Cell phenotype classification using M-phase features in live-cell bright field time-lapse microscopy

Makes a first approach to the classification of living cells in four phenotypes related to chemosensitivity: cells that were able to repair their damage in DNA, in cell cycle arrest, with DNA damage proliferation, and apoptotic cells; in order to obtain individualized results, depending on the subpopulation to which the cell belongs. Brightfield microscopy images with very long contrast and very high cell density are used as the only input. To overcome the difficulty of working with images of this type, a pattern recognition algorithm that does not follow the conventional flow was devised. The traditional segmentation of objects in the background of the image was replaced by a detection of mitosis events and the feature extraction is done by tracking the cell in previous and subsequent frames in order to describe the M phase with: duration, number of daughter cells, radio average, average intensity, distance traveled. Using these features, a classifier based on linear discriminant analysis is able to achieve general classification rates with a F1-Score greater than 75%