Computerized colony classification of induced pluripotent stem cells using gaussian naïve bayes model on phase contrast images

This study aims to develop a computerized software tool that automatically separates the colony contour region of induced pluripotent stem cells (iPSCs) and classifies the health conditions of the colonies using the Gaussian naïve Bayes (GNB) model. The occluded colony regions were automatically segmented based on the phase contrast images using image processing techniques to obtain quantitative morphological features for classification. The sequential forward selection method was utilized to extract optimized features for the identification of colony conditions. The GNB model was adapted to validate the individual colony features and their combinations using a five-fold cross validation method for classification. Furthermore, the classification performance of GNB was compared with that of the k-nearest neighbor (k-NN) method. The classification performance of the combination of features using the GNB approach presented the highest sensitivity (91.4%), specificity (88.2%), and accuracy (90.8%) for the classification of the colonies of iPSCs. Furthermore, compared with the k-NN classifier (14.3%), GNB showed lower misclassification rate (9.2%) in classification. Based on experimental results, we concluded that the proposed automated colony region segmentation and classification based on the combination of features using GNB model is precise and cost-effective for the classification of health conditions of iPSC colonies.