Automatical detecting and connecting the mitochondria from the serial EM images

It is well known that mitochondria are one of the most important organelles found in the majority of cells. In recent years, with the development of scanning electron microscope(SEM), we have had a deeper understanding of the internal structure of the cell. However, obtaining the detection and connection of mitochondria form EM images is still a great challenge. In this paper, the Faster R-CNN algorithm is put forward to detect mitochondria, which differs from other algorithms, the classifier constructed is used to detect directly. Then we fuse the multi-layer information to obtain the connected relationship. On the basis of that, some improvement measures have been taken to achieve better results. Furthermore, compared with Adaboost, the method proposed in this paper proved to be more effective according to the results of experiment on automated tape-collecting ultramicrotome scanning electron microscopy(ATUM-SEM) images. Finally, partial results of 3D visualization are also shown in the end of this paper.