BCSnet: A U-Net-based Model for Segmentation of Brain Cells in Trypan Blue Images

One of the methods for quantifying the survival of brain cells is the direct counting of stained neurons after they are fixed on the plate. This method requires a lot of time and effort associated with the need for manual cell counting and qualification in histology to detect visual differences between alive neuron cells and non-alive ones. The article deals with the problem of semantic segmentation of images of rat brain cells stained with trypan blue for automatic counting of alive brain cells. To solve the problem, a mathematical model has been developed in the form of a convolutional neural network based on the U-Net architecture. As a result of research, the best model has a metric of 0.9189 (Sorensen-Dies coefficient). The trained neural network makes it possible to evaluate the alive neurons three orders of magnitude faster than with manual calculation and with an accuracy that is not inferior to a histologist.