BigNeuron

该组BigNeuron基准测试数据将包括来自不同物种 (包括果蝇和其他昆虫，鱼，乌龟，鸡，小鼠，大鼠和人类) 的神经元图像堆栈以及神经系统区域，例如皮质和皮质下区域，视网膜和周围神经系统。这些神经元中有许多来自大型神经信息学项目，例如艾伦小鼠和人类细胞类型项目，台湾FlyCircuits和Janelia Fly Light，但许多数据集也直接由全球神经科学家贡献。还使用替代方法重建了一些数据集，并且手动整理和/或校对了一些重建。
测试数据将来自多种光学显微镜方式，尤其是激光扫描显微镜 (共聚焦/2p) 和明场或落射荧光成像。神经元使用不同的方法进行标记，例如遗传标记和病毒/染料/生物细胞素注射，并且将跨越广泛的类型 (例如单极，多极，释放不同的神经递质，并具有多种电生理特性)。
当前版本的BigNeuron将仅考虑单个神经元或在其树木化模式中具有相对清晰分离的神经元的3D图像堆栈。如果适合可靠的分色处理，则将包括使用Brainbow型技术标记的神经元。将大规模台架测试集中在3D单神经元图像堆栈上，将最大限度地提高可行性。未来的BigNeuron版本将考虑其他挑战，例如在密集标记的样本中分离神经元，解决连通性，延时追踪，电子显微镜数据等。

The BigNeuron benchmark dataset collection will encompass neuronal image stacks and nervous system regions derived from a diverse range of species, including fruit flies and other insects, fish, turtles, chickens, mice, rats, and humans. The covered nervous system regions include cortical and subcortical areas, retina, and the peripheral nervous system. A large portion of these neuronal samples originate from prominent large-scale neuroinformatics projects such as the Allen Mouse and Human Cell Type Project, Taiwan FlyCircuits, and Janelia Fly Light, while many datasets are also directly contributed by neuroscientists worldwide. Additionally, some datasets have been reconstructed using alternative approaches, and certain reconstructions have undergone manual curation and/or proofreading.

The test datasets will be acquired using multiple optical microscopy modalities, particularly laser scanning microscopy (confocal/2-photon microscopy) and brightfield or epifluorescence imaging. Neurons are labeled via diverse methods, such as genetic labeling and viral/dye/biocytin injection, and span a wide spectrum of neuronal types, including unipolar, multipolar neurons that release distinct neurotransmitters, and cells with diverse electrophysiological properties.

The current version of BigNeuron will only consider 3D image stacks of individual neurons or neurons with relatively clearly separated arborization patterns. Neurons labeled using Brainbow-style techniques will be included if they are suitable for reliable chromatic separation. Focusing large-scale benchmarking efforts on 3D single-neuronal image stacks will maximize the feasibility of the benchmark. Future iterations of BigNeuron will address additional challenges, such as neuron segmentation in densely labeled samples, connectivity analysis, time-lapse tracing, and electron microscopy data.

搜集汇总

背景与挑战

背景概述

BigNeuron是一个包含多种物种和神经系统区域神经元图像堆栈的数据集，数据来源于多个大型神经信息学项目和全球神经科学家的贡献，专注于3D单神经元图像堆栈的研究。