An automatic method for spine detection and spine tracking in in vivo images

The variation in dendritic branch morphology and spine density offers the scientists information about the function of the treatment to neuron disease. In particular we study the passive immunotherapy treatment on dendritic spine loss of Alzheimer's disease. This paper presents an automated approach for the detection of spines and tracking of spine evolution at different time points. Most automated processing methods are developed for in vitro images. Here we investigate the possibility of automated detection and tracking of the spines on lower contrast in vivo confocal microscopy images. We propose a curvilinear structure detector to determine the medial axis of the dendritic backbone and the spines connected to the backbone. In addition, we present a maximum likelihood based technique optimized through dynamic programming to find the graph homomorphism between two image graph structures at different time points to track the growth or loss of spines. Our results show that on eight data samples, we can achieve accuracies of 91.2% for detecting spines and 78.3% for tracking spine correspondences at different time points.