Influences of Microscopic Imaging Conditions on Accuracy of Cell Morphology Discrimination Using Convolutional Neural Network of Deep Learning.

Recently, automated cell culture devices have become necessary for cell therapy applications. The maintenance of cell functions is critical for cell expansion. However, there are risks of losing these functions, owing to disturbances in the surrounding environment and culturing procedures. Therefore, there is a need for a non-invasive and highly accurate evaluation method for cell phenotypes. In this study, we focused on an automated discrimination technique using image processing with a deep learning algorithm. This study aimed to clarify the effects of the optical magnification of the microscope and cell size in each image on the discrimination accuracy for cell phenotypes and morphologies. Myoblast cells (C2C12 cell line) were cultured and differentiated into myotubes. Microscopic images of the cultured cells were acquired at magnifications of 40× and 100×. A deep learning architecture was constructed to discriminate between undifferentiated and differentiated cells. The discrimination accuracy exceeded 90% even at a magnification of 40× for well-developed myogenic differentiation. For the cells under immature myogenic differentiation, a high optical magnification of 100× was required to maintain a discrimination accuracy over 90%. The microscopic optical magnification should be adjusted according to the cell differentiation to improve the efficiency of image-based cell discrimination.