Identifying heterogeneity in cell culture through machine learning-enabled lens-free microscopy

Cellular heterogeneity is the hallmark of many cancers, referring to the co-existence of different phenotypes with very distinct biological behaviours in single isolates. Automatically detecting single-cell heterogeneity is therefore critical, and can provide important information on cancer initiation. We present a clustering algorithm that allows identifying heterogeneity in cell culture from time-lapses of lensless microscopic images. A preliminary segmentation and tracking pipeline extract quantitative features (morphology, motility and reproduction cycle) for each cell. An unsupervised learning algorithm then clusters the time-series of the cell tracks measurements, in two steps. We validate our approach on co-cultures of mixed cells lines, and on murine fibroblasts isolated from genetically modified mice, where the modified genome promotes the establishment of cancers and heterogeneous cell morphologies and behaviours