Machine Learning Identification and Classification of Mitosis and Migration of Cancer Cells in a Lab-on-CMOS Capacitance Sensing platform.

Cell culture assays play a vital role in various fields of biology. Conventional assay techniques like immunohistochemistry, immunofluorescence, and flow cytometry offer valuable insights into cell phenotype and behavior. However, each of these techniques requires labeling or staining, and this is a major drawback, specifically in applications that require compact and integrated analytical devices. To address this shortcoming, CMOS capacitance sensors capable of conducting label-free cell culture assays have been proposed. In this paper, we present a computational framework for further augmenting the capabilities of these capacitance sensors. In our framework, identification and classification of mitosis and migration are achieved by leveraging observations from measured capacitance time series data. Specifically, we engineered two time series features that enable discriminating cell behaviors at the single-cell level. Our feature representation achieves an area under curve (AUC) of 0.719 in the receiver operating characteristic (ROC) curve. Additionally, we show that our feature representation technique is applicable across arbitrary experiments, as validated by a leave-one-run-out test yielding an F-1 score of 0.803 and a G-Mean of 0.647.