Detection of neuron membranes in electron microscopy images using a serial neural network architecture.

摘要: Study of nervous systems via the connectome, map connectivities all neurons in that system, is a challenging problem neuroscience. Towards this goal, neurobiologists are acquiring large electron microscopy datasets. However, shear volume these datasets renders manual analysis infeasible. Hence, automated image methods required for reconstructing connectome from very collections. Segmentation images, an essential step reconstruction pipeline, because noise, anisotropic shapes and brightness, presence confounding structures. The method described paper uses series artificial neural networks (ANNs) framework combined with feature vector composed intensities sampled over stencil neighborhood. Several ANNs applied allowing each ANN to use classification context provided by previous network improve detection accuracy. We develop serial show learned does traditional ANNs. also demonstrate advantages membrane methods. results significant towards system connectome.