An efficient method for neuronal tracking in electron microscopy images

With the introduction of deep learning, a wave of artificial intelligence research has been set off again. Scientists focus on brain-inspired intelligence, namely, try to get inspiration from the brain nervous system and cognitive behavior mechanism, to develop intelligent computing models as well as algorithms with stronger information representation, processing and learning ability. So, the study of neurons and the connections between neurons of brain are needed. One major obstacle of reconstruction lies in segmenting and tracking neuronal processes. Electron microscopy is producing neurons images rapidly. In response, we propose an efficient method for neuronal tracking in electron microscopy images to help scientists reconstruct complex neurons. First, we track neurons by kernelized correlation filter to get candidate neuron; then we calculate overlap area and distance of the contours between two consecutive images to get final neuron. We evaluate the performance of our method on a public electron microscopy dataset. The method is superior in accuracy and efficiency.