Scholarlink

<p>Accurate reconstruction of neuronal morphology is a cornerstone for understanding the structure and function of neural circuits. Despite significant advances in imaging and reconstruction techniques that have greatly expanded the scale and quality of neuronal data, existing public datasets still lack large-scale, high-quality, and well-annotated resources, limiting their utility for training and evaluating modern AI models. In this study, we present an open, graded neuronal dataset covering the entire mouse brain. Leveraging high-resolution optical microscopy, we combined automated reconstruction with multi-user collaborative proofreading to generate high-precision neuronal annotations. Our dataset was derived from 10,547 neurons across 258 whole-brain samples, and approximately 8,092,547 standardized data units with graded difficulty levels were generated through an interactive collaborative reconstruction process. Moreover, the dataset supports both API access and bulk downloads, offering researchers convenient data acquisition methods. This work fills a critical gap in standardized, annotated whole-brain neuronal resources, provides a high-quality benchmark for AI-driven automated neuronal reconstruction, and lays a solid foundation for building large-scale neural circuit maps.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

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

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Mingwei Liao,
  • Wu Chen,
  • Shengda Bao,
  • Xueyan Jia,
  • Xiaowei Chen,
  • Chi Xiao,
  • Qingming Luo,
  • Hui Gong,
  • Anan Li

摘要

Accurate reconstruction of neuronal morphology is a cornerstone for understanding the structure and function of neural circuits. Despite significant advances in imaging and reconstruction techniques that have greatly expanded the scale and quality of neuronal data, existing public datasets still lack large-scale, high-quality, and well-annotated resources, limiting their utility for training and evaluating modern AI models. In this study, we present an open, graded neuronal dataset covering the entire mouse brain. Leveraging high-resolution optical microscopy, we combined automated reconstruction with multi-user collaborative proofreading to generate high-precision neuronal annotations. Our dataset was derived from 10,547 neurons across 258 whole-brain samples, and approximately 8,092,547 standardized data units with graded difficulty levels were generated through an interactive collaborative reconstruction process. Moreover, the dataset supports both API access and bulk downloads, offering researchers convenient data acquisition methods. This work fills a critical gap in standardized, annotated whole-brain neuronal resources, provides a high-quality benchmark for AI-driven automated neuronal reconstruction, and lays a solid foundation for building large-scale neural circuit maps.