaptability to diverse VOoC data, thereby improving the robustness of the network. As VOoC continues to advance, there is significant potential for leveraging deep learning techniques to detect the growth status of microvasculature, diagnose diseases and test drugs (Fig. 7). By combining deep learning with VOoC, researchers can unlock new opportunities for automated analysis, improved disease understanding, and more effective drug development. Further exploration and development of deep learning methods in the context of VOoC hold promise for advancing our understanding of vascular biology and enhancing healthcare outcomes.

Discussion and future perspectives

In this review, we provide an overview of various chip designs, vascularization culturing strategies, and imaging methods utilized in VOoC, as well as classic and deep learning-based vascular analysis methods. The essential information is summarized in Table 1.

The complex vascular environment in vivo is characterized by intricate cell–cell and cell–matrix signaling networks, coupled with diverse flow status (e.g., flow rate, flow direction, and shear stress), and it varies across different organs and vascular structures [184]. For instance, the vascular walls in the arterial system are thicker, enabling them to withstand higher blood pressure and maintain a relatively unidirectional flow, while the distal capillaries typically exhibit slow and multidirectional blood flow, facilitating efficient substance exchange; whereas lymphatic vessels incorporate blind-ended structures that permit only one-way flow, optimizing waste removal and supporting immune functions [185]. However, the reported VOoC encounters some limitations in mimicking human physiological conditions and the environment, for example, the hydrogel channel height of reported VOoC is typically 100–200 μm, the flow pattern is generally simple, and the types of cells used for vascularization are also limited. Therefore, constructing a more realistic vascular network remains challenging, requiring targeted adjustments to the biochemical and mechanical environment based on the types of vascularized organoids. Common approaches involve regulating concentrations and types of growth factors, establishing appropriate flow status, or selecting co-cultured cell types [186]. Additionally, some researchers are addressing this challenge from the perspective of microvascular structure, aiming to build microvascular networks with complex hierarchical structures and lymphatic vessels featuring blind-ended structures [97, 187].

Furthermore, the increased thickness of tissues poses challenges in terms of imaging time and information acquisition in deep tissue. LSFM and TOC techniques have provided valuable assistants in imaging thick tissues. However, LSFM’s performance can be influenced by the chip’s structure and the on-chip TOC is primarily applicable to fixed samples, limiting their applications in real-time dynamic observation of microvascular networks. Therefore, optimizing the design of VOoC to enhance compatibility with imaging systems and realizing TOC for imaging living organoids are directions that warrant further exploration.

In terms of data processing, classic methods that rely on much manual work are predominantly employed for VOoC analysis currently. Consequently, the integration of deep learning is expected to improve vascular analysis efficiency and accuracy in VOoC significantly. Continued advances in VOoC technology and deep learning analysis methods will contribute to a better understanding of microvasculature and its role in human health and disease.

The human body is a complex and dynamic system that relies on the coordinated interactions of multiple cells and tissues. Therefore, it is of great importance to explore suitable cell interactions and develop a multi-cell co-culture OoC system for drug development and disease diagnosis. The incorporation of a vascular network into a microphysiological system on chips to build VOoC can further enhance the authenticity of OoC platforms. However, the structure of the VOoC system is relatively simple due to limitations in micro/nanofabrication technology and cell culture, typically with a standard thickness of 100–200 μm. While a simple monolayer vascular network can be constructed within this thickness, constructing a thicker and more complex vascular layer that closely resembles human physiology is a crucial object in this direction [83]. However, realizing an ideal thick vascular network presents challenges in imaging and data analysis. Researchers should consider how to strike a balance between these aspects, which is a topic worthy of careful consideration.

In conclusion, it is crucial to establish a comprehensive and efficient pipeline encompassing VOoC fabrication, culturing, clearing, imaging, and analysis methods to study microvasculature growth and enable quantitative analysis of related diseases.