Automatic acquisition of immune cells location using deep learning for automated analysis
Shoya Kusunose, Yuki Shinomiya, Takashi Ushiwaka, Nagamasa Maeda, Yukinobu Hoshino · 2020
Immune cells are analyzed for the elucidation of the human immune system by researchers. They take a cells' movie and track manually the cells to analyze. But the manual process is hard for them. The aim of this study was reducing the load of theirs with automated cells' picking up. To pick up cells, “Recognition Frequency Space” was generated using immune cells' classifier that was made by CNN that a method of deep learning. The classifier was trained while comparing several conditions to decide the best performance classifier. Recognition Frequency Space was indicated the number of times cells are recognized from the frame image of the movie. Then, it was checked if cells were picked up from high recognized points of the space. As the result, 9 immune cells were picked up from the first frame image. The result indicates that multiple immune cells can be picked up automatically if it is used the space, and the method can reduce the load of the cell's analyst.