A new algorithm for 3D dendritic spine detection

It has been shown in recent research that there is a close relationship between neurological functions of neuron and its morphology. As manual analysis of large data sets is too tedious and may be subjected to user bias, a computer aided processing method is urgently desired. In this paper, we propose an automatic approach for 3D dendritic spine detection, which can greatly help neuron-biologists to obtain morphological information about a neuron and its spines. The work mainly consists of segmentation and spine component detection. The segmentation of dendrite and spine components is carried out by means of 3D level set based on local binary fitting model, which yields better results than global threshold method. As for spine component detection, an efficient approach is presented which consists of backbone extraction, detached and attached spine components detection. The detection is robust to noise and the detected spines are well represented. We validate our algorithm with real 3D neuron images and the result reveals that it works well.