Computer Vision for Substrate Detection in High‐Throughput Biomaterial Screens Using Bright‐Field Microscopy

• Published in: Advanced intelligent systems Vol. 7; no. 5
• Main Authors: Owen, Robert, Nasir, Aishah, Amer, Mahetab H., Nie, Chenxue, Xue, Xuan, Burroughs, Laurence, Denning, Chris, Wildman, Ricky D., Khan, Faraz A., Alexander, Morgan R., Rose, Felicity R. A. J.
• Format: Journal Article
• Language: English
• Published: Weinheim John Wiley & Sons, Inc 01.05.2025; Wiley
• Subjects: Artificial intelligence; Artificial neural networks; Automation; Biomedical materials; CellProfiler; Cells; Chemistry; Computer vision; Cytoskeleton; Datasets; detectron2; feature identification; Fluorescence; image analysis; Image segmentation; Machine learning; Medical research; Neural networks; polymer microarray; Polymers; Screens; Software; TopoChip; Topography
• ISSN: 2640-4567; 2640-4567
• DOI: 10.1002/aisy.202400573

High‐throughput screening (HTS) can be used when ab initio information is unavailable for rational design of new materials, generating data on properties such as chemistry and topography that control cell behavior. Biomaterial screens are typically fabricated as microarrays or “chips,” seeded with the cell type of interest, then phenotyped using immunocytochemistry and high‐content imaging, generating vast quantities of image data. Typically, analysis is only performed on fluorescent cell images as it is relatively simple to automate through intensity thresholding of cellular features. Automated analysis of bright‐field images is rarely performed as it presents an automation challenge as segmentation thresholds that work in all images cannot be defined. This limits the biological insight as cell response cannot be correlated to specifics of the biomaterial feature (e.g., shape, size) as these features are not visible on fluorescence images. Computer Vision aims to digitize tasks humans do by sight, such as identify objects by their shape. Herein, two case studies demonstrate how open‐source approaches, (region‐based convolutional neural network and algorithmic [OpenCV]), can be integrated into cell‐biomaterial HTS analysis to automate bright‐field segmentation across thousands of images, allowing rapid, spatial definition of biomaterial features during cell analysis for the first time. It is known that biomaterial properties (e.g., shape, chemistry) direct how the body reacts to implanted materials. However, what features cause specific responses is not well known. High‐throughput screening can identify instructive properties, but manual feature segmentation in large brightfield image datasets limits discovery. Herein, computer vision is used to rapidly and spatially define biomaterial features for downstream biological analysis.