Segmenting neurons in electronic microscopy via geometric tracing

Abstract

In this paper we describe a system that is being used for the segmentation of neurons in images obtained from electronic microscopy. These images are extremely noisy, and ordinary active contours techniques detect spurious objects and fail to detect the neuron boundaries. The algorithm here described is based on combining robust anisotropic diffusion with minimal weighted-path computations. After the image is regularized via anisotropic diffusion, the user clicks points on the boundary of the desired object, and the algorithm completes the boundary between those points. This tracing is based on computing paths of minimal weighted distance, where the weight is given by the image edge content. Thanks to advanced numerical algorithms, the algorithm is very fast and accurate. We compare our results with those obtained with PictureIt, a commercially available general purpose image processing package developed by Microsoft.