A microphysiological system is a 3D in vitro model that mimics human tissue environments with high physiological relevance. This approach enhances our understanding of disease mechanisms, drug responses, and treatment development. In this presentation, we focus on the usage of an in-house microvascular network (MVN) model for biopharmaceutical research and describe an associated image-based analysis method. While many tools exist to quantify fluorescent MVN images, they often employ multiple image processing steps and may lack the needed specificity to best capture (segment) the 3D morphology of microvessel structures. Additionally, these tools fall short in automating the workflow and output of desired metrics. In contrast, deep learning techniques provide a highly trainable and flexible approach for MVN segmentation, enabling adaptable identification of image features based on morphology, texture, and vessel structure. Leveraging these advantages, we have developed a 3D U-Net segmentation model and complete workflow for automated analysis on large MVN image datasets. This approach allows us to assess the permeability of various molecules with high precision and throughput. In conclusion, our 3D analysis method offers insights beyond traditional 2D approaches, enhancing study possibilities. The deep learning model accurately segments microvessel structures, making it valuable for measuring molecule permeability and in the application of MVNs to other areas of research.

Learning Objectives:

1. Describe the role and importance of microphysiological systems (MPS), specifically microvascular network (MVN) models, in enhancing biopharmaceutical research.

2. Analyze the limitations of traditional image analysis techniques for MVN segmentation and understand how deep learning, specifically the 3D U-Net model, addresses these challenges.

3. Demonstrate how the automated 3D image analysis workflow enables high-precision and high-throughput measurement of molecule permeability in MVN research.