Automation of Foci-Cell-State Judgement with Regression Models
Keiko Itano, Koji Ochiai, Takahide Matsushima, Hiroshi Asahara, Koichi Takahashi
Image analysis of cells is commonly used to judge cell state and cell phenotype. In a previous paper, we reported the manual foci-cell identification process’s automation by applying image processing and machine learning methods to fluorescent foci-cell images. Here, we present the details of our approach to improving the proposed automated system. Specifically, we use the Gaussian mixture model (GMM) for image segmentation, depict dead cells as outliers, and add new features not included in scikit-learning regionprops. Thus, we defined new features related to foci cells’ properties, which were not included in the scikit-learn regionprops. Through the new approaches, we improved the accuracy of the regression models to an adequate level. In addition, an analysis of fitted model information showed that the new features were useful for foci-cell identification.