Temporal Matching of Dendritic Spines in Confocal Microscopy Images of Neuronal Tissue Sections

Recent research has revealed that morphological characteristics of neuronal structure are closely related to normal neural functions such as learning and memory. Neuronal dendrites and their spines define the inter-connectivity of the neural network and hence, are important predictors of its function. Many neuronal functions are observed to be correlated with the appearance or disappearance of neural structures. In this paper, we temporally track the evolution of spines on the dendrite. Three-dimensional images were acquired by the digitization of neuronal tissue sections using a two-photon laser scanning microscopy with a 40X objective, 0.8 NA, over transgenic mice expressing EGFP. We describe a pipeline of processing algorithms for solving the spine correspondence problem automatically. Our framework requires the segmentation and extraction of the skeleton of the dendrite that is aligned with a rigid registration algorithm. We then build a graph model that is spatially aligned with the dendrite, the nodes of which represent the spine branches. A maximum likelihood estimation framework is employed to match graph models from different time-points. We report the results from the automated matching and validate it with manual measurements. The sensitivity and specificity of the pipeline is evaluated with available ground-truth.