A Million-Scale, Difficulty-Stratified Optical-Microscopy Image Annotation for Neuron Reconstruction

Neuronal morphology reconstruction is fundamental to neuronal classification, neural circuit analysis, and the construction of brain-wide connectomes. However, achieving accurate reconstruction remains highly challenging due to the complex morphology of neuronal fibers, dense structural crossings, and interference from imaging background noise and artifacts.In this study, we established a large-scale standardized dataset comprising 8,092,547 difficulty-graded data blocks. Each standardized data block contains three-dimensional image data, a corresponding neuronal skeleton reconstruction in SWC format, as well as the initial seed points and reconstruction direction information required for neuronal tracing. All data have undergone rigorous quality control and standardization procedures and are accompanied by gold-standard reconstruction results.This dataset provides a large collection of high-quality annotated data for the development, training, validation, and benchmarking of automated neuronal reconstruction algorithms. It can facilitate a wide range of applications, including automated tracing, path-finding algorithms, deep learning-based reconstruction models, and reconstruction performance evaluation, thereby providing a valuable data foundation for fully automated and high-precision neuronal morphology reconstruction.

Five representative subsets are publicly available in this repository. The complete dataset is hosted in the HiNeuron Ver2 database at https://atlas.brainsmatics.cn/HiNeuron/(URL). Detailed information regarding the dataset, data visualization, access instructions, and download procedures is available through the database portal.