神经皮的多层地图与细分引导的对比学习
Sven Dorkenwald1,2,3, Peter H Li1, Michał Januszewski4
1Google Research, Mountain View, CA, USA.
Nature methods
|November 21, 2023
概括
我们开发了分段引导对比学习的表示 (SegCLR),这是一种用于分析神经回路的机器学习方法. SegCLR精确地用最小的标记数据从3D大脑图像中注释细胞及其组件.
科学领域:
- 神经科学是一个神经科学.
- 机器学习 机器学习
- 计算生物学 计算生物学
背景情况:
- 绘制神经回路需要详细的细胞识别,包括类型,亚细胞组件和连接性.
- 纳米分辨率成像产生了大量的数据,但推断细胞注释仍然是一个重大挑战.
- 现有的方法通常需要广泛的标记数据来准确地分析细胞和亚细胞结构.
研究的目的:
- 引入一种新的自我监督机器学习技术,SegCLR,用于神经电路数据的自动分析.
- 为了能够准确地分类细胞分组,并从3D脑图像中推断细胞类型.
- 为了减少对大规模手动注释数据集的依赖,用于神经电路映射.
主要方法:
- 开发了以分割为指导的对比学习表示 (SegCLR),一种自我监督的学习方法.
- 应用SegCLR对人类和小鼠皮层的3D图像体积.
- 利用3D图像和细分来生成细胞表示.
主要成果:
- 通过SegCLR实现了细胞子组件的准确分类.
- 性能相当于监督方法,但需要400倍少的标记数据.
- 从小的神经片段 (10微米) 启用了细胞类型推断,并促进了突触伙伴的分析.
结论:
- SegCLR提供了一种高效准确的方法,用于从高分辨率成像数据中注释神经电路.
- 该技术增强了大规模神经数据集的实用性,特别是那些神经结构不完整的神经数据集.
- SegCLR支持神经连接和细胞亚型的自动化,大规模分析.
更多相关视频
相关概念视频
01:11Cerebrum: Anatomical Overview II
Each cerebral hemisphere can be divided into three main regions. The outermost region, the cerebral cortex, is a thin layer (2 to 4 millimeters thick) made up of gray matter, consisting of neuron cell bodies, dendrites, glial cells, and blood vessels. The middle region, or white matter, is primarily composed of myelinated nerve fibers organized into three types of large tracts: association fibers, commissures, and projection fibers. Association fibers connect different areas within the same...
01:16Spinal Cord: Cross-sectional Anatomy
The cross-sectional anatomy of the spinal cord offers a detailed view of its complex structure and function within the central nervous system. At the core of the spinal cord lies the gray matter, characterized by its butterfly or "H"-shaped appearance in cross-section. This central region is enveloped by white matter, with the overall structure divided into symmetrical halves by the dorsal median sulcus and the ventral median fissure.
Gray Matter and its Components
Central to the gray matter is...
Gray Matter and its Components
Central to the gray matter is...