Supervised-Learning-Driven Interrogation of Organ-on-a-Chip Quality from Microscopy Images

We leverage supervised learning to realize high-throughput and objective interrogation of organ-on-a-chip quality from microscopy images. More than 600 images of two lung-related cell types in good and bad quality lung-on-a-chip models are assessed. We demonstrate that trained classifiers can predict different cell types and lung-on-a-chip quality with AUC and accuracy of above 95% and 83%, respectively. Coupled with a dimensionality reduction approach, the predictive capacity of certain classifiers can be enhanced, while the computational cost of resource-intensive algorithms can be significantly reduced. This study is anticipated to further encourage the implementation of machine learning for automating organ-on-a-chip development.