Moritz Helmstaedter

Many image segmentation algorithms first generate an affinity graph and then partition it. We present a machine learning approach to computing an affinity graph using a convolutional network (CN) trained using ground truth provided by human experts. The CN affinity graph can be paired with any standard partitioning algorithm and improves segmentation accuracy significantly compared to standard hand-designed affinity functions. We apply our algorithm to the challenging 3D segmentation problem of reconstructing neuronal processes from volumetric electron microscopy (EM) and show that we are able to learn a good affinity graph directly from the raw EM images. Further, we show that our affinity graph improves the segmentation accuracy of both simple and sophisticated graph partitioning algorithms. In contrast to previous work, we do not rely on prior knowledge in the form of hand-designed image features or image preprocessing. Thus, we expect our algorithm to generalize effectively to arbitrary image types.

Large-scale connectomics requires dense staining of neuronal tissue blocks for electron microscopy (EM). Here we report a large-volume dense en-bloc EM staining protocol that overcomes the staining gradients, which so far substantially limited the reconstructable volumes in three-dimensional (3D) EM. Our protocol provides densely reconstructable tissue blocks from mouse neocortex sized at least 1 mm in diameter. By relaxing the constraints on precise topographic sample targeting, it makes the correlated functional and structural analysis of neuronal circuits realistic.

The dense circuit structure of mammalian cerebral cortex is still unknown. With developments in three-dimensional electron microscopy, the imaging of sizable volumes of neuropil has become possible, but dense reconstruction of connectomes is the limiting step. We reconstructed a volume of ~500,000 cubic micrometers from layer 4 of mouse barrel cortex, ~300 times larger than previous dense reconstructions from the mammalian cerebral cortex. The connectomic data allowed the extraction of inhibitory and excitatory neuron subtypes that were not predictable from geometric information. We quantified connectomic imprints consistent with Hebbian synaptic weight adaptation, which yielded upper bounds for the fraction of the circuit consistent with saturated long-term potentiation. These data establish an approach for the locally dense connectomic phenotyping of neuronal circuitry in the mammalian cortex.