An Unsupervised Machine Learning Algorithm to Detect Undifferentiated Cell Clusters of Immortalized Human Cervical Epithelial Cell

Abstract

Cell differentiation is a progressive process and hard to quantitate without advanced biotechnological methods. In this study, a machine learning (ML) algorithm is introduced to detect the undifferentiated cell clusters and improve time and labor efficiencies by clustering image features extracted from the changing morphology of immortalized cervical cells. The methodology involves taking phase-contrast image data from the monolayer cell culture of the human cervical epithelial cell. The normalized histogram features and Haralick texture features from each dividing tile of input images are used in a simple k-means clustering training. The resulting colored maps are generated by filling each tile with a specific color according to its classification label. The targeted color representing the undifferentiation is selected automatically. Then simple image processing techniques are applied to analyze the colored map and outline the contour of undifferentiated cell clusters on the input images. The results showed that the undifferentiated cell clusters are indicated clearly in the images. After visually comparing to the ground truth cell morphology, the proposed method could accurately pinpoint the major undifferentiated cell clusters with minimal costs.