Abstract
Introduction
The chicken chorioallantoic membrane (CAM) is a widely used in vivo model for studying angiogenesis and tumor growth in accordance with the 3R principles. However, its multilayered vascular organization complicates quantitative analysis because overlapping superficial and deeper vascular compartments are difficult to distinguish in conventional two-dimensional imaging. Although artificial intelligence (AI)-based methods improve vascular quantification, many existing workflows require extensive preprocessing or user-side model training.
Methods
We established an AI-assisted protocol that restricts vascular quantification to the superficial CAM capillary plexus, the vascular compartment most responsive to angiogenic and anti-angiogenic stimuli. Functional separation of this layer was achieved by intra-CAM injection of commercially available bovine whipped cream (minimum 30% fat), creating a diffuse white background that optically masks deeper vessels without altering vascular morphology. A U-Net-based model trained on manually annotated vessel masks was used for automated two-dimensional quantification of vessel area, length, branching points, and thickness.
Results
The segmentation model achieved a Dice similarity coefficient of 0.831. Application at embryonic days 11 and 15 revealed remodeling of the superficial CAM vasculature, including increased branching and changes in perfusion area. The workflow reduced preprocessing complexity and enabled standardized analysis using a pretrained segmentation model.
Conclusion
This method combines experimental layer isolation with automated vessel segmentation to enable two-dimensional quantification of superficial CAM vasculature without requiring custom model training. It supports practical and reproducible CAM angiogenesis studies by providing a standardized analysis workflow.
Keywords: Chorioallantoic membrane, Artificial intelligence, Vascular quantification, Chick embryos
Introduction
The chorioallantoic membrane (CAM) of the chicken embryo is a widely used in vivo model in cancer and vascular biology research due to its dense vascular network, rapid development, and experimental accessibility [1]. The CAM enables investigation of tumor growth, angiogenesis, extracellular matrix remodeling, metastatic dissemination, and responses to anticancer therapies, while allowing interactions between tumor, stromal, and immune components to be studied in a physiologically relevant environment [2–4]. In addition, the CAM assay aligns with the 3R principles of animal research – Replacement, Reduction, and Refinement – providing an ethically favorable, nonmammalian platform for preclinical studies [5–7].
Tumor-induced angiogenesis in the CAM reproduces key features of neovascularization observed in human cancers, making it a valuable model for investigating vascular remodeling and evaluating anti-angiogenic strategies [8]. Owing to the transparency of the CAM, vascular parameters such as vessel density, branching architecture, and morphological complexity can be quantified using optical imaging approaches [9].
Before the widespread application of AI-based image analysis, CAM angiogenesis was assessed using conventional approaches that combine imaging techniques with manual or semi-automated quantification. These include manual vessel counting, threshold-based analysis (e.g., ImageJ), and imaging modalities such as fluorescence microscopy, Doppler ultrasound, and laser speckle contrast imaging [8, 10]. Fluorescence and intravital imaging allow visualization of vascular architecture, while Doppler ultrasound and laser speckle contrast imaging provide functional information on blood flow and microvascular perfusion [11–15]. Despite these advances in imaging, quantitative analysis often still relies on manual or semi-automated methods, which are time-consuming, subject to observer bias, and difficult to standardize, thereby limiting scalability in routine CAM angiogenesis studies.
To address these limitations, artificial intelligence (AI)-based image analysis approaches have been introduced to automate and standardize CAM vascular quantification [14, 16–21]. Deep learning-based frameworks and automated platforms enable objective extraction of multiple vascular features and have been applied to detect tumor-induced and treatment-associated vascular changes. Despite their potential, many existing AI-based methods require specialized computational expertise, extensive model training, or complex imaging pipelines, which restrict their routine use by the broader CAM research community.
A further, often underappreciated challenge arises from the intrinsically multilayered organization of the CAM vasculature. Histological and anatomical studies show that the CAM consists of a superficial capillary plexus located directly beneath the ectoderm and a deeper hierarchy of medium- and large-caliber vessels embedded within the mesoderm [22, 23]. These vascular compartments form a three-dimensional architecture rather than a single planar network. The superficial capillary plexus undergoes continuous remodeling and is the primary site of angiogenic sprouting, whereas deeper vessels function mainly as feeding and draining conduits and exhibit comparatively limited short-term structural plasticity [22, 24]. Quantitative analyses that do not distinguish between these compartments, particularly when based on two-dimensional imaging, risk conflating dynamic angiogenic changes with relatively stable vascular structures.
Based on this anatomical and functional organization, we hypothesized that restricting vascular analysis to the superficial CAM capillary plexus would provide a biologically appropriate framework for studying angiogenic remodeling while reducing analytical complexity. Accordingly, we developed an AI-assisted protocol for selective two-dimensional quantification of superficial CAM vasculature.
Materials and Methods
Embryos
Fertilized chicken eggs (Gallus gallus domesticus) were obtained from the Campus Frankenforst of the Faculty of Agricultural, Nutritional and Engineering Sciences of the University of Bonn, and incubated at 37.8°C with 50–60% humidity in a standard incubator, as described previously [25]. All experiments were conducted in accordance with institutional guidelines and national regulations for the care and use of animals in research. Embryos were staged according to the Hamburger and Hamilton criteria to ensure experimental consistency [26].
Preparation of CAM
On embryonic day 3 (ED3), fertilized eggs were carefully windowed on the lateral side of the shell under sterile conditions. Following eggshell fenestration, the eggs were sealed and returned to the incubator to allow continued embryonic development. CAM assays were subsequently performed on ED11 and ED15 to assess the developmental changes in the superficial vascular network and to compare vascular remodeling between an earlier and a later stage of CAM development.
Isolation of Superficial CAM Vasculature
At the time of analysis, the CAM was exposed through the preformed shell window, which was carefully enlarged to approximately 1.5 cm × 2 cm to allow optimal visualization and imaging of the vascular network. To selectively isolate the superficial blood vessels while minimizing interference from deeper vascular layers, 2–4 mL of commercially available bovine whipped cream (minimum 30% fat, ultra-high temperature treated) was gently injected into the mesodermal stromal layer beneath the superficial CAM layer using a fine-tipped syringe. The anatomical principle of the intra-CAM injection is shown in Figure 1. The product was purchased from a local supermarket and used without further dilution, sterilization, or preparation at room temperature.
Fig. 1.
Schematic illustration of intra-CAM injection. Bovine whipped cream (30% fat) is injected into the mesodermal stromal layer beneath the superficial vascular plexus to achieve optical masking of deeper vessels. The schematic is adapted from the graphical abstract by Dünker and Jendrossek [27].
Bovine whipped cream was selected as the intra-CAM masking agent because its lipid-rich emulsion provides strong diffuse optical scattering and high opacity. Suspended fat globules and casein micelles generate significant light scattering due to refractive index differences between these particles and the surrounding aqueous phase, producing the characteristic white, nontransparent appearance. This scattering effect increases optical separation between the superficial blood vessels and the deeper vascular layers by masking underlying structures without directly altering the morphology of the superficial vasculature. Similar optical masking principles are widely used in biomedical imaging with lipid emulsions such as Intralipid, which serve as tissue-mimicking scattering media [28, 29]. Compared with intralipid suspensions or opaque gels, whipped cream is inexpensive, sterile, readily available, easy to inject, and does not require additional preparation or specialized reagents. Importantly, the cream remains localized beneath the superficial layer and does not interfere with standard stereomicroscopic imaging or automated vessel segmentation.
In preliminary experiments, fresh bovine milk with 3.8% fat content was tested as a background masking agent; however, its lower lipid content resulted in insufficient optical opacity and weaker masking of deeper vascular layers. The higher fat content of whipped cream provided stronger diffuse light scattering, producing a more uniform and stable white background beneath the CAM surface. Therefore, whipped cream with a minimum fat content of 30% was selected for all subsequent experiments.
In our observations, the bovine whipped cream injection did not induce visible distortion of vessel morphology or acute disruption of vascular architecture, and the diffuse white background stabilized within less than 1 min after injection. This white background optically masked the underlying vascular structures without mechanically disturbing or altering the morphology of the superficial blood vessels. As a result, only vessels located at the CAM surface were visualized and subjected to quantitative analysis (shown in Fig. 2). In contrast, images acquired without cream injection showed overlapping vascular structures from multiple depth planes, which could complicate two-dimensional image analysis and could confound quantitative and morphological assessment of vascular development.
Fig. 2.
Isolation of superficial CAM vasculature by optical masking of deeper vascular layers. Representative images of the same CAM region of a day-11 embryo acquired sequentially before (a) and after (b) injection of bovine whipped cream beneath the superficial CAM layer (×70 magnification). a In the absence of cream, blood vessels from multiple depth planes are simultaneously visible, resulting in overlapping vascular structures. b Injection of bovine whipped cream creates a uniform diffuse white background that optically masks deeper vascular layers, enabling selective visualization of the superficial blood vessels. This functional separation allows consistent two-dimensional analysis of surface vessels while excluding contributions from deeper, more mature vasculature.
In this study, “functional separation” refers to selective visualization of the superficial vascular plexus for quantitative analysis without physical disruption of the deeper CAM structure. This is achieved by “optical masking,” in which the injected lipid-rich white background reduces visualization of deeper vessels by diffuse light scattering.
Imaging
After layer isolation, the CAM surface was imaged using a stereomicroscope (Nikon SMZ1500, 1.0x objective lens, zoom setting of 2.0x) equipped with a digital camera (Nikon DS-Fi3). The imaging system was calibrated using a stage micrometer, yielding a spatial resolution of 2.23 µm per pixel. Images were acquired with consistent illumination settings across samples to ensure reproducibility.
For each embryo, only one region was imaged in this study. Image acquisition parameters, including exposure time, resolution, and contrast, were kept constant for all experimental groups to enable direct comparison across developmental stages and treatment conditions.
Training of CAM Segmentation Model
A U-Net-based convolutional neural network with a MiT-B1 encoder backbone derived from the SegFormer architecture was used for automated vessel segmentation [30, 31]. The model was trained using manually annotated ground-truth vessel masks distinguishing vessels from the background.
For each embryo, a region of interest (ROI) of 1,024 × 1,024 pixels was first extracted from the CAM image and used as the unit of analysis. A subset of these ROIs was randomly selected from each developmental stage (ED11, n = 5; ED15, n = 5) and manually annotated by trained medical personnel to generate ground-truth vessel masks. These annotated ROIs were used for model training. Given the limited number of manually annotated ROIs, reflecting the labor-intensive nature of high-resolution microvascular annotation, transfer learning was applied. The encoder was initialized with weights pretrained on a publicly available retinal fundus vessel segmentation dataset [24] and subsequently fine-tuned on the CAM images.
During model training, images were preprocessed using contrast-limited adaptive histogram equalization and gamma correction, followed by intensity normalization. To increase the number of training samples, annotated ROIs were randomly cropped into multiple patches of 512 × 512 pixels. Data augmentation included random horizontal and vertical flipping.
The dataset was divided at the ROI level into training, validation, and test subsets (80%, 10%, and 10%, respectively). Model training was performed using a combined Dice and Focal loss function, and model selection was based on the highest validation Dice similarity coefficient. Segmentation performance on the held-out test subset yielded a Dice similarity coefficient of 0.831, indicating good agreement between automated segmentation and manual annotations (shown in Fig. 3).
Fig. 3.
Comparison of ground-truth and model-predicted vessel masks. A region of interest (ROI; 1,024 × 1,024 pixels) was extracted from a representative image of superficial CAM vasculature of an ED11 embryo and used as the unit of analysis. a Original stereomicroscopic image of the CAM surface. b Manually annotated ground-truth vessel mask. c Model-generated vessel segmentation obtained using the trained segmentation model. d Overlay of the ground-truth vessel mask (red) and model prediction (green), showing spatial agreement between ground-truth and automated segmentation. Minor discrepancies were primarily observed in very fine capillaries and at vessel edges.
For inference, images were processed without augmentation. A single ROI of 1,024 × 1,024 pixels per embryo was analyzed using the final trained model to generate vessel segmentation masks for quantitative analysis.
Quantification of Vasculature
The previously defined ROI (1,024 × 1,024 pixels) was used for quantitative analysis. To avoid bias, quantitative measurements were performed exclusively on ROIs obtained from an independent cohort of embryos that was not used during model training or segmentation performance evaluation. Vessel segmentation was performed using the trained CAM segmentation model to generate binary vessel masks. The training code for the CAM segmentation model is publicly available at https://github.com/yihu111/cam-vessel-segmentation. The resulting masks were subjected to automated quantitative analysis to extract vascular morphometric parameters (shown in Fig. 4). All images were analyzed using identical parameter settings without manual correction.
Fig. 4.
Segmentation and quantitative analysis of superficial CAM vasculature. A region of interest (ROI; 1,024 × 1,024 pixels) was extracted from a representative CAM image of an ED11 embryo. a Original image of the superficial CAM vasculature following optical masking of deeper layers. b Corresponding segmentation and quantitative analysis output. The segmented vessel mask is shown as a cyan overlay. Vessel centerlines, obtained by skeletonization, are displayed in yellow, and branching points are indicated by red markers. This representation serves as the basis for automated extraction of vascular parameters.
Total vessel area was calculated as the percentage of image pixels classified as vessel. Total vessel length was quantified by skeletonizing the binary vessel mask and counting skeleton pixels. Branching points, defined as junctions where three or more vessel segments converged, were identified from the skeletonized network. Mean vessel thickness was estimated using distance-transform analysis; for each skeleton pixel, the distance to the nearest vessel boundary was calculated, multiplied by 2 to obtain a local diameter estimate, and averaged across all skeleton pixels.
Statistical Analysis
Quantitative data are presented as mean ± standard deviation (SD). Comparisons between ED11 and ED15 groups were performed using Welch’s two-tailed t test. A p value <0.05 was considered statistically significant. Multiple vascular parameters were analyzed as complementary descriptors of the same superficial vascular network. To control the family-wise error rate across these comparisons, p values were adjusted using the Holm-Bonferroni method.
Results
Biological Validation across Development
To validate the performance of the AI-based quantification pipeline, vascular parameters were compared between ED11 (n = 15) and ED15 (n = 15). The total vessel area increased significantly from 22.51 ± 3.79% on ED11 to 28.93 ± 3.32% on ED15 (p < 0.001) (shown in Fig. 5a). Consistently, the total vessel length increased also significantly from 18,539 ± 2,589 pixels to 21,909 ± 1,716 pixels (p < 0.001) (shown in Fig. 5b). Branching complexity also demonstrated a marked significant increase, with the number of branching points rising from 123.13 ± 30.27 to 160.80 ± 24.51 (p < 0.001), reflecting progressive capillary network expansion (shown in Fig. 5c). Mean vessel thickness showed a modest but statistically significant increase from 14.48 ± 1.41 pixels on ED11 to 16.43 ± 1.62 pixels on ED15 (p < 0.01) (shown in Fig. 5d).
Fig. 5.
Quantitative analysis of superficial CAM vascular remodeling across development. Comparison of vascular parameters between ED11 and ED15 is shown as box plots. Each dot represents an individual sample; box plots show median, interquartile range, and range. a Total vessel area (%). b Total vessel length (px). c Number of branching points. d Mean vessel thickness (px). Statistical significance was assessed using Welch’s t test with Holm-Bonferroni correction for multiple comparisons. Significance levels are indicated as follows: *p < 0.05, **p < 0.01, ***p < 0.001.
Recommended Workflow for Superficial CAM Vascular Quantification
Following validation of the AI-based quantification approach, we propose a standardized workflow for two-dimensional analysis of superficial CAM vasculature. The training code and application have been made publicly available to facilitate reproducibility.
The workflow:
-
1.
Isolation of superficial vasculature: the CAM vasculature is exposed, and deeper vascular layers are optically masked by injection of bovine whipped cream, creating a uniform white background for superficial vessels.
-
2.
Image acquisition: the CAM surface is imaged using stereomicroscopy at ×70 magnification under standardized and constant illumination settings across all experimental groups to ensure comparability between developmental stages and treatment conditions.
-
3.
ROI selection: a defined ROI is selected for analysis. ROI size and position may vary depending on study design, and multiple ROIs per image may be analyzed if required. In the present study, a single ROI of 1,024 × 1,024 pixels was selected per image.
-
4.
Vessel segmentation: segmentation is performed automatically using the trained CAM segmentation model to generate binary vessel masks. The training code for the CAM segmentation model is available via the following link: https://github.com/yihu111/cam-vessel-segmentation.
-
5.
Quantitative parameter extraction: from the segmented masks, vascular parameters including total vessel area, total vessel length, number of branching points, and mean vessel thickness are computed.
-
6.
Statistical analysis: quantitative measurements are compared between experimental groups using appropriate statistical methods.
A schematic overview of the complete workflow is shown in Figure 6, illustrating the sequential steps from CAM preparation and layer isolation to automated vessel segmentation and quantitative vascular analysis.
Fig. 6.
Schematic illustration of the workflow for superficial CAM vascular quantification. a Fertilized chicken eggs are incubated at 37.8°C and 60% humidity. b On ED3, the egg shell is fenestrated after removal of 1–2 mL of albumen through the air sac. c Functional isolation of the superficial vasculature is achieved by intra-CAM injection of bovine whipped cream (30% fat). d The CAM surface is imaged using a stereomicroscope, and a region of interest (ROI; 1,024 × 1,024 pixels) is selected. e Vessel segmentation is performed automatically using a trained segmentation model. f Vascular morphometric parameters, including total vessel area, total vessel length, number of branching points, and mean vessel thickness, are extracted from the segmented masks. g Quantitative measurements are compared between experimental groups using appropriate statistical methods. *, p value.
Discussion
In this study, we established an AI-assisted protocol for quantifying the superficial CAM capillary plexus, the vascular compartment most responsive to angiogenic and anti-angiogenic stimuli. Functional isolation of this layer was achieved by intra-CAM injection of bovine whipped cream (30% fat), enabling selective imaging and automated two-dimensional quantification. The segmentation model showed reliable performance (Dice similarity coefficient: 0.831), and application on ED 11 and 15 revealed clear remodeling of the superficial vascular network. These findings are consistent with established descriptions of CAM vascular development, which involve angiogenic sprouting, intussusceptive growth, and progressive network densification rather than substantial thickening of preexisting vessels [8, 32, 33]. Together, this supports the biological relevance and analytical sensitivity of the proposed workflow.
The CAM vasculature is organized into distinct layers, with a dense superficial capillary plexus located beneath the ectoderm and deeper, larger-caliber vessels embedded within the mesoderm [22, 34]. The superficial plexus serves as the primary interface for gas exchange and nutrient delivery and undergoes active remodeling during both physiological development and experimentally induced angiogenesis. Accordingly, most CAM assays quantify changes in vessel density, branching, and morphology within this layer, where endothelial proliferation and sprouting are most prominent in response to stimuli such as tumor xenografts or growth factors [10, 35, 36]. In contrast, deeper vessels form a more stable feeding and draining network and exhibit slower short-term remodeling. Inclusion of these vessels in two-dimensional analyses may therefore dilute angiogenesis-associated changes with relatively stable vascular structures. Restricting analysis to the superficial plexus thus provides a biologically appropriate framework for capturing dynamic angiogenic responses.
A key feature of the present approach is that layer separation is achieved experimentally rather than computationally. By optically masking deeper structures prior to imaging, overlapping vascular layers and depth-related heterogeneity are minimized, enabling robust two-dimensional analysis without three-dimensional reconstruction or depth-resolved imaging, which are often difficult to standardize across laboratories [9, 10]. In addition, the segmentation model was trained and validated in advance using manually annotated vessel masks, allowing direct application of the analysis pipeline without user-side model development, annotation, or parameter tuning. This reduces technical barriers and supports a more standardized and reproducible workflow for routine CAM angiogenesis studies. Together, this protocol reduces workflow complexity compared to existing approaches (see Table 1).
Table 1.
Comparison of conventional CAM vascular analysis and the proposed protocol
| Feature | Conventional analysis | Proposed protocol |
|---|---|---|
| Vessel layer analyzed | Mixed superficial + deep vessels | Superficial vessels only |
| Layer separation | Computational/post-processing | Experimental isolation prior to imaging |
| User-side model training | Often required | Not required |
The segmentation model was trained on a limited number of annotated images, reflecting the labor-intensive nature of microvascular annotation. Despite this, the achieved Dice similarity coefficient of 0.831 indicates good agreement with manual ground truth under standardized conditions. Similar performance has been reported in related microvascular segmentation tasks [16, 37]. Nevertheless, the limited dataset may introduce bias and restrict generalizability to images acquired under different conditions, and future studies using larger and more diverse datasets will be important to further evaluate robustness.
This approach also has inherent limitations. By design, the analysis is restricted to a two-dimensional representation of the superficial vasculature and is therefore not suitable for assessing volumetric parameters, vertical sprouting, or depth-resolved vascular architecture. Comprehensive characterization of vascular networks requires complementary three-dimensional imaging modalities such as microcomputed tomography (MicroCT) or combined MRI/MicroCT approaches [38, 39]. Accordingly, the present method is best suited for assessing dynamic angiogenic remodeling at the CAM surface rather than total vascular volume or deep vascular architecture.
Finally, a direct benchmarking against alternative image analysis platforms such as ImageJ, Ilastik, or CellProfiler was beyond the scope of this study. Future comparative analyses will be important to evaluate quantitative agreement, user requirements, and practical implementation across different CAM vascular analysis workflows.
Conclusion
We developed an AI-assisted protocol for two-dimensional quantification of superficial CAM vasculature. By isolating the superficial blood vessels and applying automated image analysis, this workflow enables reliable measurement of key angiogenic parameters, including vessel area, length, branching points, and thickness. The method is well suited for studying angiogenic remodeling, such as tumor growth and responses to anti-angiogenic treatments. It also supports broader use of standardized vascular quantification in the CAM model and may facilitate its application in preclinical angiogenesis research.
Acknowledgments
We thank Dr. Jörg Wilting for the valuable discussion. We thank Mr. Heinz Biörnsen for technical support and the Campus Frankenforst of the Faculty of Agricultural, Nutritional and Engineering Sciences of the University of Bonn, for providing fertile eggs.
Statement of Ethics
Ethical approval was not required for this study because, according to the German Ordinance on the Protection of Animals Used for Scientific Purposes (Tierschutz-Versuchstierverordnung), specific permission is not required for in ovo experiments on avian embryos prior to hatching. Nevertheless, all procedures were performed in accordance with institutional guidelines and the 3R principles (Replacement, Reduction, and Refinement) to ensure responsible and ethical use of the CAM model.
Conflict of Interest Statement
The authors have no conflicts of interest to declare.
Funding Sources
The study is financially supported by the Deutsche Forschungsgemeinschaft (DFG) – German Research Foundation (Hu 729/22-1 to R.H.). The open access publication fee was covered by the Open-Access-Fonds, University of Bonn. The funder had no role in the design, data collection, data analysis, and reporting of this study.
Author Contributions
Conceptualization: Y.H., N.F.J., S.K., R.H., and Q.P. Investigation and writing: Y.H., R.H., and Q.P. Funding acquisition: R.H.
Funding Statement
The study is financially supported by the Deutsche Forschungsgemeinschaft (DFG) – German Research Foundation (Hu 729/22-1 to R.H.). The open access publication fee was covered by the Open-Access-Fonds, University of Bonn. The funder had no role in the design, data collection, data analysis, and reporting of this study.
Data Availability Statement
All data generated and analyzed during this study are publicly available in the GitHub repository (https://github.com/yihu111/cam-vessel-segmentation). The repository contains the complete set of CAM images used in this study (ED11 and ED15), the manually annotated segmentation masks used for model training, and the code required for vessel segmentation. An installable application for automated vessel segmentation is also provided.
References
- 1. Fischer D, Fluegen G, Garcia P, Ghaffari-Tabrizi-Wizsy N, Gribaldo L, Huang RY, et al. The CAM model-Q&A with experts. Cancers. 2022;15(1):191. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Ribatti D. The chick embryo chorioallantoic membrane in the study of tumor angiogenesis. Rom J Morphol Embryol. 2008;49(2):131–5. [PubMed] [Google Scholar]
- 3. Balke M, Neumann A, Szuhai K, Agelopoulos K, August C, Gosheger G, et al. A short-term in vivo model for giant cell tumor of bone. BMC cancer. 2011;11(1):241. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Ribatti D. The CAM assay in the study of the metastatic process. Exp Cel Res. 2021;400(2):112510. [DOI] [PubMed] [Google Scholar]
- 5. Russell WMS, Burch RL. Principles of humane experimental technique. Springfield, IL: Charles C. Thomas; 1959. [Google Scholar]
- 6. Schneider-Stock R, Ribatti D. The CAM assay as an alternative in vivo model for drug testing. Handb Exp Pharmacol. 2021;265:303–23. [DOI] [PubMed] [Google Scholar]
- 7. Ranjan RA, Muenzner JK, Kunze P, Geppert CI, Ruebner M, Huebner H, et al. The chorioallantoic membrane xenograft assay as a reliable model for investigating the biology of breast cancer. Cancers. 2023;15(6):1704. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Ribatti D. The chick embryo chorioallantoic membrane (CAM). A multifaceted experimental model. Mech Dev. 2016;141:70–7. [DOI] [PubMed] [Google Scholar]
- 9. Staton CA, Reed MW, Brown NJ. A critical analysis of current in vitro and in vivo angiogenesis assays. Int J Exp Pathol. 2009;90(3):195–221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Nowak-Sliwinska P, Segura T, Iruela-Arispe ML. The chicken chorioallantoic membrane model in biology, medicine and bioengineering. Angiogenesis. 2014;17(4):779–804. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Zhu D, Zhang J, Cui H, Mao Z, Li P, Luo Q. Short-term and long-term effects of optical clearing agents on blood vessels in chick chorioallantoic membrane. J Biomed Opt. 2008;13(2):021106. [DOI] [PubMed] [Google Scholar]
- 12. Said SS, Yin H, Elfarnawany M, Nong Z, O’Neil C, Leong H, et al. Fortifying angiogenesis in ischemic muscle with FGF9-loaded electrospun poly(ester amide) fibers. Adv Healthc Mater. 2019;8(8):e1801294. [DOI] [PubMed] [Google Scholar]
- 13. Chen L, Wang S, Feng Y, Yu J, Coudyzer W, Van Ongeval C, et al. Development and characterization of a chick embryo chorioallantoic membrane (CAM) based platform for evaluation of vasoactive medications. Microvasc Res. 2022;142:104372. [DOI] [PubMed] [Google Scholar]
- 14. Kuri PM, Pion E, Mahl L, Kainz P, Schwarz S, Brochhausen C, et al. Deep learning-based image analysis for the quantification of tumor-induced angiogenesis in the 3D in vivo tumor model-establishment and addition to laser speckle contrast imaging (LSCI). Cells. 2022;11(15):2321. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Chen L, Wang M, Feng Y, Gao L, Yu J, Geng L, et al. Multi-parametric investigations on the effects of vascular disrupting agents based on a platform of chorioallantoic membrane of chick embryos. Quant Imaging Med Surg. 2024;14(2):1729–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Faihs L, Firouz B, Slezak P, Slezak C, Weißensteiner M, Ebner T, et al. A novel artificial intelligence-based approach for quantitative assessment of angiogenesis in the ex ovo CAM model. Cancers. 2022;14(17):4273. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Annese T, Tamma R, Ribatti D. IKOSA® CAM assay application to quantify blood vessels on chick chorioallantoic membrane (CAM). Methods Mol Biol. 2023;2572:129–39. [DOI] [PubMed] [Google Scholar]
- 18. Salvante ERG, Popoiu AV, Barb AC, Cosma AA, Fenesan MP, Saxena AK, et al. Artificial intelligence (AI) based analysis of in vivo polymers and collagen scaffolds inducing vascularization. Vivo. 2024;38(2):620–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Salvante ERG, Popoiu AV, Saxena AK, Popoiu TA, Boia ES, Cimpean AM, et al. Glycosaminoglycans modulate the angiogenic ability of type I collagen-based scaffolds by acting on vascular network remodeling and maturation. Bioeng. 2024;11(5):423. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lawate S, Kulloli A, Shetty S, Martande S, Deshetti K, Mante N. The effect of different time intervals of diode laser irradiation on the in Ovo chick embryo yolk sac membrane for angiogenic differentiation - an in Ovo pilot experimental study. Lasers Med Sci. 2025;40(1):525. [DOI] [PubMed] [Google Scholar]
- 21. Fontenele Gomes JG, de Sousa Barbosa B, Timah AB, Ferreira de BLN, Alves Damaceno do Lago JP, Alves de Sousa A, et al. Dihydromyricetin: a multitarget anticancer agent exhibiting cytotoxic and anti-angiogenic activities in preclinical models. Curr Top Med Chem. 2026. [DOI] [PubMed] [Google Scholar]
- 22. Fuchs A, Lindenbaum ES. The two- and three-dimensional structure of the microcirculation of the chick chorioallantoic membrane. Acta anatomica. 1988;131(4):271–5. [DOI] [PubMed] [Google Scholar]
- 23. Lindsay J, Kabir M, Gilroy N, Dyer G, Brice L, Moore J, et al. Epidemiology of complementary and alternative medicine therapy use in allogeneic hematopoietic stem cell transplant survivorship patients in Australia. Cancer Med. 2016;5(12):3606–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Ribatti D, Nico B, Cimpean AM, Raica M, Crivellato E, Ruggieri S, et al. B16-F10 melanoma cells contribute to the new formation of blood vessels in the chick embryo chorioallantoic membrane through vasculogenic mimicry. Clin Exp Med. 2013;13(2):143–7. [DOI] [PubMed] [Google Scholar]
- 25. Zhou P, Cheng L, Tao H, Hintze M, Wang Y, Pu Q, et al. Fibroblast growth factor 8 promotes in vitro neurite outgrowth of placode-derived petrosal and nodose ganglia to varying degrees. Ann Anat - Anatomischer Anzeiger. 2024;256:152323. [DOI] [PubMed] [Google Scholar]
- 26. Hamburger V, Hamilton HL. A series of normal stages in the development of the chick embryo. J Morphol. 2005;88(1):49–92. [PubMed] [Google Scholar]
- 27. Dünker N, Jendrossek V. Implementation of the chick chorioallantoic membrane (CAM) model in radiation biology and experimental radiation oncology research. Cancers. 2019;11(10):1499. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Flock ST, Jacques SL, Wilson BC, Star WM, van Gemert MJ. Optical properties of Intralipid: a phantom medium for light propagation studies. Lasers Surg Med. 1992;12(5):510–9. [DOI] [PubMed] [Google Scholar]
- 29. Michels R, Foschum F, Kienle A. Optical properties of fat emulsions. Opt Express. 2008;16(8):5907–25. [DOI] [PubMed] [Google Scholar]
- 30. Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention. Cham: Springer international publishing; 2015. p. 234–41. [Google Scholar]
- 31. Xie E, Wang W, Yu Z, Anandkumar A, Alvarez JM, Luo P. SegFormer: simple and efficient design for semantic segmentation with transformers. Adv Neural Inf Process Syst. 2021;34:12077–90. [Google Scholar]
- 32. Ausprunk DH, Knighton DR, Folkman J. Differentiation of vascular endothelium in the chick chorioallantois: a structural and autoradiographic study. Dev Biol. 1974;38(2):237–48. [DOI] [PubMed] [Google Scholar]
- 33. Djonov V, Schmid M, Tschanz SA, Burri PH. Intussusceptive angiogenesis: its role in embryonic vascular network formation. Circ Res. 2000;86(3):286–92. [DOI] [PubMed] [Google Scholar]
- 34. Melkonian G, Munoz N, Chung J, Tong C, Marr R, Talbot P. Capillary plexus development in the day five to day six chick chorioallantoic membrane is inhibited by cytochalasin D and suramin. J Exp Zool. 2002;292(3):241–54. [DOI] [PubMed] [Google Scholar]
- 35. West DC, Thompson WD, Sells PG, Burbridge MF. Angiogenesis assays using chick chorioallantoic membrane. In: Murray JC, editor. Angiogenesis protocols. Totowa, NJ: Humana Press; 2001. p. 107–29. [DOI] [PubMed] [Google Scholar]
- 36. Deryugina EI, Quigley JP. Chapter 2. Chick embryo chorioallantoic membrane models to quantify angiogenesis induced by inflammatory and tumor cells or purified effector molecules. Methods Enzymol. 2008;444:21–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Jin K, Huang X, Zhou J, Li Y, Yan Y, Sun Y, et al. Fives: a fundus image dataset for artificial intelligence based vessel segmentation. Sci Data. 2022;9(1):475. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Woloszyk A, Wolint P, Becker AS, Boss A, Fath W, Tian Y, et al. Novel multimodal MRI and MicroCT imaging approach to quantify angiogenesis and 3D vascular architecture of biomaterials. Sci Rep. 2019;9(1):19474. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Zhou Y, Hou P, Wang F, Li B, Gao J. Primary hepatic malignant vascular tumors: a follow-up study of imaging characteristics and clinicopathological features. Cancer Imaging. 2020;20(1):59. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
All data generated and analyzed during this study are publicly available in the GitHub repository (https://github.com/yihu111/cam-vessel-segmentation). The repository contains the complete set of CAM images used in this study (ED11 and ED15), the manually annotated segmentation masks used for model training, and the code required for vessel segmentation. An installable application for automated vessel segmentation is also provided.
