A Quantitative Evaluation Method Using Deep Learning for Quality Control of Chimeric Mice with Humanized Livers

In this paper, we propose a classification method of a grading of human hepatocytes for quality control of chimeric mice with humanized livers. Currently, the grade of chimeric mice with humanized livers are classified into five categories. That is Grade S, A, B, C, and D. The five grades are currently determined by visual inspection under a stereo microscope by an experienced person. This causes problems such as variation among observers and difficulty in quantifying each grade. In this paper, we develop a method to classify the grade of human hepatocytes quantitatively by using deep learning. In cell grade classification, it is difficult to learn with a single labeling because the state of cells in a single microscopic image is not uniform. Therefore, in this paper, we introduce two preprocessing methods to clean the training data. As a result, an accuracy of 83.4% was obtained for 5-class classification.