Tracking on tree-like structures in 3-D confocal images

Confocal microscopy is qualified to perform volume scans of nerve cells with dendrites and spines. Length and diameter of the dendrite branches and the spines should be determined to analyze the influence of learning processes. A prerequisite for that is the recognition of the dendritic structure with branching off and spines. Because the microscope operates at the resolution limit the images are blurry, noisy and only poorly sampled. In contrast to other methods which are based on binary images and thinning algorithms, our method tracks the dendritic tree faster and in the gray-level domain using simple geometric models. An explicit segmentation is unnecessary and knowledge about shape and structure of the dendrite is included as a-priori information (nearly circular cross-section, approximate diameter and branching off angle). For large trees, first a low resolution scan is captured to create a rough model. The algorithm allows to refine this model using higher resolution scans for interesting regions along the dendrite. The large unimportant areas between the dendrite branches are not scanned at high resolution to save time and disc space. In a second step, the parameters of the model are adapted to the microscope image by minimizing the deviation of the microscope image from the model image convolved by the microscope point spread function. Features like number, diameter, length and position of the dendrite branches and spines can be easily calculated from the model. An interactive user intervention is possible at the model domain.