A proposal for automatic tracking of immune cells using deep learning

Abstract

On immune cells' analysis, analyst has hard working such as manual cell tracking to analyze in cell's movies. The aim of this study was reducing the load of theirs with automated tracking cells using the classifier. Cell's initial positions in the video was picked up from points of high recognized frequency in "Recognized Frequency Space". Recognized Frequency Space was generated from the positions that was recognized cells by the classifier. The classifier was trained by CNN which is one of deep learning methods. Then, cell's next positions were picked up using the cell's positions in the previous frame to track cells. As the result of used "Recognized Frequency Space" made by the classifier, 8 single cells were picked up from the frame image. And 5 of 8 cells could be tracked for 100 frames. This indicates that isolated immune cells can be tracked automatically if it is followed the method, and the method can reduce the load of cell's analyst.