Zhou Zhi, Sorensen Staci, Zeng Hongkui, Hawrylycz Michael, Peng Hanchuan
Allen Institute for Brain Science, Seattle, WA, USA.
Neuroinformatics. 2015 Apr;13(2):153-66. doi: 10.1007/s12021-014-9249-y.
It is important to digitally reconstruct the 3D morphology of neurons and brain vasculatures. A number of previous methods have been proposed to automate the reconstruction process. However, in many cases, noise and low signal contrast with respect to the image background still hamper our ability to use automation methods directly. Here, we propose an adaptive image enhancement method specifically designed to improve the signal-to-noise ratio of several types of individual neurons and brain vasculature images. Our method is based on detecting the salient features of fibrous structures, e.g. the axon and dendrites combined with adaptive estimation of the optimal context windows where such saliency would be detected. We tested this method for a range of brain image datasets and imaging modalities, including bright-field, confocal and multiphoton fluorescent images of neurons, and magnetic resonance angiograms. Applying our adaptive enhancement to these datasets led to improved accuracy and speed in automated tracing of complicated morphology of neurons and vasculatures.
对神经元和脑脉管系统的三维形态进行数字重建非常重要。之前已经提出了许多方法来实现重建过程的自动化。然而，在很多情况下，相对于图像背景的噪声和低信号对比度仍然阻碍了我们直接使用自动化方法的能力。在此，我们提出一种自适应图像增强方法，该方法专门设计用于提高几种类型的单个神经元和脑脉管系统图像的信噪比。我们的方法基于检测纤维结构（如轴突和树突）的显著特征，并结合对检测到这种显著性的最佳上下文窗口的自适应估计。我们针对一系列脑图像数据集和成像模态测试了该方法，包括神经元的明场、共聚焦和多光子荧光图像以及磁共振血管造影。将我们的自适应增强应用于这些数据集，提高了自动追踪神经元和脉管系统复杂形态的准确性和速度。
