Machine Learning-Aided Three-Dimensional Morphological Quantification of Angiogenic Vasculature in the Multiculture Microfluidic Platform

A plethora of in vitro models have been the focus of intense research to mimic the native physiological system more accurately. Among them, organ-on-a-chip or microfluidic devices gave notable results in reconstructing reproducible three-dimensional vascularized microenvironments specific to certain organs, using various approaches. However, current strategies for quantifying the morphological variation of on-chip microvascular networks (MVNs) merely remain in the 2D domain with limited indicators, which might result in misinterpretations and the loss of biologically significant information. To this end, we introduce a novel machine learning-assisted 3D analysis pipeline that is capable of extracting major assessment parameters quantifying on-chip MVNs' morphological variation. We utilized the MV-IMPACT (MicroVascular Injection-Molded Plastic Array 3D Culture) platform for data acquisition, a high-throughput experimental device that offers standardized form factor compatibility. Meso-skeletal depiction of the microvasculature and skeleton segmentation via improved graph convolutional network allowed for a more accurate structural analysis of the angiogenic network than any other approach. We show that our method outperforms conventional projection-based analysis by providing satisfactory concordance with manual investigation. Our approach offers a different avenue for analyzing the 3D structure of MVNs compared to conventional voxel-based methods as it allows for greater flexibility in handling complex structures, including the lumen. With its robustness and potential for future applications, we anticipate that our method can provide an opportunity to uncover the fundamental aspects of vessel physiology that may have been overlooked, and aid in the development of reliable preclinical models for biopharmaceutical applications.