Applying a Convolutional Neural Network for Cancer Cell Image Segmentation

The recent resurgence in AI has enabled scientists to use neural networks for efficiently identifying trends in large data sets and has also brought about new ways that biological processes can be studied. Specifically, analyzing the behaviors of certain macrophage and breast cancer cocultures grown in vitro. A convolutional neural network (UNET) is trained on time series images of MDA-MB-231 breast cancer cells and their respective ground truths, according to specific color channels, so that the network can learn to segment the cell clusters. The neural network then generates masks for all images which are then used to extract information about the behaviors of the cell clusters over time. Time series graphs of area, perimeter, solidity, and eccentricity are found and show the effect of different macrophages on clusters of MDA-MB-231 cancer cells. An interesting feature was found in the graphs, at a certain time point there is a jump or dip in all the graphs due to an external permutation. The trends revealed in these graphs will be used to learn about how these clusters respond to the macrophages.