A higher level for neuron reconstruction

为了更好地了解我们认知的基础——大脑，我们需要详细了解它的结构，特别是它的“接线图”。不幸的是，到目前为止，甚至连大脑最微小部分的精确图都没有！由于体积电子显微镜的巨大进步，人类专家现在可以可靠地追踪单个神经元。不幸的是，即使是对大脑的一小部分进行密集追踪的努力也需要几十年的时间。在图像质量较差的地方，自动化方法仍然会失效。与人类不同，这种自动化方法还不能利用形状和环境方面的知识。目前的建议将开发能够学习形状和结构的高阶模型。这样的学习将依赖于在这个应用领域中有意义的新的误差度量，也将在这里开发，以及对局部不确定性的估计。这种不完全局部的性质将导致困难的优化问题，我们有信心通过对原始数据的多个局部合理解释的有效生成来解决这些问题。总而言之，该项目将有助于自动提取大脑接线图的部分，其准确性将使新的生物学问题得以提出和回答。

To better understand the brain, the substrate of our cognition, we require detailed knowledge of its structure, in particular its "wiring diagram". Unfortunately, an exact diagram of even the tiniest part of a brain is not available to date! Thanks to great advances in volume electron microscopy, human experts can now reliably trace individual neurons. Unfortunately, the effort for a dense tracing of even a tiny part of the brain is in the order of multiple decades. Automated methods still break down in spots where the image quality is poor. Unlike humans, such automated methods cannot leverage knowledge on shape and context yet. The present proposal will develop higher order models that can learn about shape and structure. Such learning will rely on new error metrics that are meaningful in this application domain, also to be developed here, and on estimators for local uncertainty. Such not quite local properties will result in difficult optimization problems which we are confident to solve thanks to the efficient generation of multiple locally plausible interpretations of the raw data. In summary, this project will contribute to the automated extraction of parts of the wiring diagram of a brain, with an accuracy that shall enable the asking and answering of new biological questions.