Abstract
Abstract The difficulty of image labeling and analysis poses a problem for many areas of scientific research, including the recently developed Organotypic Brain Slice Culture (OBSC) tumor characterization platform. An OBSC experiment generates dozens of images which, up to this point, had to be manually quantified. In order to increase the efficiency, consistency, and accuracy of the OBSC image analysis process, we have created a computer program which automates the analysis of OBSC images. This platform uses a number of computer vision (CV) techniques to identify objects of interest and measure their corresponding signals. For fluorescent images of single tumor spots, a straightforward thresholding algorithm divides the image into foreground and background. This automated approach substantially reduces analysis time, and the thresholding algorithm measures unusually shaped tumor spots up to 9% more accurately than previously used manual methods. Other images capture an entire plate of up to 12 OBSCs, complicating the image analysis task. A machine-learning algorithm trained on the web-based platform Biodock could identify individual OBSCs, even if they were touching each other or the sides of the wells. The ML-powered methods consistently reproduced killing curves for treatment effects on tumor tissue as well as the brain tissue substrate, and these results were obtained in much less time. By combining various CV approaches to image analysis, we have completely automated the image analysis component of our OBSC platform. This approach could be adapted to other tumor co-culture platforms.
Accepted Version
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call