Spheroids are 3D cell aggregates that better recapitulate the physiological microenvironment as compared to 2D cell culture systems. This makes spheroids useful for various disease models, including cancer. The model that more closely resembles the in vivo cell environment enables better assessment of cellular behavior and responses (e.g. treatment with potential anti-cancer drugs). Although spheroids are helpful in research and testing new potential drugs, their handling is often less standardized and straightforward compared to the handling of a 2D cell culture. In order to eliminate variation and subjectivity in spheroid culturing, CytoSMART and KNIME have teamed up for a proof-of-concept project that establishes an automated image analysis pipeline to classify whether spheroids in culture plates are ready for further experiments.
KNIME builds software to create and productionize data science. Its users can be found in many departments across a wide range of industries in over 60 countries. KNIME Analytics Platform is based on openness and collaboration: it is an open-source platform where functionalities are added continuously by their own developers and their users. The CytoSMART Lux3 BR and Lux2 are compact brightfield microscopes which fit in a cell culture incubator, so that cells can be monitored without disrupting their ideal culture environment. Combined, these technologies formed the pipeline at the cutting edge of automated spheroid imaging and analysis that is presented in this case study. Here, the final decision on spheroid status was still made by a human researcher, but to demonstrate that the presented setup is also suitable for end-to-end lab digitalization and automation, the images for classification were retrieved from a Standardization in Lab Automation (SiLA) server. Afterward, the results were also reported back to the SiLA server.
An overview of the case study can be found in this video:
To read the extensive report of this case study, please refer to the application note ‘An automated pipeline for imaging and analyzing spheroid development’, which was written in collaboration with KNIME.
The figure below illustrates a part of the results. The developed image analysis model was correct in every instance where human observers annotated the presence of spheroids, as well as all images where the model predicted spheroids to be absent. The only incorrectly classified images were manually annotated to contain only single cells and cell clusters, whereas the model predicted the image to contain spheroids.
Check out KNIME article about this project: www.knime.com/blog/spheroid-detection-deep-learning